The Internet-Informed Patient and Cancer Screening Adherence: The ...

University of Miami

Scholarly Repository Open Access Dissertations

Electronic Theses and Dissertations

2012-05-02

The Internet-Informed Patient and Cancer Screening Adherence: The Role of PatientPhysician Communication Yuliya Shneyderman University of Miami, [email protected]

Follow this and additional works at: http://scholarlyrepository.miami.edu/oa_dissertations Recommended Citation Shneyderman, Yuliya, "The Internet-Informed Patient and Cancer Screening Adherence: The Role of Patient-Physician Communication" (2012). Open Access Dissertations. Paper 790.

This Open access is brought to you for free and open access by the Electronic Theses and Dissertations at Scholarly Repository. It has been accepted for inclusion in Open Access Dissertations by an authorized administrator of Scholarly Repository. For more information, please contact [email protected].

UNIVERSITY OF MIAMI

THE INTERNET-INFORMED PATIENT AND CANCER SCREENING ADHERENCE: THE ROLE OF PATIENT-PHYSICIAN COMMUNICATION

By

Yuliya Shneyderman

A DISSERTATION

Submitted to the Faculty of the University of Miami in partial fulfillment of the requirements for the degree of Doctor of Philosophy

Coral Gables, Florida

May 2012

©2012 Yuliya Shneyderman All Rights Reserved

UNIVERSITY OF MIAMI

A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy

THE INTERNET-INFORMED PATIENT AND CANCER SCREENING ADHERENCE: THE ROLE OF PATIENT-PHYSICIAN COMMUNICATION Yuliya Shneyderman

Approved: ________________ Seth J. Schwartz, Ph.D. Associate Professor Epidemiology and Public Health

_________________ Terri A. Scandura, Ph.D. Dean of the Graduate School

________________ Kristopher L. Arheart, Ed.D. Associate Professor Epidemiology and Public Health

_________________ Margaret M. Byrne, Ph.D. Research Associate Professor Epidemiology and Public Health

________________ Lila J. Finney Rutten, Ph.D., M.P.H. Senior Behavioral Scientist Clinical Monitoring Research Program SAIC-Frederick, Inc. National Cancer Institute at Frederick Frederick, Maryland

_________________ Julie Kornfeld, Ph.D. Director of Education Epidemiology and Public Health

SHNEYDERMAN, YULIYA

(Ph.D., Epidemiology and Public Health)

The Internet-Informed Patient and Cancer Screening Adherence: The Role of Patient-Physician Communication.

(May 2012)

Abstract of a dissertation at the University of Miami. Dissertation supervised by Associate Professor Seth J. Schwartz. No. of pages in text (111) Breast, cervical, and colorectal cancers have population-based effective screenings that are recommended for individuals of a certain age and sex, while prostate cancer screening is controversial. People often seek information from various sources about cancer and health behaviors, and in the information age, this information may come from the Internet. Furthermore, individuals who see their physicians may be influenced by how well they communicate with their healthcare providers when deciding on health behaviors such as cancer screenings. Getting screened for breast, cervical, and colorectal cancer at the recommended intervals is linked to decreases in mortality rates for those cancers. While prostate cancer screening is not linked to decreases in mortality, the screening is often used by men without cancer risk factors. This study seeks to study the association between information seeking, both from the Internet and other sources, and cancer screening guideline adherence for breast, cervical, and colorectal cancers, and screening rates for prostate cancer, and to determine the role that patient-physician communication plays in that relationship.

Acknowledgements I would like to acknowledge my mentor, Seth Schwartz, without whom this work would not have been possible. I also want to thank my family and friends, who supported me throughout the years I spent in graduate school.

iii

Table of Contents List of Tables

v

List of Figures

vii

Chapter 1. Introduction, Background, Specific Aims

1

Chapter 2. Methods

18

Chapter 3. Analysis

30

Chapter 4. Results

37

Chapter 5. Discussion

50

Tables and Figures

72

References

99

Appendix 1. HINTS conceptual framework

111

iv

List of Tables

Table 1.1. Summary of recommendations for cancer screenings from the USPSTF, both current and past

72

Table 2.1. Cancer and health information seeking questions from HINTS 2003 and 2007.

73

Table 2.2. Questions on the patient-physician communication scale on HINTS 2003 and 2007.

74

Table 4.1 - 2003 weighted descriptive statistics (n = 6,369)

75

Table 4.2 - 2007 weighted descriptive statistics (n = 7,623)

76

Table 4.3 - Breast cancer screening and cancer information seeking

78

Table 4.4 - Breast cancer screening and health information seeking

79

Table 4.5 – Cervical cancer screening and cancer information seeking

80

Table 4.6 – Cervical cancer screening and health information seeking

81

Table 4.7 – Colorectal cancer screening and cancer information seeking

82

Table 4.8 – Colorectal cancer screening and health information seeking

83

Table 4.9 – Prostate cancer screening and cancer/health information seeking

84

Table 4.10 – Prostate cancer screening and cancer/health information seeking in participants who did not get a recommendation from their physicians

85

Table 4.11 – Breast cancer screening and health information seeking – partial mediation model

86

Table 4.12 – Cervical cancer screening and information seeking – partial mediation model

87

v

Table 4.13 – Colorectal cancer screening and information seeking – partial mediation model

89

Table 4.14 – Prostate cancer screening and health information seeking – partial mediation model

91

Table 4.15 - Cervical cancer screening and perceived physician reaction to health information

92

Table 4.16 – Colorectal cancer screening and perceived physician reaction to health information

93

Table 4.17 – Partial mediation test of the relationship between patient-physician communication and cervical cancer screening

94

Table 4.18 – Partial mediation test of the relationship between patient-physician communication and colorectal cancer screening

vi

96

List of Figures Figure 1.1. Model tested in Specific Aim 2.

97

Figure 1.2. Model tested in Specific Aim 3.

97

Figure 3.1. Alternative model for Specific Aim 2.

98

Figure 3.2. Alternative model for Specific Aim 3.

98

vii

Chapter 1 Introduction, Background, Specific Aims Introduction The goal of this dissertation is to analyze the various links that may exist between cancer screening adherence and behavior, cancer and health information seeking, and patient-physician communication, while following theorized models of how these three health behaviors work together. Population-based cancer screening guidelines exist for breast, cervical and colorectal cancers, and while prostate cancer screening is not recommended for the population at large, many men nevertheless get screened. The guidelines for screening are different for each of the different cancers and depend on age and sex of the individual. Following the recommended timelines for screening (for breast, cervical and colorectal cancers) and getting screened at all for prostate cancer are the health behavior outcomes that are modeled in this dissertation. The outcome of on-schedule screening is analyzed in the context of patients who seek health information from various sources, but especially from the Internet, because information is freely available, abundant, and generally easy to find on the Internet. In addition, the role of patient-physician communication during healthcare visits is analyzed for an effect on cancer screening adherence, and for the role that it may play in the relationship between information seeking and screening by conducting mediation modeling. Finally, for those patients who decide to talk to their healthcare providers about the information they found on the Internet, the effect of perceived physician reaction on cancer screening behavior and the role it may play in the relationship

1

2

between to patient-physician communication and cancer screening adherence and behavior. Background Cancer is one of the leading causes of death in the United States. Between 2003 and 2007, age-adjusted incidence rates of cancer were highest for prostate, breast, lung, and colorectal cancers, while mortality rates were highest for lung, prostate, breast, and colorectal cancers (U.S. Cancer Statistics Working Group, 2010). For men, prostate cancer had the highest age-adjusted incidence rate of 153.5 per 100,000 and colorectal cancer was third, with an incidence rate of 57.1 per 100,000 (Kohler et al., 2011). For women, breast cancer had the highest age-adjusted incidence rate of 120.7 per 100,000, colorectal cancer was third with an incidence rate of 42.4 per 100,000, and cervical cancer had an incidence rate of 8.1 per 100,000 (Kohler et al., 2011). Death rates followed similar rankings for men and women. For men, prostate cancer had the second highest age-adjusted death rate of 24.7 per 100,000, with colorectal cancer ranking next with a death rate of 21.2 per 100,000 (Kohler et al., 2011). Finally, for women, breast and colorectal cancers had the second and third highest death rates (24.0 and 14.9 per 100,000 respectively), while cervical cancer had lower death rates of 2.4 per 100,000 (Kohler et al., 2011). Prevention is an important tool that can be used to reduce the high mortality rates associated with cancer. Primary prevention is chiefly concerned with stopping a condition or disease before clinical and pre-clinical changes have evolved (Reisig & Wildner, 2008). Secondary prevention is a way to identify persons with a disease before clinical signs and symptoms have manifested (Wildner & Nennstiel-Ratzel, 2008), and in the case

3

of cancer, often refers to early detection or screening. Although primary prevention of most cancers is not yet available because causes and catalysts are not known or may be unpreventable, secondary prevention in the form of screening and early detection can be effective at lowering mortality and morbidity (Meissner et al., 2004). Many cancers listed as leading causes of death have effective screening tools available - in particular, mammography for breast cancer, the Pap test for cervical cancer, and multiple screening modalities for colorectal cancer. These screenings are also recommended on a population level, meaning that all people of a certain age and sex should be screened regardless of their cancer risk (U.S. Preventive Services Task Force, 2003, 2008b, 2008d, 2009). Evidence supporting population screening for prostate cancer to date has been insufficient, therefore, recommendations around prostate screening focus on an informed and shared decision making process between age-appropriate men and their healthcare providers (U.S. Preventive Services Task Force, 2008c). Recently, the United States Preventive Services Task Force (USPSTF) had begun to get public comment on a statement that recommends against population prostate cancer screening, although other professional societies may still recommend it (American Cancer Society, 2011b; U.S. Preventive Services Task Force, 2011a). Factors that affect adherence to recommended screening guidelines are an important area of investigation. This study examines screening adherence and behavior for breast, cervical, colorectal and prostate cancers in a large nationally representative dataset. Additionally, it examines some of the correlates of screening guideline adherence for the breast, cervical and colorectal cancers and participation in screening for prostate cancer. These screenings were selected because they are either recommended for the entire population of a certain sex and within

4

specified age ranges or widely used by physicians and patients (in the case of prostate cancer). Cancer Screening Efficacy The Pap test checks for changes in the cells in the cervix that can lead to cervical cancer (Office on Women's Health, 2010). The Pap test has been known to be effective in decreasing incidence of invasive cervical cancer and mortality (Eddy, 1986) and estimates of the effect range from a 60% reduction of cancers in women aged 40 to an 80% reduction at age 64 (Sasieni, Castanon, & Cuzick, 2009). Mammograms, or X-rays of the breast, are used either as a screening tool or as a diagnostic tool to detect breast cancer tumors (National Cancer Institute, 2010). This screening has been shown to be effective at lowering rates of mortality by about 16% (Mandelblatt et al., 2009). Colorectal cancer screening can lead to early detection and removal of pre-cancerous polyps (2011). Tests for colorectal cancer fall into two categories – tests that can detect cancer by using fecal testing or tests that can detect both cancer or pre-cancerous polyps by using a variety of endoscopic procedures (R. A. Smith, Cokkinides, Brooks, Saslow, & Brawley, 2010). These screenings, when tested in randomized controlled trials, have been found to reduce the relative risk for colorectal cancer mortality by 15% (Hewitson, Glasziou, Watson, Towler, & Irwig, 2008). Prostate cancer screening is performed using a blood test called the prostate-specific antigen test (PSA) and using a digital rectal examination (DRE) (Schroder, 2009). However, in contrast to other cancer screening tests, randomized trials of prostate cancer screening have shown mixed results – in one study, no difference in mortality was found between the intervention and control groups (Andriole et al., 2009), whereas another showed a 20% reduction in the death rate but a

5

high risk of false positive diagnoses (Schroder et al., 2009). Further study of PSA testing is ongoing. Cancer Screening Guidelines Screening guidelines are provided by a variety of professional organizations and non-profit organizations, such as the American Cancer Society (R. A. Smith et al., 2010). Another such organization, the United States Preventive Services Task Force (USPSTF), conducts rigorous scientific evidence reviews on a variety of clinical preventive health services, including cancer screening. This panel consists mainly of experts in prevention and evidence-based medicine. They release recommendations for clinicians and constantly review scientific evidence and update these recommendations based on the latest research available. In order for the USPSTF to recommend a procedure, the benefits must outweigh the harms and quality of life must be maintained (U.S. Preventive Services Task Force, 2011b). For the current study, recommendations from the USPSTF that were current at the time of data collection were used to define adherence. For breast cancer, average-risk women were recommended to receive counseling and mammography every 1-2 years from their healthcare providers at age 40 and older (U.S. Preventive Services Task Force, 2002). For cervical cancer, Pap screening is recommended in women who are sexually active and who have an intact cervix. Screenings should begin 3 years after sexual initiation or age 21 and continue until women reach the age of 65, with the exception of women who are at high risk for cervical cancer (U.S. Preventive Services Task Force, 2003). For colorectal cancer, the guidelines to screening are more varied and flexible – screening should begin at age 50 and include either an annual fecal occult blood test

6

(FOBT), a flexible sigmoidoscopy (or other types of endoscopy not including colonoscopy) every 5 years, or a colonoscopy every 10 years. Screenings should continue until the age of 75 years (U.S. Preventive Services Task Force, 2008a). Finally, prostate cancer screening is not recommended at the population level because the evidence is insufficient to determine whether PSA testing has the appropriate balance of harms and benefits (U.S. Preventive Services Task Force, 2008c). An informed decision making process should be undertaken with a healthcare provider when the patient reaches age 50 and is expected to live for at least 10 more years based on his health status (men at higher risk for prostate cancer can begin receiving information at age 40 or 45, and men not expected to live 10 years should not be offered prostate cancer screening) (R. A. Smith et al., 2010). For a summary of screening recommendations that were correct at the time of data collection and the screening recommendations that are up to date, see Table 1.1. Trends in Cancer Screening Rates Rates of adherence to screening guidelines vary for different cancers. In 2008, 78% of women reported getting a Pap smear in the past 3 years, whereas 53% of women reported having a mammogram in the past year (Smith et al., 2010). Another report indicated a drop in rates of past-two-year mammography from 70% in 2000 to 66% in 2005 (Breen et al., 2007). These rates vary by ethnicity (Hispanic and Black women were less likely to be screened regularly), age (younger women and much older women received fewer screenings), insurance status (HMO membership was associated with higher screening rates), and other characteristics like marital status (Black, McCulloch, Martin, & Kan, 2011; Jennings-Dozier & Lawrence, 2000; A. S. O'Malley, Forrest, & Mandelblatt, 2002; Rigal, Saurel-Cubizolles, Falcoff, Bouyer, & Ringa, 2011). Within

7

different ethnicities, differences in screening uptake also exist based on education, insurance status, age, and other socio-demographic factors. For example, Black women with a high school degree were more likely to be screened, whereas Hispanic women age 40 or older were less likely to be screened (Jennings-Dozier & Lawrence, 2000). Screening rates for colorectal cancer (CRC) have been increasing since 2005 (R. A. Smith et al., 2010), although this test still has the lowest rate of adherence when compared to other cancer screening tests (Gierisch & Bastian, 2010). A review of empirical results (Subramanian, Klosterman, Amonkar, & Hunt, 2004) found that the rates of utilization of CRC tests were between 40-60% for FOBT, and about 50-75% for sigmoidoscopy or colonoscopy, though it is not known how many people get both types of tests. According to the CDC (2011), a third of the population still does not present for colorectal cancer screening. Some factors affecting adherence to these tests include education, health insurance coverage, and knowledge about and attitudes toward CRC screening tests (Subramanian et al., 2004). Race and ethnicity, gender, and disability status have also been found to have an influence on rates of CRC screening – whites, men, persons with disability, and person with health insurance are all more likely to get screened (Rim, Joseph, Steele, Thompson, & Seeff, 2011). Prostate cancer screening is different from other routine cancer screenings in that the recommended procedure involves an element of communication between patient and provider known as informed decision making (IDM). IDM occurs when an individual understands the disease, the test, and the test’s consequences; has considered his preferences; has participated in the decision making process with his healthcare provider; and then has made a decision consistent with his preferences (Briss et al., 2004).

8

Informed decision making is particularly important because the effects of widespread PSA testing are uncertain (Andriole, 2005), and because stage of cancer at diagnosis and mortality are similar for those who do and do not choose to undergo PSA testing (Andriole et al., 2009). Even though prostate cancer screenings have uncertain benefits, many men choose to utilize this screening modality. In a randomized trial of the benefits of routing PSA screening, compliance was very high in the intervention group (85%) and fairly high in the control group (40-50% for PSA and DRE), indicating that men chose to utilize PSA screening as part of their routine care (Andriole et al., 2009; Andriole et al., 2005). Prostate cancer screening rate is about 44% in the US population, though no surveillance system exists to track IDM in association with screening uptake (R. A. Smith et al., 2010). Population Knowledge of Cancer Screening Early efforts to promote cancer screenings for cervical and breast cancers involved awareness campaigns and community outreach strategies (Marcus & Crane, 1998). These efforts were intended to serve as a way to increase people’s knowledge about early cancer detection and its benefits. However, recent findings indicate that public knowledge about cancer screening and screening guidelines tends to be low. For example, in a study using data from the Health Information National Trend Survey, only about 5% of adults selected cancer screening as a strategy for preventing cancer in an open-ended question about reducing cancer risk. Of those participants who did endorse cancer screening as a form of cancer prevention, more than half mentioned mammography, about 40% mentioned the Pap test and endoscopy, more than a third mentioned the PSA exam, and all other cancer screenings were associated with

9

knowledge rates of less than 15% (Hawkins, Berkowitz, & Peipins, 2010). In more specific populations, such as adults over age 65, more than 25% of individuals indicated that they were not familiar with the fecal occult blood test, and about 17% had not heard of either sigmoidoscopy or colonoscopy (Berkowitz, Hawkins, Peipins, White, & Nadel, 2008). A small study, using an ethnically diverse sample, found that few participants could answer open-ended questions about causes of CRC, screening for CRC, and cancer screening in general (Shokar, Vernon, & Weller, 2005). In a large, multi-center cohort study of HIV-infected and at-risk women, fewer than 50% of all women were able to answer some of the questions correctly about the Pap test, although about three-quarters knew the purpose of the test and what an abnormal result meant (Massad et al., 2010). A study of African American women found that only 30% of women reported feeling wellinformed about breast cancer and fewer than a third were able to identify breast cancer screening guidelines (Sadler et al., 2007). A study utilizing in-depth interviewing with African American and white men found that men with low educational attainment almost universally did not know about the causes of prostate cancer, about different screening modalities, or about why these modalities were used (Winterich et al., 2009). Health Information Seeking Due to the changing nature of how people obtain information in the digital age, many people now look to the Internet for health information. According to research conducted by the Pew Internet and American Life Project (Fox, 2011a), 59% percent of all adults in the US search for health information online. That number rises to 80% when only regular Internet users are considered. People most often search for information on specific diseases or medical problems, or on medical treatments and procedures (Fox,

10

2011a). An example of this type of search topic is colonoscopy, which is one of the top ten topics searched on WebMD, a popular health website (Fox, 2011a). Although doctors, nurses, and other health professionals remain the most important source of health information for US adults (Hesse, Moser, Rutten, & Kreps, 2006), the Internet is consistently rated as the second most important tool (Fox, 2011b; Hesse et al., 2006; Koch-Weser, Bradshaw, Gualtieri, & Gallagher, 2010). Some of the main advantages of accessing health information on the Internet, whether in general or specifically about cancer, include the short amount of time it takes when compared to waiting for a visit with a healthcare professional, the flexibility of being able to search at the patient’s convenience, and access to abundant and usually free information (Cline & Haynes, 2001). Women, Whites, those with higher income and education, and younger people are all more likely to use the Internet to search for healthrelated information (Hesse et al., 2005; Rice, 2006). The majority of participants in a large nationally representative survey (HINTS) reported having a lot or some trust in Internet health information, specifically when they searched for information about cancer (Hesse et al., 2005), and in a smaller study, most participants rated the quality of Internetobtained information as high (Diaz et al., 2002). Moreover, both people with and without illnesses use the Internet to search for health information (Baker, Wagner, Singer, & Bundorf, 2003; Fox, 2011b); and people tend to use the Internet for other health related reasons such as creating and accessing support groups, looking for information on specific hospitals or physicians, and planning diet and physical activity regimens (Baker et al., 2003; Ferguson, 2007).

11

Although some concerns exist about the quality, accessibility, and accuracy of health information on the Internet (Berland et al., 2001; Jadad & Gagliardi, 1998) and about patients ignoring treatment and advice from their own physicians in favor of information on the Internet (Weaver, Thompson, Weaver, & Hopkins, 2009), a systematic review found that very few people have been harmed by inaccurate Internet health information (Crocco, Villasis-Keever, & Jadad, 2002). Another study found that health information relating specifically to cancer is quite accurate (Huang & Penson, 2008), although certain websites, such as those that provide information on alternative cancer therapies, are linked to higher levels of inaccuracies on the sites (Bernstam et al., 2008). Only about 11% of patients were found to be non-adherent to physician advice because of information found on the Internet in one cross-sectional study (Weaver et al., 2009). More importantly, patients who seek Internet health information are also more likely than non-health information seekers to consult with their physicians and to report that physicians’ advice drives their medical decision making (Fox & Fallows, 2003). Association Between Information Seeking and Cancer Screening An association between seeking health information and rates of cancer screening has been documented in several studies. In a study which used data from HINTS, information seeking specifically about cancer from any source was found to be associated with getting colorectal cancer screenings on the schedule recommended by guidelines (Ling, Klein, & Dang, 2006). Cancer information seeking was also found to be associated with getting PSA screenings and health information scanning (a construct that assessed how much attention participants paid to health information) was associated with getting mammograms within the past 2 years (Shim, Kelly, & Hornik, 2006). However, Internet-

12

specific health and cancer information seeking was not examined separately in those studies – all types of health and cancer information seeking was combined into one variable. Patient-Physician Communication Some concerns have also arisen about how Internet health information seeking will affect the relationship between the patient and the health-care provider. The Institute of Medicine has called for health care that empowers the patient to make choices, that provides the patient with all the information he or she needs, and that conforms to the patient’s needs and value (Committee on Quality Health Care in America, 2001). Patientphysician communication is a key aspect of this type of health care, given that informed patients are more likely to participate in their own care. Methods to convey medical information to patients that have been recommended to patients include relationship building, delivering information, and validating understanding (Epstein, Alper, & Quill, 2004). Patient-physician communication is important for building trust between patient and provider, gathering information during an office visit, and for making health-care decisions (Baile & Aaron, 2005; Beck, Daughtridge, & Sloane, 2002; Ong, de Haes, Hoos, & Lammes, 1995). Good patient-physician communication has been linked to several positive health outcomes, such as symptom resolution, physical measures like blood pressure, and patient satisfaction (Stewart, 1995; Zachariae et al., 2003). Perceived control over a disease and adherence to treatment were also linked to good doctor-patient communication (Zachariae et al., 2003; Zolnierek & Dimatteo, 2009). Not all people who look for health information online choose to bring that information to their physician or other health-care provider. In one study, a majority of

13

Internet information seekers did not bring this information to their physician (Diaz et al., 2002). Research shows that Internet health information is used by people to supplement the information they receive after a doctor’s visit or simply to research an issue that does not lead to a physician visit (Baker et al., 2003; Diaz et al., 2002; Koch-Weser et al., 2010). Those who do talk to their physicians may encounter a range of positive validating responses from the healthcare providers or resistance in the form of denial or dismissal. The latter is more likely if the doctor views the patient as challenging his or her authority (Helft, Hlubocky, & Daugherty, 2003; Malone et al., 2004; McMullan, 2006; Wald, Dube, & Anthony, 2007). It is important to investigate the extent to which these responses from healthcare providers, as perceived by patients, affect health behaviors such as cancer screening. Links between Health Information Seeking and Patient-Physician Communication Some studies have been devoted to linking Internet health information with patient-physician communication. In one study utilizing a focus group setting with physicians, participants expressed some concerns about Internet-informed patients. These concerns included reports that some patients experienced distress because of the information they found, and that physicians felt constrained by time and could not address all the information that was brought to them, but that they generally felt positively about self-educated patients who did not challenge or act adversarial toward physicians (Ahmad, Hudak, Bercovitz, Hollenberg, & Levinson, 2006). In another study, conducted from the patients’ perspectives, participants reported a mostly positive response from physicians, especially when a non-confrontational or “face-saving” method of communication was used (Bylund et al., 2007). An especially important

14

feature of Internet information-based conversations in health-care settings seems to be the physician taking the patient seriously, which is linked to higher satisfaction with the medical encounter (Bylund, Gueguen, D'Agostino, Li, & Sonet, 2010). Patient-physician communication overall tended to be positively affected by Internet health information unless the physician or patient felt challenged or if communication skills were reported as low by either patient or healthcare provider (Broom, 2005; Murray et al., 2003; Stevenson, Kerr, Murray, & Nazareth, 2007; Wald et al., 2007). Many authors of the studies listed above took the opportunity to recommend that physicians pay attention to their Internet-informed patients and attempt to find and recommend resources to help them obtain accurate and trusted information. Study Goals Internet health information seeking, patient-physician communication, and cancer screening adherence have all been studied extensively separately, but they have not been studied together, and as a result, their interrelationships are not well understood. The links between communication and health outcomes such as cancer screenings have been established, as have associations between accessing Internet health information and patient physician communication. There remains a question about whether Internetinformed patients are more or less likely to adhere to cancer screening guidelines and whether patient-physician communication acts as a possible mediation mechanism in this relationship. Mediation analysis is one way to assess this possibility. Another important influence in the pathways linking the above three constructs may be the response that patients perceive when they talk about Internet health information during a visit with a doctor or other provider. Those who perceive a more positive response may be more

15

likely to perform better self-care, which in the case of cancer prevention may mean adherence to cancer screening guidelines. Finally, it is important to note that subgroup differences exist in all of the above constructs. Women, whites, younger people, and those with higher income and educational attainment tend to use Internet health information more than other subgroups (Hesse et al., 2006; Koch-Weser et al., 2010). Likewise, significant disparities exist in cancer screening adherence and screening rates are affected by race and ethnicity, health insurance status, health status, and socio-economic status (SES). For example, it is well documented that White and Black women are more likely to get cervical cancer screening than Hispanic women; women who have attained higher education are more likely to get mammograms; and health insurance availability is linked with higher rates of screening for all cancers (R. A. Smith et al., 2010). Finally, patient-physician communication is also affected by factors such as gender, ethnicity, SES, and health status, with women and minority patients reporting lower levels of patient-physician communication and of participatory decision making during a healthcare visit (Cooper-Patrick et al., 1999; Cooper et al., 2003). For that reason, it is important to include covariates like age, sex, and socio-economic status in any analysis of these relationships. Specific Aims Specific Aim 1. To examine the relationship between (a) health and (b) cancer Internet information seeking and cancer screening guidelines adherence for breast, cervical, and colorectal cancer. Furthermore, to examine the relationship between health and cancer Internet information seeking and prostate cancer screening.

16

Hypothesis 1a: There is a positive relationship between Internet health information seeking and cancer screening guidelines adherence (breast, cervical, colorectal) or screening receipt (prostate), when controlling for sex, age, health, family history of cancer, health insurance status, and income. Hypothesis 1b: There is a positive relationship between Internet cancer information seeking and cancer screening guidelines adherence (breast, cervical, colorectal) or screening receipt (prostate), when controlling for sex, age, health, family history of cancer, health insurance status, and income. Specific Aim 2. To determine the role patient-physician communication plays in the relationship between (a) health and (b) cancer information seeking and cancer screening adherence (See Figure 1.1). Hypothesis 2a: Patient-physician communication will partially mediate the relationship between health information seeking and cancer screening guidelines adherence (breast, cervical, colorectal) or screening receipt (prostate). Hypothesis 2b: Patient-physician communication will partially mediate the relationship between cancer information seeking and cancer screening guidelines adherence (breast, cervical, colorectal) or screening receipt (prostate). Specific Aim 3. To analyze the role perceived physician reaction to Internet information plays in the relationship between patient-physician communication and cancer screening adherence (cervical, colorectal) among those patients who talk to their physicians about Internet information (see Figure 1.2). Hypothesis 3a: Perceived physician reaction will be associated with cancer screening adherence (cervical, colorectal).

17

Hypothesis 3b: Perceived physician reaction will partially mediate the relationship between patient-physician communication and cancer screening adherence (cervical, colorectal).

Chapter 2 Methods Data Source, Data Collection, Response Rates, and Sampling Data used for this study are from the Health Information National Trends Survey (HINTS). HINTS is a biennial, cross-sectional survey of a nationally representative sample of American adults that includes data about health information, information technology, and health behaviors, with a focus on cancer-related information and prevention. It was developed by the Health Communication and Informatics Research Branch in the National Cancer Institute (NCI) Division of Cancer Control and Population Sciences. The purpose of developing HINTS was to create a monitoring tool on the rapidly changing field of health communications. In particular, the focus of the survey is on cancer: its risks, prevention, early detection (e.g. screening), diagnosis, and how people access relevant health information. HINTS also assesses use of health information from the Internet and from other sources (Cantor et al., 2009; Cantor, Covell, Davis, Park, & Rizzo, 2005; "HINTS 2003 Final Report," 2003). HINTS data are publically available from three administrations: 2003, 2005 and 2007. Data from 2003 and 2007 are analyzed in this study. The mode of administration and sampling strategy were slightly different between years. In 2003, random digit dialing (RDD) was used to select a probability sample of phone numbers in the US. Telephone exchanges with high numbers of Black and Hispanic people were oversampled to assure adequate representation of these households to create a nationally representative sample. Households that were sampled and that had a mailing address attached to the phone number received a letter prior to the call. A household screener was

18

19

administered prior to the interview to identify the number of adults living in the household, and one adult per household was randomly selected for the interview. Prescreeners were administered by personnel from Westat (a contractor), and interviews were conducted in English or Spanish using Computer Assisted Telephone Interviewing (CATI). The response rate calculations include the screener response rate (i.e. percentage of participants who responded to the pre-interviewer screener) and the extended interview response rate (i.e. the percentage of participants who began or completed the full CATI). The response rate at the household screener level was 55%, and the response rate for extended interview was 62.8%. A total of 6,369 participants completed or partially completed the questionnaire ("HINTS 2003 Final Report," 2003). Greater details on the sample or sampling design of HINTS 2003 have been published elsewhere (D. E. Nelson et al., 2004). In 2007, two modes of administering the survey and two separate sampling frames – telephone and mail – were used in an attempt to obtain better response rates and population coverage than were found in previous administrations of HINTS. The twomode design also allows for access to people who solely use cell-phones and to those without a telephone. The telephone mode used the same techniques as the 2003 HINTS RDD to select a sample for the CATI survey administration. A total of 3,767 participants fully completed interviews and an additional 325 partially completed the interviews. The final response rate for participants who received a phone call was 25.3%. HINTS 2007 also sent out a mail survey. A random sample of addresses was provided by a vendor. The sampling unit for the mail sample was an individual address, and the sampling frame was all residential addresses in the US in a particular database.

20

The sampling frame was stratified into a high-minority (where Hispanics and African Americans comprise more than 24% of all residents) and low-minority strata, and highminority strata were over-sampled. All adults in the household were asked to complete the survey (thus, this administration mode provided a stratified cluster sample in which the household is the cluster). To reduce duplication, a question was included to discern whether the household received mail in multiple ways (e.g., P.O. boxes as well as home addresses). Advance letters were sent to all randomly selected households. Nonresponders received a reminder postcard and another set of questionnaires. A total of 3,582 surveys were returned (3,473 fully completed and 109 partially completed) (Cantor et al., 2009). The household response rate for the mail survey was 40%. Data from the mail and telephone sample were merged into one dataset. Participants who completed a HINTS questionnaire during any administration received a sampling weight and a set of replicate weights at each administration, which are used for jackknife variance estimation. Each sample-specific weight consists of three components: a base weight, which is the reciprocal of the probability of the person being sampled; an adjustment for non-response; and a calibration adjustment using auxiliary information from the Current Population Survey or the U.S. Census to ensure representativeness within the US population. This sample-specific weight is used when calculating point estimates. The jackknife technique is used to estimate variance in HINTS with a set of 50 replicate weights. This technique consists of systematically deleting portions of the original sample, treating these subsamples as if they were the complete sample, and then recalculating sampling weights (i.e. taking 50 random subsamples from the full sample and calculating the variance using them all). In 2007, each

21

sampled adult received three full-sample weights (address, RDD, and composite) and three sets of replicate-sample weights. In the present study, the composite sample was used (i.e. mail and phone samples were combined) because the relationships of interest were similar in significance level and magnitude of association, which is the acceptable way of determining which sample to use in an investigation (Cantor & McBride, 2007). HINTS Questionnaires The items in the first questionnaire (2003) were developed by NCI investigators and the subcontractor, Westat. Before developing the questionnaire for HINTS, many sources were reviewed to ensure that the data to be covered by this questionnaire were not already being collected elsewhere, i.e. that no other major surveys were currently collecting data with the same focus on both cancer knowledge and behavior and health communication. These sources included major national surveys such as the Behavioral Risk Factor Surveillance System (BRFSS) and the National Health Interview Survey (NHIS), as well as instruments that focused on communication media, such as the Pew Charitable Trust and surveys administered by smaller institutions like departments within states or universities. While these surveys collected information on separate aspects of health and cancer behavior and Internet usage, none of them contained all of the information on all the focus areas of HINTS. A conceptual framework was constructed using research on health communication, medical informatics, psychology, and social ecology (for a graphical representation of the HINTS framework, see Appendix 1). When possible, questions were adapted from other federal surveys. New constructs were drafted and tested within Westat’s cognitive laboratory by in-depth interview sessions with nine respondents that

22

probed these participants about their comprehension and process of answering questions. Scientific validity was considered in the following ways: questions had to be wellestablished as assessing cancer-related information, self-reports could be expected to be reasonably accurate, and the sample size would ultimately be adequate to detect the expected effect size. Data utility was also considered, i.e. questions had to be monitored for health trends and Healthy People 2010 goals, and questions also had to fit into specific criteria requested by NCI staff who were going to analyze the data. Finally, issues of implementation of the questionnaire were considered, especially in regards to the telephone method of question administration, respondent burden, and equitable distribution of questions among main topics selected. The full questionnaire was tested using one-on-one interviews with an additional nine respondents. A telephone field test was also conducted with 172 respondents before the survey was administered. This test was used to check CATI programming and survey administrator training issues, and resulted in questions that were not considered high-priory being deleted from the survey because of issues with interview length. In 2007, several working groups chose priority areas for the survey, and a pool of questions was assembled using HINTS 2003 and 2005 and other questions that addressed the areas of health communication, health services, behaviors and risk factors, cancer, and health status and demographics. Then, the instrument underwent three rounds of semi-structured cognitive interviewing followed by focus groups. Participants (9 volunteers) reported on their comprehension of the questions and any difficulties they had with the instrument. Finally, the mail version of the survey was developed by re-wording items for self-administration. The mail questionnaire also underwent cognitive testing to

23

ensure that the skip pattern was understandable to respondents and that the wording and format were appropriate for self-administration. Outcome Variables Breast Cancer Screening: The outcome variables that were used to test the main hypothesis are dichotomous variables indicating whether participants report being adherent to cancer screening guidelines. Breast cancer screening guidelines recommend clinical breast exams every three years and mammograms starting at age 50; however, the guidelines were revised in 2009 to raise the minimum age from 40 to 50, therefore we used age 40 as the cutoff for screening adherence (U.S. Preventive Services Task Force, 2009). There are no questions about clinical breast exams in HINTS; however, there are questions about mammography. The outcome variable for breast cancer screening combines two questions (ever having a mammogram and if so, when) to create an outcome variable with two possible response levels – women who had a mammogram within two years of the survey, and those who had one more than two years ago or never had a mammogram. The questions used for this variable were asked for women aged over 35 years (this study only marked women non-adherent if they were over age 40 and did not receive mammograms at the appropriate intervals) and who had never been diagnosed with breast cancer at the time of assessment. Participants were also allowed to refuse to answer the question or to response that they did not know. This variable was constructed according to recommendations from the US Preventive Services Task Force (U.S. Preventive Services Task Force, 2009). Cervical Cancer Screening: In the case of cervical cancer, two questions were asked of all women and then combined to create a final dichotomous variable: whether

24

participants have ever had a Pap test, and when was the last time participants had a Pap test. These two responses were then recoded into a single variable with two possible response levels – women who had a Pap smear within the past three years, and women who had a Pap smear more than three years ago or have never had a Pap test. Women were also able to refuse to answer or indicate that they did not know. These response choices were determined based on the recommendations of the US Preventive Services Task Force. The recommendations state that women should get Pap smears from the onset of sexual activity or after age 21 and until the age of 65, and that screening should occur at least every 3 years (Vesco et al., 2011). Because sexual initiation was not ascertained in HINTS, adherence was calculated for women over the age of 18. Colorectal Cancer Screening: The outcome variable for colorectal cancer screenings combines responses to several questions asking about the different modalities of screening. All respondents aged over 45 years were asked questions about whether and when they had undergone a home fecal occult blood test (FOBT), sigmoidoscopy or colonoscopy. These questions were then combined into one variable with two possible levels – participants who have either had a FOBT in the past year, a sigmoidoscopy in the past five years or a colonoscopy in the past ten year were considered adherent to screening guidelines, and participants who had the tests outside the recommended intervals or never participated in any of the test modalities were considered non-adherent. Participants were able to refuse to answer or to respond “Don’t know”. These variables were constructed according to recommendations from the USPSTF (U.S. Preventive Services Task Force, 2008b). Even though the guidelines were updated in 2008, similar

25

guidelines for colorectal cancer screening for adults aged over 50 years were first presented in 2002, before HINTS was first administered. Prostate Cancer Screening: Finally, prostate cancer screening guidelines do not recommend the test for all men. Rather, during the timeline relevant to the data, the USPSTF recommended an informed decision making process should occur between the patient and the healthcare provider. Furthermore, screening was not recommended for men over age 75 (U.S. Preventive Services Task Force, 2008d) Therefore, the outcome variable for prostate cancer screening involved questions about provider conversation and getting a PSA test. There are three questions in HINTS 2003 dealing with prostate cancer screening. Two ask about provider factors – whether respondents talked with a provider the about PSA tests and whether the provider recommended that a PSA test be conducted. The other question refers to having ever undergone the PSA test. These questions were asked of men over the age of 45 who had never had a diagnosis of prostate cancer and who had heard of PSA tests (n = 911). The outcome variable that was used in testing Specific Aims 1, 2, and 3 was having ever had a PSA test. However, the physician recommendation questions were also used to test a different sub-group of men – those who did not get a recommendation (since almost all of the men who got a recommendation went on to get the blood test). The sub-group was defined as men who did not get a recommendation for the PSA test and the outcome variable was having ever had a PSA test. Explanatory Variables Cancer and Health Information Seeking: Cancer and health information seeking was defined using slightly different versions of the same question. In both years,

26

participants were asked whether they ever sought cancer information anywhere and, if so, where they first looked. In 2003, participants who responded affirmatively to a question about general Internet use were also asked whether they looked for health information on-line. However, in 2007, the questions about general health information seeking followed the same format as the cancer information seeking questions. The wording of these questions can be found in Table 2.1. Possible responses to the follow-up question about places where participants looked for cancer information included possibilities such as books, family and friends, doctors, etc. These responses were used to create a variable with three possible response levels: participants who did not look for any cancer information, those who looked to the Internet as their first source of cancer information, and those who used all other sources first. Refusals to answer or answers of “I don’t know” were re-coded as missing. In 2007, the health information seeking question was also re-coded into three categories, but in 2003, the question was a simple yes/no dichotomous variable. These three-level variables were further manipulated to create two indicator (dummy) variables which compared participants who looked for cancer and health information from either the Internet or other sources to those who did not look for information at all (i.e. not seeking health or cancer information from any source was used as the reference category for these analyses). Patient-Provider Communication: The patient-provider communication scale consisted of five questions in 2003 and six questions in 2007.These questions were asked of those participants who reported seeing a healthcare provider in the past 12 months. These questions asked about nurses, doctors, or other health professionals and their behavior during clinical visits – the full list of questions is listed in Table 2.2. The

27

responses to these questions were on a four-point Likert-type scale and ranged from 1 (always) to 4 (never). These responses were re-coded so that ‘never’ became the lowest value and ‘always’ the highest, which meant higher values on the scale indicated better patient-physician communication. This made interpretation easier. The Cronbach’s alpha for the scale in the 2003 administration (using listwise deletion for participants who did not respond to any of the questions in the scale) was α = .86 and was similar in 2007, α = .80. This scale, which was combined into one latent variable, was used to test hypotheses for Specific Aim 2. In order to test hypotheses in Specific Aim 3, a variable that asks participants whether they ever talked to a health care provider (physician, nurse, or others) about any health information they found online was used to split the file into sub-samples of those who did and did not do so. This question was asked of those participants who reported both seeking Internet health information and seeing a health care provider in the past 12 months. A dichotomous yes/no response scale was used. Participants were also permitted to refuse to answer or to state “I don’t know.” Those who responded yes were asked a follow-up question about the level of interest that the patient perceived: “In the past 12 months when you talked with a health care professional, how interested were they in hearing about the information you found online?” This question was adapted from the Pew Internet & American Life Project, Health Care Callback Survey, August 2001 (National Cancer Institute). The responses were on a four-point Likert-type scale with possible responses ranging from 1 (very interested) to 4 (not at all interested). Like the patient-physician communication variables, these responses were re-coded so that higher

28

number indicated a more positive reaction from the healthcare provider to make interpretation easier. This question was used as a mediator variable for Specific Aim 3. Socio-demographic variables: Several socio-demographic variables were used in analyses because they have been previously found to be associated with the cancer screening adherence and behavior. These socio-demographic variables are: age, gender, and household income (as a proxy for socio-economic status). Age was measured as a continuous variable. It was later used to stratify participants into age ranges (corresponding to cancer screening guidelines) in which hypotheses were tested separately (see Analysis section). This was done because age is highly associated with both the independent variables (information seeking, particularly Internet information seeking) and the outcome variables (older people tend to utilize more health care in general and to receive more cancer screenings). Income was also measured as a continuous variable with participants providing an exact number for the household income. These responses were later categorized into 5 levels: less than $25,000, $25,000 to < $35,000, $35,000 to < $50,000, $50,000 to < $75,000, and $75,000 or more were the levels used in 2003. In 2007, the lowest category was changed to be “less than $20,000” while the next highest was listed as “$20,000 to < 35,000.” This was done because the income questions were asked differently on the mail and telephone surveys – all responses were re-coded into the categories that were used in the mail survey. Indicator variables were constructed using these income levels with the lowest income category serving as a reference. Control variables: Control variables were used in the analyses to address possible confounds. Having a family history of cancer was measured by asking the participant

29

whether any family members had ever had cancer. Participants could select “yes,” “no” or “no family” as possible responses. Those who reported no family had their responses marked as missing, given that these participants likely did not know the cancer status of their parents and other relatives. Self-reported health was also used as a control variable, given that those participants who feel in good health may have less reason to see physicians and look for health information online, or to obtain cancer screenings even when they are recommended. The self-rated health item is present in all three years of the questionnaire with the same phrasing (“In general, would you say your health is…”). Responses to this question are on a 5-point Likert scale ranging from “excellent” to “poor” health. Health insurance status was also used as a covariate, because participants who do not have health care coverage may be more likely to use Internet health information as a substitute for seeking health care from a provider. Another possibility is that participants who did not health insurance also did not have access to the Internet and would therefore be less likely to use Internet health information. Finally, those without insurance coverage are less likely to get screened for cancer. This dichotomous yes/no question asks participants whether they have any kind of health coverage (health insurance, HMO, others) and was asked on all administrations of the HINTS questionnaire.

Chapter 3 Analysis HINTS was administered three times, in 2003, 2005, and 2007. The 2003 and 2007 data sets are used in this study because the questions asked and scales used during those administrations are most consistent with the current study’s specific aims. Specifically, the patient-physician communication scale was included only in the 2003 and 2007 administrations. Some of the outcomes were measured in both years: cervical and colorectal cancer screening. Other outcomes were only measured in 2003 – breast and prostate cancer screening. The questions asking about bringing Internet health information to a healthcare provider, and about his or her reaction to that information were only asked in 2007. Finally, all demographic variables, health status, family history of cancer, health insurance status, and Internet health and cancer information seeking were measured in both administrations of HINTS. The 2007 version of HINTS was administered in two modes. The data were collapsed across different modes of administration, as recommended by the HINTS administration team (Ilic, 2010), because no major mode differences were found for the relationships tested in the present study. Preliminary estimations of structural relationships were conducted using the telephone and mail weights, as well as the composite weight which uses data from both modes. The path estimates and model fit were compared across the different modes by looking at whether significant and nonsignificant coefficients remained such across the different data sets and whether the signs and magnitudes of associations were similar. Invaraince testing across modes of administration was no possible due to limitations of the analytical program used and the 30

31

type of variance estimation used by HINTS (jackknifing). All significance levels remained the same and all paths were similar in sign and magnitude, between the two modes of survey administration. As a result, data were combined across the two modes. The analyses addressing the Specific Aims of this study were conducted separately on data from the 2003 and 2007 administrations. Because not all data were available at each administration, not all Specific Aims were addressed with both HINTS data sets. Specific Aim 1 used data from 2003 to test relationships for all cancer screening outcomes, whereas data from 2007 was used to examine associations of the independent variables and only cervical and colorectal cancer screening outcomes. Similarly, for Specific Aim 2, the 2003 data were used to test the extent to which patientphysician communication mediated the relationships between independent variables and all cancer screening outcomes, whereas the 2007 data was used to test possible partial mediation of the relationship between information seeking and cervical and colorectal cancer screening guideline adherence. Finally, for Specific Aim 3, only data from 2007 were used, because the questions that explore perceived physician reaction to health information brought forth by their patients were only asked that year. The patient-provider communication scale consists of five questions in 2003 and six questions in 2007 about the way healthcare providers talk and behave during clinical visits. In the present study, the dimensionality of this scale was tested by performing a confirmatory factor analysis (CFA). The construct of patient-provider communication is hypothesized to be one-dimensional. Scoring higher on this variable is taken to indicate that the participants perceived more positive communication, both verbal and non-verbal, with their healthcare providers. To that end, the responses on the questions comprising

32

these scales were re-coded to reflect this direction of scoring. Latent variables (with five or six indicators), one for each question on the scale were constructed and fitted. Goodness of fit was assessed using Root Mean Square Error of Approximation (RMSEA), and Standardized Root Mean Square Residual (SRMR). The suggested cutoff values for acceptable model fit are RMSEA ≤ .05 and SRMR ≤ .06 (Kline, 2005). Because the analysis using jackknife replication can only use one type of estimation (maximum likelihood for the CFA), the only other fit indices provided are the Akaike Information Criteria (AIC) and the Bayesian Information Criteria (BIC). These indices only allow relative fit comparisons and do not have suggested cutoff values. Furthermore, the relative fit can only be compared when using similar data sets because they are sample-size dependent. Therefore, the AIC and BIC cannot be used to compare the separate scales that were used in 2003 and 2007. Specific Aim 1 and its hypotheses were tested using data from 2003 and 2007, with 2007 data serving as a possible confirmation of the relationships found in the earlier survey administration. The relationships hypothesized between health and cancer information seeking and cancer screening guidelines adherence were tested using probit regression. Because age is highly associated with both the independent and dependent variables, but in opposite directions (i.e. younger participants were more likely to seek cancer/health information and specifically Internet cancer/health information, while the older participants were more likely to get screened for cancer), the data were stratified on age using two or three strata (depending on the cancer in question) and the relationships were tested in each stratum. For breast cancer, participants were separated into women aged 40-50, those aged 51-65, and finally those aged 66 and up. The 40 year old cutoff

33

was used because of the USPSTF guidelines for breast cancer screening at the time of survey administration. For cervical cancer, participants were separated into three strata: ages 18-35, ages 36-50, and ages 51-65. Data from 18 year old were used because the USPSTF guidelines recommended getting screened at most 3 years after first sexual contact and no information was available about the women’s sexual initiation – using the lower age cutoff would capture those women. For colorectal cancer, participants were stratified into 2 age strata: 50 to 60, and 61-75. Finally, for prostate cancer, men over age 45 were used in the analyses in two groups – ages 45-60 and ages 61 and older. These analyses were further stratified by physician advice, because men whose physicians recommended getting tested almost uniformly got tested, whereas those who did not get a recommendation nevertheless received PSA tests at a fairly high rate. Simple unadjusted estimates of association were first examined, and then control variables were added into the analysis. The hypothesized model for Specific Aims 2 and 3 was tested using structural equation modeling (SEM), a technique that allows for estimation and testing of multiple theorized or hypothesized processes at once. The model tested is pictorially represented in Figure 1.1 for Specific Aim 2 and in Figure 1.2 for Specific Aim 3. SEM is a confirmatory technique and should therefore have strong theoretical and prior research grounding (Byrne, 2012; Kline, 2005). It is also important to test plausible alternative models that postulate different relationships between and among the variables selected for analysis (Vandenberg & Grelle, 2009). The alternative model tested reverses the order of mediation and is pictorially represented in Figures 3.1 and 3.2. Model fit for these models was assessed using only one fit index, the Weighted Root Mean Square Residual

34

(WRMR), because this fit index is the only one available from Mplus for SEM with jackknife weights (weighted least square parameter estimates using a diagonal weight matrix with mean- and variance-adjusted standard errors, or WLSMV). The accepted cutoff values for good fit for WRMR is less than 1.0 (Yu, 2002). Because SEM is highly theory-driven and because removing important variables could confound the associations of interest, I did not remove non-significant covariates from analyses. All of the covariates (health insurance status, self-reported health, family history of cancer, income, sex, and patient-physician communication) were hypothesized to relate to cancer screening adherence or have been previously found to be important when analyzing cancer screening rates. However, the decision to leave non-significant covariates in the model may have impacted model fit negatively. A mediation SEM framework was used to test the hypothesis that patientphysician communication partially mediates the relationship between Internet health and cancer information seeking and cancer screening adherence (Specific Aim 2). Full mediation would imply that the direct association between the predictor (Internet health information) and the outcome (cancer adherence) becomes non-significant when a mediating variable (patient-physician communication) is added to the model (MacKinnon, 2008). Both full and partial mediation models were tested (Holmbeck, 2002). A similar test was used to test the hypothesis that the relationship between patientphysician communication and cancer screening adherence may be partially mediated by the health-care provider’s response to Internet health information presented during a health-care visit (Specific Aim 3).

35

The outcome variables for most of the hypotheses of the present study are dichotomous. Therefore the models in this study were tested using probit modeling. Probit regression uses the cumulative normal function as the link function between a binary outcome and various predictors. Generally, both probit and logit link functions can be used to analyze outcome variable with a binomial distribution (Dobson, 2002). Logit and probit functions fit very similarly and provide basically identical conclusions, especially when estimated using data with many observations (Powers & Xie, 2008), but the interpretation of the regression coefficients is different for the two types of models. The coefficient for the predictors in a probit model essentially provides an indication of change in the z-score of the outcome for a one unit change in the independent variable. Although logistic regression lends itself to easier interpretation in the form of odds ratios, the probit function was used in this analysis because it is the only type of analysis available in Mplus when using jackknife replication variance estimation, as is needed with HINTS data (Muthen & Muthen, 2010). All analyses, including descriptive statistics, confirmatory factor analysis, and structural equation modeling, were carried out using Mplus 6.11 (Muthen & Muthen, 2010). Partial mediation was tested using the asymmetric distribution of products test (MacKinnon, 2008) and the PRODCLIN software (MacKinnon & Fritz, 2007) for only those relationships in which the paths between either the independent variable and mediator or mediator and outcome were significant. This technique evaluates mediation by multiplying the unstandardized path coefficients from the SEM and computing the standard error for this product using the coefficients and standard errors for the two paths and reporting a 95% confidence interval. The calculation uses the unstandardized direct

36

path coefficients and their standard errors and the correlation between the two variables. The proportion of the overall effect that is mediated effect was also calculated in cases where significant partial or full mediation was found (MacKinnon & Fritz, 2007).

Chapter 4 Results Participant demographics and descriptive statistics Demographic and descriptive statistics were calculated using full samples from each of the two years of HINTS that were used in this study (i.e. in 2007, combined mail and phone samples were used). Furthermore, all statistics were calculated using the provided jackknife replicate weights in order to adjust for sampling, non-response, and national representativeness. Tables 4.1 and 4.2 list the means and standard deviations, or proportions, of all the variables used in the analyses and the racial/ethnic make-up of these samples. The samples from 2003 and 2007 were similar demographically by design, since the estimates are weighted to be nationally representative. The mean age was close to 45 in both samples. Women made up about 51% of the sampled participants. Hispanics comprised about 12-13% of the participants, while Blacks comprised 8-11% and Whites comprised 65-67%. These percentages are similar to national estimates of the racialethnic make-up of the United States population. About a third of the participants reported a household income of less than $35,000, and close to a quarter reported a household income of over $75,000. Participants who were asked questions about their overall rating of patientphysician communication seemed to be largely satisfied. They endorsed high agreement with the items on this scale, with very few participants marking the “Sometimes” or “Never” options. This may reflect the general nature of the questions that comprise this scale. Many people visit multiple physicians’ offices and may interact with a variety of 37

38

healthcare providers, including nurses, physician’s assistants, and others, during their healthcare visits. The answers to the questions on this scale may indicate the satisfaction patients feel in their communication with any of these healthcare providers. Patients with health insurance coverage also tend to have at least some freedom in their choice of physicians/healthcare providers, and the positive rating of patient-physician communication may be due to patients choosing the doctors they like. The majority of participants reported having health insurance (85% in 2003 and 82% in 2007), although a substantial number of participants did not have any type of health insurance (including Medicare, Medicaid, or other forms). Participants also responded fairly positively when asked to rate their health, with the mean levels of health falling between “Good” and “Very Good.” Finally, few participants who had seen their doctors in the past 12 months reported talking to them about Internet health information – only 25% in 2007 (the only time that question was asked on HINTS). Measurement model for patient-physician communication A confirmatory factor analysis (CFA) was conducted on the patient-physician communication items, r to create a single latent variable to be used in all subsequent analyses. The scale consisted of either five (in 2003) or six (in 2007) questions about patients’ perceptions of health providers’ communication during visits. We estimated the CFA using data from all participants, but list-wise deletion of missing data was used for those who did not receive questions about patient-physician communication (i.e. all participants who did not report seeing a physician in the last 12 months). The CFA was conducted using maximum likelihood estimation.

39

For the 2003 data, the CFA indicated that a model with a single patient-physician communication latent variable fit the data well: AIC = 50,485.14, BIC = 50,583.89, RMSEA = .04 (95% CI = .03 – .05), SRMR = .01. These model fit statistics all fall under the recommended indication of good fit (below .05 for RMSEA and SRMR (Kline, 2005)). The standardized path coefficients for the individual questions that comprised the scale ranged from .71 to .77. For the 2007 data, the CFA indicated a fairly good fit to the model of a single patient-physician communication latent variable: AIC = 78,636.46, BIC = 78,759.09, RMSEA = .07 (95% CI = .06 – .08), and SRMR = .02. The standardized path coefficients for the individual variables ranged from .71 to .81. The path coefficient for the final question on the scale (“In the last 12 months, how often did you feel you could rely on doctors, nurses or other health professionals to take care of your health care needs?”) was consistently lower than the others, but it was nevertheless above the commonly accepted threshold of at least .6 (Kline, 2005). Specific Aim 1 – Associations between information seeking and cancer screening The associations between cancer or health information seeking and cancer screening adherence were conducted using simple and multivariable probit regressions, accounting for weights used by HINTS. Cancer and health information seeking were tested as separate predictors of cancer screening adherence in each data set. Furthermore, because different cancers have different ranges of ages at which screening is recommended, analyses were conducted separately for each cancer and age range. For breast cancer, the unadjusted models regressing cancer information seeking on cancer screening adherence fit the data extremely well – WRMR = 0.003 for women

40

aged 40-50; WRMR = 0.005 for women aged 51-65; and WRMR = 0.007 for women aged 66 and up. The fully adjusted models fit the data fairly well for the younger and middle age groups (WRMR = 1.036 and WRMR = 1.067 respectively), but not well for the older age group (WRMR = 1.153). This difference in fit suggests that the inclusion of non-significant covariates hurt the fit of the model. The probit regression coefficients for both models are reported in Table 4.3. While the unadjusted regression showed a significant positive association between cancer Internet information seeking and onschedule mammography is women aged 40-50 and 51-65 and information seeking from other sources and mammography is women aged 40-50, those associations became nonsignificant with the addition of control variables. The unadjusted models regressing health information seeking on mammography adherence fit the data extremely well – WRMR = 0.003 for women aged 40-50; WRMR = 0.000 for women aged 51-65; and WRMR = 0.002 for women aged 66 and up. However, the fully adjusted models did not fit the data adequately: WRMR = 1.230 for the youngest group; WRMR = 1.125 for the middle group; and WRMR = 1.194 for the oldest group. Path coefficients are reported in Table 4.4. No significant adjusted associations between health information seeking and breast cancer screening adherence were found. For cervical cancer screening and cancer information seeking, the unadjusted models fit the data very well both in 2003 and 2007, with WRMR ranging from 0.002 or 0.005 for all age ranges in both years. However, the fully adjusted models showed good or adequate fit in only two cases, women aged 66 and older in 2003 (WRMR = 1.077) and women aged 18-35 in 2007 (WRMR = 0.924). The probit regression coefficients are

41

reported in Table 4.5. Internet cancer information seeking when compared to no information seeking was found to be significantly positively associated with getting Pap smears on schedule among women aged 18-35 and 36-50 in the 2003 data when adjusted for control variables. In the analyses that regressed health information seeking on Pap smear adherence, the unadjusted simple regression models showed good fit, with WRMR ranging from 0.001 to 0.007 for both 2003 and 2007. In the adjusted analyses, only one age strata showed adequate fit to the data – women aged 18-35 in 2007, with WRMR = 1.024. The regression coefficients are reported in Table 4.6. In the 2003 data, Internet health information seeking was positively associated with higher rates of on-schedule Pap smears for women aged 36-50 and a similar association was found in the 2007 data for women aged 18-35. In the analyses that examined the associations between cancer information seeking and colorectal cancer screening adherence, model fit was quite good. WRMR was equal to 0.002 for both age groups in 2003 and 2007. In the adjusted analyses, the model fit was not very good: WRMR = 1.192 for people aged 50-60 in 2003; WRMR = 1.273 for people aged 61-75 in 2003; WRMR = 1.457 for people aged 50-60 in 2007; and WRMR = 1.274 for people aged 61-75 in 2007. The path coefficients for the relationships of interest in these models are reported in Table 4.7. In the 2007 data, seeking cancer information from the Internet or from other sources was associated with higher rates of getting screened for colorectal cancer according to guidelines. The unadjusted associations between health information seeking and colorectal cancer screening adherence fit to the data well, with WRMR ranging from 0.002 to 0.006

42

for all age categories in both years. Good and adequate fit was exhibited in the fully adjusted models in 2003 – WRMR = 1.059 for participants aged 50-60 and WRMR = 0.954 for participants aged 61-75. However, model fit was not very good in 2007 – WRMR = 1.355 for ages 50-60 and WRMR = 1.304 for ages 61-75. All of the regression coefficients for the relationships of interest are listed in Table 4.8. In the 2007 data, Internet health information seeking was associated with higher rates of colorectal cancer screening adherence in people aged 61-75. Finally for prostate cancer screening, the unadjusted models regressing cancer information seeking on ever getting screened fit the data well with WRMR = 0.002 for the both age groups. However, the fully adjusted model did not fit the data well: WRMR = 1.480 for men aged 45-60 and WRMR = 1.210 for men aged 61 and up. The models testing the association between health information seeking and prostate cancer screening fit the data well: WRMR = 0.002 for younger men and WRMR = 0.001 for older men. The fully adjusted health information seeking models also provided good fit with WRMR = 0.863 for the younger group and WRMR = 0.960 for the older. Regression coefficients for both types of models are displayed in Table 4.9. Men aged 45-60 were found to get PSA screening more often if they reported Internet health information seeking. Because almost all men over age 45 who get a recommendation from their healthcare provider to get a PSA screening do so, we also separately analyzed the group who did not receive such recommendation. These men were not stratified by age because the group was quite small (n = 264). Once again, the unadjusted models provided good model fit: WRMR = 0.001 for the association between cancer information seeking and screening and WRMR = 0.002 for health information seeking and screening. The fully

43

adjusted models also had good fit: WRMR = 0.849 for cancer information seeking and WRMR = 0.944 for health information seeking. Results of the analysis are displayed in Table 4.10. No significant associations were found between information seeking and PSA screening. Specific Aim 2 – Mediation of relationship between information seeking and cancer screening To test the hypothesis that patient-physician communication would partially mediate the relationship between information seeking and cancer screening by patientphysician communication, SEM models to all subgroups of participants were estimated. This was done to address possible issues of confounding of the direct relationship between information seeking and cancer screening adherence. There may have been indirect paths between independent variable and outcome that had an opposite relationship than the one found by testing the direct paths. Decomposing that path allowed the understanding of the actual effect of patient-physician communication as a mediator. All paths that were found to be significant were reported, even if partial mediation was not significant, in order to determine the effect that the mediator may have on the outcome and the relationship between independent variable and outcome. In addition, since SEM is a confirmatory technique, an alternative model of mediation was tested as pictured in Figure 3.1 for all sub-groups. There was no way to directly compare the alternative and main models in terms of fit criteria, but the paths and fit were examined. None of the alternative models had good fit, except in the case of one model that demonstrated adequate fit – in the model that tested the relationship between the three variables of interest for cervical cancer screening adherence in women aged 18-

44

35. However, none of the paths were significant in this model, and no paths were significant in all of the alternative models that weren’t also significant in the main models. Therefore, we concluded that the main model tested was the one that would be retained for this work. Model fit was determined the same way as in Specific Aim 1, by using the WRMR fit index, which should be below 1 to indicate adequate fit. The only model with any significant paths in which the relationship between mammography and information seeking were partially mediated was for women aged 40-50, and only for health information seeking. WRMR in this model was 1.196, showing poor fit. The models testing mediation of relationships for cervical cancer screening mostly did not fit the data adequately except for the model testing the association of cancer information seeking and screening in women aged 51-65 (2003) where the WRMR was close to the cutoff at 1.043. The model linking colorectal cancer screening and health information seeking had close to adequate and good fit respectively: WRMR = 1.018 for the group aged 50-60 and WRMR = 0.894 for the group aged 61-75 using 2003 data. The models specifically examining cancer information seeking did not fit the data well, nor did any of the mediation models using data from 2007. Finally, prostate cancer screening mediation models fit the data adequately when general health information seeking was used as a mediator (WRMR = 0.850 for the younger men and WRMR = 0.958 for the older men), but not when health information seeking was used as a mediator. The results of the analyses testing for partial mediation with breast cancer screening adherence are displayed in Table 4.11. The tables in this section present the three components of mediation: the regression of the independent variable on the

45

mediator, the regression of the mediator on the outcome, and the regression of the independent variable on the outcome (MacKinnon, 2008). The results of the partial mediation testing of the relationship between information seeking and cervical cancer screening adherence are shown in Table 4.12. The results of the analyses testing partial mediation of the relationships between cancer/health information seeking and colorectal cancer screening adherence are displayed in Table 4.13. The regression coefficients for all paths in the partial mediation model for prostate cancer screening and health information seeking are displayed in Table 4.14. All of the mediation analyses were conducted while controlling the path between either type of information seeking and cancer screening for self-reported health, health insurance status, family history of cancer, income, and sex (for colorectal cancer screening). Although the direct relationships between information seeking (whether cancer or health information) and cervical cancer screening were significant in the multivariable regression model and therefore warranted a partial mediation test, no mediation was found in any of the age subgroups. As presented in Table 4.12, no direct paths between independent variable and mediator, or between mediator and outcome, were significant, and the direct relationships between information seeking and cancer screening in many cases remained significant. This may be one of the reasons for poor model fit. One instance of partial mediation was found. The relationship between cancer information seeking and colorectal cancer screening adherence was partially mediated by patient-physician communication for participants aged 50-60 (in 2007). The indirect unstandardized path between Internet information seeking and outcome was B = -0.024 (95% CI = -0.048 to -0.005) and the path between other types of information seeking and

46

outcome was B = -0.025 (95% CI = -0.054 to -0.003). No other significant partial mediation was found, although there were significant positive associations between patient-physician communication and screening adherence in the rest of the sub-groups. The relationship between health information seeking and prostate cancer screening was tested for partial mediation, but there was no indication of patientphysician communication serving as a partial mediator. The significant negative relationship between Internet health information seeking and PSA screening remained for men aged 45-60 and there was a significant negative association between patientphysician communication and PSA screening for men over 61. Specific Aim 3 – Testing the impact of physician response to health information brought by the patient This specific aim was tested only with participants who reported bringing health information to their physicians. Specific Aim 2 was also only tested using data from the 2007 survey administration, because it was the only one that asked about bringing information to a physician and the physician’s reaction. Participants were once again stratified by age, as in previous analyses. The test of hypothesis 3a used multivariable probit regression with the physician’s reaction serving as the independent variable and cancer screening serving as outcome. Other control variables include the patient-physician communication latent variable, income, health insurance status, family history of cancer, and self-reported health. Model fit was good for the fully adjusted model using the sub-group of women aged 18-35 and testing the relationship between perceived physician reaction and cervical cancer screening – WRMR = 0.847, but was not adequate for any of the other outcomes or age

47

groups (including all tests involving colorectal cancer adherence). The regression coefficients for perceived physician reaction are reported in Table 4.15 for cervical cancer screening adherence and in Table 4.16 for colorectal cancer screening adherence. No significant associations were found in the relationship between perceived physician reaction and either cancer screening adherence. Furthermore, it was hypothesized that the relationship between patient-physician communication and cancer screening would be partially mediated by perceived physician reaction in those participants who talked to their healthcare providers about health information they found. Because there was an indication that patient-physician communication was related to cancer screening in most cases, all of the subgroups were tested. For models testing mediation of the relationship between patient-physician communication and cervical cancer screening adherence, model fit was good: WRMR = 0.879 for women aged 18-35, WRMR = 0.853 for women aged 36-50, and WRMR = 0.904 for women aged 51-65. The path coefficients for each of the paths relevant to the partial mediation model are listed in Table 4.17. A significant negative relationship was found between perceived physician reaction to Internet health information and cervical cancer screening adherence for women aged 18-35. There was also a positive relationship between patient-physician communication and perceived physician reaction to Internet health information for women aged 36-50 and 51-65. Since SEM is a confirmatory technique, the partial mediation hypothesis was tested using an alternate model pictorially represented in Figure 3.2, in which patientphysician communication acted as the mediator of the relationship between perceived physician reaction to Internet health information and cancer screening adherence for

48

cervical and colorectal cancers. Once again, the fit of the alternative models were similar to the fit exhibited by the main theorized models, and no different path significances were found. Therefore, the alternative models were discarded for the purposes of this study. Similarly to Specific Aim 2, the significance of the indirect path between independent and dependent variable (through the mediator) can be calculated with PRODCLIN. As can be seen in Table 4.16, at least one of the paths between independent variable and mediator or mediator and outcome proved significant for cervical cancer screening; therefore, the partial mediation test was done for all three age subgroups. However, there was no indication of significant partial mediation. The models testing partial mediation of the relationship between patient-physician communication and colorectal cancer screening adherence by the reaction of the physician to health information also fit the data well – WRMR = 0.845 for participants aged 50-60 and WRMR = 0.881 for participants aged 61-75. Regression coefficients for all paths of interest in the models are displayed in Table 4.18. Significant positive relationships were found between patient-physician communication and perceived physician reaction in both age groups, and the relationship between patient-physician communication and screening adherence remained positive for people aged 50-60. The indirect path between patient-physician communication and colorectal cancer screening adherence was once again found and tested for significance, but no partial mediation was found. Post hoc analysis When examining the results from the analyses in Specific Aim 3, I noticed that rates of screening varied by whether patients brought Internet health information to

49

physicians or not, so this was tested for statistical significance. Results indicated that, in a model testing the relationship between bringing information to a physician and cervical cancer screening, women aged 18-35 were more likely to receive the Pap test on-schedule if they brought Internet health information to their physicians, compared to those women who did not (WRMR = 1.026, B = 0.492, p = 0.017). No other significant relationship between bringing Internet health information to a physician and cancer screening adherence emerged in the other age strata for cervical cancer screening or for colorectal cancer screening adherence.

Chapter 5 Discussion The effect of use of the Internet for cancer and health information seeking on cancer screening adherence and behavior appears to differ by type of cancer and by patient age group. Patient-physician communication, while proving not to serve as a mediating variable in most cases, has been shown to directly predict cancer screening adherence. And finally, the reaction of a healthcare provider to Internet health information presented by a patient also has some effect on cancer screening. These three factors combine to create a complex picture of cancer screening, information seeking, and physician visit behavior in a nationally representative sample of adults. Breast cancer screening Even though there was some indication of significant associations between cancer information seeking (whether from the Internet or other sources) and adherence to mammography screening guidelines when not adjusted for other factors, the addition of variables like health insurance, self-reported health, income, family history and patientphysician communication reduced these associations to non-significance. The strongest predictor of getting a mammogram on schedule was health insurance in the younger (4050) and middle-aged (51-65) groups, whereas income played somewhat of a role in mammography screening for the older group (over 65). Breast cancer is somewhat different from the other cancers whose screening was tested in this study. Awareness of breast cancer screening has long been a goal of some very visible and influential organizations, such as the Susan G. Komen Foundation. These organizations have spent much time, money and energy to increase women’s

50

51

awareness of breast cancer screening and treatment ("Susan G. Komen for the Cure," 2012) and have brought attention to cancer through activism, research funding, and corporate sponsorships. The philosophy of these organizations is to focus on early detection through screening, both by clinical breast exams, self-exams, and mammography (American Cancer Society, 2011a; "Susan G. Komen for the Cure," 2012). Research has shown that campaigns like Breast Cancer Awareness month have an effect on Internet searches about breast cancer (Glynn, Kelly, Coffey, Sweeney, & Kerin, 2011) and that women receive memorable messages about breast cancer awareness and screening at quite high rates from media and family which in turn are associated with higher rates of screening behaviors (S. W. Smith et al., 2009). The visibility of breast cancer that has come about because of Komen and other organizations and the pink ribbon used to symbolize the disease may have made it difficult to detect an effect of information seeking, or any other independent variable. There have been criticisms of the use of breast cancer awareness for political (King, 2010) and commercial gain ("Breast-cancer awareness: too much of a good thing?," 2007), but there is no doubt that the disease has become well known to a wide population. The challenge remains that health insurance availability and socio-economic status still affect mammography utilization. Another recent development that may affect breast cancer screening rates is the pushing back of the recommended age for population screening from 40 to 50 (H. D. Nelson et al., 2009). The controversy that this decision generated, and the fact that organizations like Komen ("Susan G. Komen for the Cure," 2012) and the American Cancer Society still recommend that women get screened

52

starting at age 40 (American Cancer Society, 2011a), will warrant further study of women’s mammography utilization in the future. Cervical cancer screening Use of the Internet for cancer information seeking (when contrasted to no information seeking of any kind) was positively associated with cancer screening adherence in women aged 36-50 and 51-65 in HINTS 2003. Similarly, the more general “health information seeking” variable was associated with cancer screening adherence in women aged 36-50 in 2003 and women aged 18-35 in 2007. Seeking information from sources other than the Internet was not significantly associated with the outcome. This result seems to suggest that, across age ranges, those women who seek Internet information about health in general or cancer specifically are more likely to report receiving a Pap smear within the last 3 years. As in the models examining breast cancer screening, having health insurance was positively significantly associated with screening in women aged 36-50 and 51-65, as was income. In particular, among younger women (age 18-35) participants who reported household incomes of between $50,000 and $75,000 and above $75,000 were much more likely to get screened on schedule than those women who earned below $25,000. Again, the importance of having health insurance is highlighted, as is the effect of SES. The significant positive associations between predictor and outcome remained when patient-physician communication was entered into the model as a potential mediating variable. The magnitudes of the associations between information seeking and cervical cancer screening adherence remained almost the same with the addition of the mediating variable. This implies that no mediation was present, which was further

53

confirmed through formal mediation testing. The lack of mediation notwithstanding, these results seem to uphold the conclusion that Internet information seeking is indeed associated with cervical cancer adherence for women age 50 and younger. The question lies in why the association between Internet information seeking and screening was not found in 2007, in the third administration of HINTS. Data from the 2005 administration of HINTS, which was not used in the present analyses, suggested a similar result in that women who had ever searched for cancer information were more likely to be cervical cancer screening “maintainers” (i.e. have been screened in the past 3 years and intended to get screened in the next 3 years) when controlling for a variety of variables including age (W. Nelson, Moser, Gaffey, & Waldron, 2009). However, although source of cancer information seeking was available in the data, the study only looked at information seeking from any source. Factors associated with cancer screening adherence in 2007 were once again health insurance and, in the case of women aged 36-50, self-reported health (women who reported better health were more likely to be adherent to screening guidelines). These two factors were strongly associated with the outcome, accounting for most of the variance explained in the Pap screening variable. In the 2003 analysis, self-reported health was not associated with adherence in the multivariable probit regression. Interestingly, the rates of screening in 2003, the year in which the significant associations were found, seem to be about equal in all the age strata – about 80% of women reported getting screened within the past 3 years. However, in 2007, screening rates differed by age group. Women aged 18-35 and 51-65 reported getting screened similarly to those in the first administration of HINTS, about 78 or 79%, whereas the 36-

54

50 age group reported higher rates of Pap smears in the past 3 years – almost 88%. In the 2005 administration of HINTS, W. Nelson et al. (2009) found that the highest rates of cervical cancer screening “maintainers” were in the 25-34 age group (88%) and lowest in the 46-64 age group (82%). Another study using data from the National Health Interview Survey (NHIS) from 2000 found a similar pattern to the HINTS 2007: past 3-year screening rates were 84% for women aged 50-70, 88% for women aged 30-50, and 80% for younger women (Solomon, Breen, & McNeel, 2007). It may be that slight variations in rates of screening are evident in most years of data collection, and that the 2003 screening rates which were almost exactly equal are the anomaly. It is not entirely clear how this difference affects the regression and structural equation models being tested in Specific Aims 1 and 2. The Papanicolaou test was first introduced in the 1950’s and 1960’s and resulted in large decreases in incidence and mortality of cervical cancer (Waxman, 2005). This test became widely known shortly after it was introduced, especially once it became yoked to a regular gynecological visit, which was needed for the prescription of contraceptives in the 1960’s (Walton et al., 1976). In that respect, the Pap test is similar to mammography and breast cancer screening, because they are both well-known and the rates of screening are generally quite high. However, given that the Pap test has been in use longer, it may be that the knowledge about this type of screening is gained not through active seeking on the part of the test recipient, but rather through other sources (family, health education in schools). Breast cancer screening is currently very visible because of the ongoing campaigns to raise awareness and funds by Komen and other groups, whereas cervical cancer screening may perhaps be assumed to be known to

55

everyone. Research has found, however, that there may be an increase in some barriers to screening, especially in younger women – particularly, awareness about the Pap test seems to be decreasing (L. Finney Rutten, Nelson, & Meissner, 2004). This may point to a need for intervention in specific vulnerable groups. It does nevertheless seem that women who are active participants in their own healthcare (through Internet information seeking) are more likely to be screened for cervical cancer. This result implies that it is important to create more involved patients, perhaps through public health campaigns or direct conversation with physicians. Another implication of the findings is the importance of accurate and clear Internet health information, since there is an impact of looking for this type of information on health behaviors like cancer screenings. Another issue in cervical cancer is human papillomavirus (HPV). This virus is known to cause many common types of cervical cancer, and with the advent of the HPV vaccine, screening patterns may have changed or may change in the future. In fact, cervical cancer screening guidelines were recently changed to include HPV screening (U.S. Preventive Services Task Force, 2012). The HPV vaccine was first approved by the U.S. Food and Drug Administration in 2006 (Food and Drug Administration, 2006). Since that time, much media attention has been allocated to the vaccine and cervical cancer and there has been a national conversation about the vaccine and its acceptability. This may have had some repercussions on the information gathered on HINTS 2007. It may be that the link between HPV and cancer was not yet widely known during the 2003 and 2005 data collections, and participants therefore may not have endorsed the cancer information seeking item when they were looking for or hearing about information on the HPV vaccine. Or it may be that since participants were hearing a lot about the HPV

56

vaccine and cervical cancer, they did not seek health or cancer information relating to this subject. Either of these may be reasons why no association was found between cancer information seeking and screening adherence in 2007, though the link was found in 2003. Colorectal cancer screening Colorectal cancer screening adherence was positively associated with Internet cancer and health information seeking in 2007, but not in 2003. Interestingly, for participants aged 50-60, specifically cancer information seeking was found to be associated with the colorectal cancer screening adherence, whether the source listed was the Internet or other sources. But for participants aged 61-75, it was more general health information seeking that proved to be significantly associated with the outcome, again both from the Internet and from other sources. It seems that, unlike cervical cancer screening, for which Internet information seeking specifically was the important predictor, colorectal cancer screening adherence is higher among people who seek information from any source. A previous study using HINTS 2003 data by Ling et al. (2006) found an association between cancer information seeking and colorectal cancer screening adherence, but these authors did not separate the information seeking by source. As in previous analyses, health insurance was once again a strong predictor of cancer screening adherence for all age groups, and for both types of information seeking in 2007. Family history of cancer was also significantly predictive of colorectal cancer screening adherence. Having a family history of cancer was positively associated with getting screened for colorectal cancer on schedule, but only for the younger age group. In 2007, patient-physician communication was also associated with both cancer and health

57

information seeking. This pattern of findings suggests that participants who reported better communication with their healthcare providers in general were more likely to follow colorectal cancer screening guidelines. In 2007, it is also apparent that the older participants (aged 61-75) were, in general, more adherent to colorectal cancer screening guidelines (FOBT, sigmoidoscopy or colonoscopy), with a rate of 73%, compared to the lower rate of 56% in the younger age group. This supports findings from the 2008 National Health Information Survey in which participants aged 66-74 were 1.64 times more likely to get either endoscopy in the last 10 years, or FOBT at home in the last 1 year, compared to their counterparts aged 5064 years (Shi, Lebrun, Zhu, & Tsai, 2011). Further analysis of of the colorectal cancer screening adherence variable indicated evidence of partial mediation by patient-physician communication in the younger age stratum. The mediated effect, which can be computed by multiplying the direct paths between cancer information seeking and the mediator and the mediator and screening, was negative. This was because the relationship between cancer information seeking (from any source) and the latent patient-physician communication was also negative, implying that participants who sought out cancer information reported worse patientphysician communication levels. However, patients who reported higher patientphysician communication levels were more likely to get screened for colorectal cancer on-schedule (positive direct path). Overall, the total unstandardized mediated effect was small but difficult to interpret because of the unequal scales for information seeking and patient-physician communication and a significant relationship remained between independent variable and outcome. Because the unmediated path between cancer

58

information seeking and colorectal screening adherence did not change much with the addition of the mediator, a reasonable assumption is that the mediated effect is small. Practically, this may mean that patient-physician communication is important for colorectal cancer screening adherence, especially given that in the other mediation models tested it also showed a significant positive association with the outcome for all age groups and both cancer and health information seeking. One possible interpretation of a negative indirect path between information seeking and screening adherence may be that those patients who talked to their physicians about health information they searched for tended to get negative reactions from their physicians and this may have led to a lowered likelihood of getting screened on schedule. This may have specifically affected colorectal cancer screening rates because the type of tests performed (endoscopy, specifically) require more discussion and planning than other tests like the Pap test or mammography. Colorectal cancer screening rates are lower than those of other universal screenings like cervical and breast cancer, although they have been rising – in 2000, only about 34% of the eligible US population was getting screened for colorectal cancer, whereas in 2008 the rate rose to over 50% (Rim et al., 2011; Shi et al., 2011). In this study, we found the rates to be between 53% and 58% in 2003 and between 56% and 73% in 2007. Unlike breast cancer, which receives a lot of attention from non-profit and public health foundations, and cervical cancer screening, which has been used for more than 50 years, colorectal cancer does not seem to receive the same level of advocacy and attention. Colorectal cancer screening guidelines also seem to be more complex than the others – there is a choice of modalities which patients may not feel equipped to make

59

properly and a different schedule of screening for each modality, which may be confusing. Patients who view colorectal cancer screening as difficult to arrange, and/or those who do not know about screening methods, are less likely to adhere to screening guidelines (Berkowitz et al., 2008). On the other hand, when patients receive a recommendation for screening by a physician, they tend to be screened on-schedule (Subramanian et al., 2004). It seems that age may also play a role in who gets screened and who does not, but this may simply be an artifact of patients being enrolled in Medicare once they turn 65, given that health insurance (or lack thereof) is highly associated with screening adherence, both in the current study and in others (Rim et al., 2011). Although colorectal cancer screening rates seem to be rising, they are still much lower than those of the other universally-recommended cancer screenings. In the current study, I found that both information seeking and patient-physician communication are positively associated with colorectal cancer screening adherence. This implies that patients who are engaged in their healthcare and who have good relationships with their physicians and other healthcare providers are also more likely to get screened, since screening is an important part of caring for oneself. Similar results were found for cervical cancer screening in terms of Internet information seeking. Given that the present study was conducted using cross-sectional data, further prospective or experimental studies are needed to shed further light on the research questions examined in the present study. There are three possible modalities for colorectal cancer screening: FOBT, sigmoidoscopy, and colonoscopy. The current study does not distinguish among patients who choose the different types of screening, instead classifying them as adherent if they

60

chose any of the three methods and followed the timeline of the screening guidelines. However, patients who choose a procedure that must be performed in a doctor’s office (i.e. sigmoidoscopy or colonoscopy) may be different from those patients who choose to perform the screening at home (i.e., FOBT). The screenings are qualitatively different, and thus the associations between information seeking, patient-physician communication and cancer screening adherence may vary across the different screening modalities. Further study is needed. Prostate cancer screening Prostate cancer screening at the universal level is a controversial topic. There are currently large randomized controlled trials that are being done to determine whether PSA testing should be recommended to all men over a certain age. The United States Preventive Services Task Force is currently accepting public comment on a draft recommendation which would state that the agency is against the screening for men who do not exhibit symptoms of prostate cancer (U.S. Preventive Services Task Force, 2011a). The USPSTF recommendations state that there is not enough evidence for or against PSA screening (U.S. Preventive Services Task Force, 2008d). Nevertheless, men are being screened using this modality at fairly high rates – in this study, about 67% of younger men (aged 45-60) were screened and more than 85% of older men (aged 61 and older) were screened. In another study suing HINTS 2003 data, L. J. Finney Rutten, Meissner, Breen, Vernon, and Rimer (2005) found that about 55% of men received PSA screening and in a large, randomized controlled trial of universal PSA screening, both the experimental and control groups had fairly high rates of screening – 85% in the experimental group and 52% in the control group in the 6th year of the study (Andriole et

61

al., 2009). The higher screening rates in the current study may have emerged because of the ways in which prostate cancer screening was operationalized – only men who were asked, and who answered questions, about information seeking and physician recommendation of the PSA test were included in the present analysis. Among men who did not receive a recommendation from a healthcare provider, about 61% of men over 45 received a PSA test. Nevertheless, these are all quite high rates for a test with uncertain benefits and known harms. Similarly to the other cancer screenings examined in the current study, Internet health information seeking was positively associated with getting a PSA screening. In the partial-mediation model, both Internet information seeking was positively associated with PSA screening rates, while patient-physician communication were negatively associated with PSA screening. Although mediation was not found in this model, the associations found are important. While it is not possible to determine causality, the results are interesting. Those who look for health information online are more likely to get screened, even though the benefits of this screening are not certain. This may reflect a lack of balanced information on harms and benefits of this screening modality on the Internet or in other informational materials. It’s not possible to certify what kind of information participants in HINTS looked for when they reported seeking cancer or health information, whether from the Internet or other places. The controversy about PSA screening has been ongoing for a few years – although the HINTS data are from 2003, the uncertainty about PSA screening and the conclusion that there was insufficient evidence to recommend screening from the USPSTF appeared before that (Lin, Lipsitz, Miller, & Janakiraman, 2008). Interestingly, patient-physician communication was also

62

linked to lower likelihood of PSA screening in the partial-mediation model, in contrast to the colorectal cancer screening results. This may again suggest that physicians are uncertain about the benefits of PSA. Another possibility is that physicians are deciding not to recommend the PSA test for the general population and that participatory decision making between healthcare providers and patients is leading to patients not choosing this test. The implication of this result is that better communication and shared decision making may prevent over-screening by a test that has uncertain benefits, like the PSA blood test. The results in this case may imply that more education is needed for men about PSA screening, from their healthcare providers and on websites and other informational resources. This will especially be important when the new USPSTF recommendations of not getting screened on a universal level become public. Public health officials and physicians will have to focus on getting the information to the public and recommend reliable sources of information on the topic. More current study on this topic is needed, and information continues to be released from large trials that are currently underway. Communication, bringing information to a physician, and his or her reaction Although the direct tests of associations between healthcare provider reaction to Internet health information and cancer screening outcomes were not significant (one was marginally significant for the cervical cancer screening in the younger age group), the partial-mediation models fit the data well and also indicated some significant associations between perceived patient-physician communication, physician reaction, and cancer screening adherence. For women aged 18-35, a more positive physician reaction to Internet health information was negatively associated with the likelihood of cervical

63

cancer screening adherence. This result seems counterintuitive, but is difficult to interpret because of how the questions were asked (it is not known, for example, what the health information that was brought to the doctor was about, or which doctor the patient was rating in the patient-physician communication scale). In the other age strata, no relationship was found between physician reaction to health information and cancer screening outcomes, but a more positive report of patient-physician communication was positively associated with a more positive perceived physician reaction to Internet health information. The same pattern of results emerged in the models looking at colorectal cancer screening. In addition, patient-physician communication was positively linked to cancer screening for participants aged 50-60. The results for Specific Aim 3 seem to indicate that different aspects of the patient physician relationship are important. Both communication and perceived physician reaction to patient engagement in the form of Internet health information are positively associated with each other, and patient-physician communication is in turn linked to higher likelihood of colorectal cancer screening adherence. Post-hoc analysis also indicated that bringing Internet information to a physician is associated with higher cervical cancer screening likelihood for women aged 18-35. This type of patient physician interaction may be another marker of a patient who is engaged in her health care and may be a good variable to use in future analyses of the patient-physician relationship and its association with cancer screening and other forms of self-care. Limitations The present results should be interpreted in light of several important limitations. First, the data used for the analyses were taken from two cross-sectional administrations

64

of HINTS. Because the data are not longitudinal, no conclusions can be drawn about the temporal occurrence of the variables in the study, nor can causality be inferred. It may be that participants rated patient-physician communication and sought health and cancer information before they received cancer screening, but in some cases the opposite may have been true. Therefore, even though some significant associations between these variables were found, it is difficult to say with certainty that people who seek health information then put that information to use by getting on-schedule cancer screenings. It is more likely that people who are generally more involved in their own healthcare are also more likely to adhere to cancer screening guidelines, whether because of their relationship with a healthcare provider or because they are simply more informed and feel empowered to make healthcare decisions. Further research is needed to examine the role of “health care participation” in the present findings. Second, the variables tested in the current study referred to health and cancer information seeking, patient-physician communication, and cancer screening behavior. These questions were asked about past behaviors and may suffer from recall bias and socially desirable responding. The use of computer assisted telephone interviewing (CATI) and mail surveys may have alleviated the social desirability bias, but recall bias remains a limitation. Participants were asked whether and when they underwent various cancer screening procedures, and some of these questions asked about events going back as far back as 10 years (colonoscopy). There may have been some uncertainty, then, on whether patients were really adherent to cancer screening guidelines or whether they remembered their screening schedule incorrectly. However, there is no indication that

65

this type of misclassification bias was differentially influential in either the screened or non-screened group, so the results likely were not pulled in either direction. The patient-physician communication scale was different between the 2003 and 2007 administrations, so the results from either of the two years may not be fully comparable to each other. In addition, this scale asked about patients’ communication with all doctors, nurses, and other health providers whom they saw in the past 12 months. This makes the scale not very specific, and it is not known whether patients had different relationships with their primary care physicians or specialists, physician’s assistants, nurses or other healthcare providers they may have seen. Also, insured patients do tend to have some control over which physicians they choose under their health plan, and it follows that they are likely to switch away from healthcare providers who do not satisfy their needs. This last fact may be one of the reasons that patients rated their communication with healthcare providers quite favorably, although uninsured patients may not have the same freedom and would perhaps have a different experience with patient-physician communication. Another reason for the high ratings may be that patients see doctors and other healthcare providers as authority figures and are therefore unwilling to challenge them. Third, the information seeking variable may also need to be examined with caution. We do not know exactly what information the participants were looking for. It is also unknown what information they found on the Internet or through other sources. Some research has looked at the accuracy and relevance of content on web pages, but more research needs to be done. Not all websites provide accurate or current information about cancer or about health topics in general. Additionally, websites devoted to

66

controversial topics, like prostate cancer screenings, or to complicated screenings like colorectal cancer screening may generate more confusion than help people make good healthcare decisions. Finally, there are some limitations in the study design. Though a large nationally representative data set, HINTS suffers from fairly low response rates. Falling response rates for RDD surveys have been observed over the past decade across all types of survey administrations, particularly with increases in cell phone-only households and the introduction of caller ID (Blumberg, Luke, & Cynamon, 2006; Fahimi, Link, Mokdad, Schwartz, & Levy, 2008). However, results from methodological research studies suggest that the threat for bias introduction from low response rates may not be as potent for health surveys as originally feared, although some differences in overall estimates may persist (Blumberg et al., 2006; Fahimi et al., 2008; Gentry et al., 1985; D. E. Nelson, Powell-Griner, Town, & Kovar, 2003). Moreover, efforts have been underway in the HINTS program to address these sources of potential error, including the use of dual frame administrations to combat modality bias, coverage bias, and sampling error (Cantor et al., 2009). In addition, the 2007 survey administration did not include information on breast or prostate cancer screening, possibly because of controversies surrounding mammography and PSA testing. It is also difficult to determine if people who refuse are also those who are less likely to be empowered in their healthcare decision making and who are less engaged in their own health care because phone and mail surveys likely miss sampling people who have a lower socio-economic status. It is also known that there is a digital divide among people with lower SES, so that those who are most vulnerable to under-screening are also those least likely to get the benefit of searching for Internet

67

health information (Viswanath & Kreuter, 2007). This type of non-response would likely dilute the findings in the current study. Finally, because more and more people are using only their cell phones and no longer have home phones, random-digit dialing sampling methods will have to include cellular phone numbers if studies are to be truly representative of the US population. Implications for future research Some significant relationships were found among information seeking, patientphysician communication and the adherence to cancer screening guidelines in the current study. These associations need to be examined further in studies with prospective and experimental designs to facilitate understanding of the temporality of the sequence of these events. People’s information seeking preferences, especially about cancer specifically, need to be closely studied and followed. With the ease of use of the Internet and the portability of mobile Internet technology, it is becoming easier to find information on health topics. However, the accuracy and relevance of that information and its impact on medical decision making needs to be assessed by skilled researchers. In addition, health literacy plays a large role in how that information is used and interpreted by people who find it, and this construct needs to be closely examined as well. Physician recommendations regarding cancer screenings have previously been found to be important for cancer screening adherence (M. S. O'Malley et al., 2001; Taylor et al., 2003). Patient-physician communication has also been shown to be important in the current study and others (Baile & Aaron, 2005; Beck et al., 2002; Ong et al., 1995). The interaction of these two constructs deserves further study as do other aspects of the patient provider relationship such as type of doctor or healthcare provider,

68

the trust that patients feel, as well as an examination of the relationship from the physician’s side. Direct observations of visits with healthcare providers may hold some insight into what is discussed and longitudinal tracking of health behaviors such as cancer screenings may show how the content of communication affects behaviors. Different organizations list different cancer screening guidelines. For example, although the USPSTF has raised its recommended starting age for mammography to 50, the American Cancer Society has not and still lists 40 as the age to start mammograms for women of average risk (R. A. Smith et al., 2010). Another example of inconsistent recommendations are the various clinician groups that recommend colonoscopy as the best modality for colorectal cancer screening, although the evidence is unclear whether the results of this type of screening are better than the home-based and much less invasive FOBT. Finally, some cancer screenings are still undergoing study, such as screening procedures for lung, prostate and ovarian cancers. Studying how patients perceive these sometimes confusing recommendations and how they make decisions on which to follow, as well as looking at how physicians interpret these varying recommendations and what actions they take when talking to their patients, will provide much needed insight into cancer screening behavior. Another avenue of future research may deal with colorectal cancer screening specifically, since there are multiple possible modalities for getting screened with different rates of adherence. In this study, adherence was defined to include any of the modalities, as long as the tests were performed on the recommended schedule. However, differences likely exist for patients who choose to perform the FOBT versus those who choose colorectal endoscopy. Different barriers may be present for these procedures and

69

the acceptability of either or both of these modalities needs to be examined by researchers if rates of colorectal cancer screening are to be raised. Finally, further study should be done with information seeking and how it relates to cancer screening while utilizing health communication theories, like the Health Belief Model. The Health Belief Model contains constructs like barriers to or benefits of screening (Becker, 1974) and the Theory of Planned Behavior looks at people’s attitudes, expectations, subjective norms and perceived behavioral control and how these individual factors affect the likelihood of performing a health behavior (Ajzen, 1991). Studying these constructs may lead to a deeper understanding of the processes involved in cancer screening adherence among the general population. Implications for practitioners In the present study, information seeking was found to serve as an important correlate of cancer screening adherence. In particular, Internet information seeking seems to be associated with cervical, colorectal and prostate cancer screening. Even though it is widely assumed that young people are using the Internet more than older people, the association between Internet-specific information seeking and different types of cancer screenings are apparent in various age groups, including in people over 60 (for colorectal cancer). This may mean that patients require more information than they are receiving from their healthcare providers. It may also mean that patients want to be involved in their own healthcare and that they are willing to research health behaviors that will prolong their lives and improve their quality of life. It is important for healthcare providers to understand that patients are interested in health information generally and cancer information specifically. One possible way to insure that patients are getting

70

accurate and relevant information in their searching may be to meet their needs in the office by spending more time providing all the information. Another way would be to recommend websites that contain accurate, up-to-date information that has been vetted by the healthcare provider so that patients can explore and learn on their own time. Patient-physician communication has also been found to be an important factor in cancer screening adherence in the present study. Patients who rate their healthcare providers highly on communication are more likely to get screened on schedule for colorectal cancer. Given that rates of colorectal cancer of screening are lower than those of other cancers, it is important to help patients to improve their adherence to screening guidelines. In this study, communication is defined by allowing patients to ask questions, involving patients in healthcare decisions, giving attention to patient needs and feelings, and making sure patients understand the things they need to do to care for their health. Together, these variables index a participatory way of dealing with patients and their healthcare needs. Therefore, more participatory visits may lead to higher levels of cancer screening, and possibly to better health outcomes. Thus, more interventions may be needed to teach both patients and physicians about the importance of participatory decision making during healthcare visits. Another important theme from the current results is the importance of health insurance. Almost every analysis from this study found that having health insurance was strongly associated with getting the cancer screenings that are recommended. Because cancer screenings are preventive in nature and thus should not only lower rates and severity of disease, but may also save money, the importance of making sure that the population has access to health insurance is paramount. The Affordable Care Act should

71

make health insurance easier to obtain for those who do not have it currently. The question of underinsurance will also have to be examined and addressed in future research and in any interventions that deal with rates of cancer screenings. Conclusion Despite some limitations, this study was able to examine the relationship between health information seeking from the Internet and other sources and cancer screening adherence (for breast, cervical, and colorectal cancers) and behavior (for prostate cancer), as well as the relationship that patient-physician communication had with these two constructs. In general, it was determined that people who look for health or cancer information are more likely to get screened on schedule and that people who report a more positive communication with their healthcare providers are also more likely to get screened. Some groups of people, however, don’t exhibit this relationship and thus may be more vulnerable to under-screening (or over-screening in the case of prostate cancer). These specific groups may benefit more from targeted interventions that attempt to involve people in their healthcare.

Tables and Figures Table 1.1. Summary of recommendations for cancer screenings from the USPSTF, both current and past. Type of cancer

2003

Current

Breast

Mammography every 1-2 years,

Biennial mammography for women

with or without a clinical breast

aged 50-74. No self breast exams. Not

exam for women aged 40 and

enough evidence to assess clinical

above. Not enough evidence for or

breast exam benefits.

against self exams. Cervical

Pap smears every 3 years, starting

Pap smears every 3 years for ages 21-

at age 21 or at most 3 years after

65, or combination of Pap smear and

sexual initiation and ending at age

HPV test every 5 years for ages 30-

65.

65. No screening for women younger than 21 or older than 65.

Colorectal

Prostate

FOBT every 1 year, flexible

FOBT every 1 year, flexible

sigmoidoscopy every 5 years, or

sigmoidoscopy every 5 years, or

colonoscopy every 10 years from

colonoscopy every 10 years from age

age 50 to 75. No routine screening

50 to 75. No routine screening above

above age 75. Benefits and harms

age 75. Benefits and harms vary by

vary by method.

method.

Insufficient evidence for men

No routine PSA-based screening for

younger than 75. No screening after

all men.

age 75.

72

2007

Cancer information seeking

2003

Health information seeking

Cancer information seeking

Health information seeking

Type of questions

Survey year

about health and medical topics, where did you go first?

from any source? The most recent time you looked for information

Have you ever looked for information about health or medical topics

you go first?

The most recent time you looked for cancer information, where did

Have you ever looked for information about cancer from any source?

yourself?

somewhere else to look for health or medical information for

In the past 12 months, did you use the Internet, whether from home or

you look first?

The most recent time you looked for information on cancer, where did

Have you ever looked for information about cancer from any source?

Content of questions

73

Table 2.1. Cancer and health information seeking questions from HINTS 2003 and 2007.

a

How often did doctors, nurses or other health professionals give the attention you needed to your feelings and emotions? How often did they involve you in decisions about your health care as much as you wanted? How often did they make sure you understood the things you needed to do to take care of your health? How often did they help you deal with feelings of uncertainty about your health or health care? In the past 12 months, how often did you feel you could rely on doctors, nurses or other health professionals to take care of your health needs?

3

4

5

6

… involve you in decisions about your health care as much as you wanted?

5

2

… spend enough time with you?

4

During the past 12 months, how often did doctors, nurses, or other health professionals give you the chance to ask all the health-related questions you had?

… show respect for what you had to say?

3

1

… explain things in a way you could understand?

2

2007

During last 12 months, how often did doctors or other health care providers a listen carefully to you?

1

2003

Content of questions

Question

Year

74

Table 2.2. Questions on the patient-physician communication scale on HINTS 2003 and 2007.

All questions on this scale started with “During the past 12 months, how often did doctor and other health care providers” and the specific question was then filled in.

75

Table 4.1 - 2003 weighted descriptive statistics (n = 6,369) Variable (range)

Mean (SD)

Age (18 – 95)

46.21 (17.12)

Percent

Sex – female

51.9

Family history of cancer

61.3

Have health insurance

84.9

Self-reported health (1=excellent – 5=poor)

2.67 (1.05)

Hispanic

11.6

Non-Hispanic Black

10.4

Non-Hispanic American Indian

1.4

Non-Hispanic Asian

2.2

Non-Hispanic Hawaiian or Pacific Islander

0.3

Non-Hispanic Multiple races indicated

2.1

Non-Hispanic White

70.8

During last 12 months, how often did doctors or other

3.47 (0.75)

health care providers listen carefully to you? … explain things in a way you could understand?

3.50 (0.73)

… show respect for what you had to say?

3.61 (0.68)

… spend enough time with you?

3.32 (0.85)

… involve you in decisions about your health care as much

3.41 (0.84)

as you wanted? Household income – below $25,000

26.5

Household income – between $25,000 and $35,000

12.2

Household income – between $35,000 and $50,000

15.8

Household income – between $50,000 and $75,000

15.9

Household income – above $75,000

20.7

Cancer information seeking - none

55.0

Cancer information seeking - Internet

21.6

Cancer information seeking – other sources

22.9

Health information seeking for oneself

50.6

76

Table 4.2 - 2007 weighted descriptive statistics (n = 7,623) Variable (range)

Mean (SD)

Age (18 – 97)

45.51 (18.08)

Percent

Sex – female

51.4

Family history of cancer

69.7

Have health insurance

82.4

Self-reported health (1=excellent – 5=poor)

2.61 (0.94)

Patients who talked to their healthcare providers about

25.7

health information Perceived physicians’ reactions to Internet health

3.07 (0.92)

information Hispanic

12.7

Non-Hispanic Black

11.4

Non-Hispanic American Indian

0.5

Non-Hispanic Asian

4.4

Non-Hispanic Hawaiian or Pacific Islander

0.2

Non-Hispanic Multiple races indicated

1.3

Non-Hispanic White

68.8

During the past 12 months, how often did doctors, nurses,

3.38 (0.80)

or other health professionals give you the chance to ask all the health-related questions you had? How often did doctors, nurses or other health professionals

3.12 (0.92)

give the attention you needed to your feelings and emotions? How often did they involve you in decisions about your

3.24 (0.89)

health care as much as you wanted? How often did they make sure you understood the things

3.42 (0.77)

you needed to do to take care of your health? How often did they help you deal with feelings of

3.08 (0.97)

uncertainty about your health or health care? In the past 12 months, how often did you feel you could rely on doctors, nurses or other health professionals to take care of your health needs?

3.31 (0.81)

77 Household income – below $20,000

18.1

Household income – between $20,000 and $35,000

15.3

Household income – between $35,000 and $50,000

12.8

Household income – between $50,000 and $75,000

17.5

Household income – above $75,000

27.8

Cancer information seeking - none

60.9

Cancer information seeking - Internet

21.4

Cancer information seeking – other sources

17.3

Health information seeking - none

30.1

Health information seeking - Internet

42.6

Health information seeking – other sources

24.8

* p < 0.05

a

** p < 0.01

*** p < 0.001

health, health insurance status, and income 776

66 and

up

859

860

51-65

40-50

in

group analysis

n used

Age

0.774

0.840

0.740

rate

Screening

0.072

0.373

0.344

rate

seeking

info

Internet

0.072

0.198

0.228

rate

seeking

info

Other

(0.387)

(0.453)

0.405

(0.205)

(0.155) 0.135

0.077

(0.178)

(0.137) 0.428**

0.414*

beta (SE)

seeking

other info

Unadjusted

0.285*

beta (SE)

seeking

info

Internet

Unadjusted

(0.359)

-0.115

(0.185)

0.199

(0.165)

-0.019

beta (SE)

seeking

info

Internet

Adjusted a

(0.443)

0.258

(0.223)

-0.139

(0.191)

0.170

beta (SE)

seeking

other info

Adjusted a

78

Table 4.3 - Breast cancer screening and cancer information seeking

^ p < 0.10

Controlling for patient-physician communication, family history of cancer, self-reported

79

* p < 0.05 a

(0.391)

-1.347^ (0.714) 0.206 0.243 up

168 66 and

0.876

(0.201)

0.867

0.725

-0.001

0.001 (0.201)

*** p < 0.001

493

624 40-50 ** p < 0.01

51-65

(0.160)

0.171 (0.184) -0.226 0.784

0.590

(SE)

beta (SE) seeking beta rate

info seeking health info seeking

Internet health Internet info analysis group

rate

n used in Age

Screening

Internet

Unadjusted

Adjusted b

Table 4.4 - Breast cancer screening and health information seeking

^ p < 0.10

In this group, n’s are different because of missing values on the Internet health info

seeking variable (question was only asked of those who reported using the Internet) b

Controlling for patient-physician communication, family history of cancer, self-reported

health, health insurance status, and income

* p < 0.05

a

** p < 0.01

*** p < 0.001

health, health insurance status, and income

(2007)

51-65

(2007) 1462

0.780

0.877

0.408

0.464

0.334

0.299

0.169

0.419*

(0.098)

0.286**

(0.155)

0.141

1141

0.215

(0.145)

36-50

0.789

0.198

(0.181)

783

0.373

(2007)

18-35

(2003)

0.799

0.571***

859

51-65

0.298 (0.115)

0.402

0.056

(2003)

0.808

0.200

seeking

info

Internet

Unadjusted

0.419***

1169

0.644

rate

seeking

info

Other

36-50

0.808

rate

seeking

info

Internet

(0.123)

1024

rate

Screening

(2003)

18-35

in

group analysis

n used

Age

(0.128)

-0.123

(0.152)

-0.033

(0.273)

0.171

(0.201)

0.517**

(0.156)

0.095

(0.172)

0.373*

beta (SE)

seeking

other info

Unadjusted

(0.117)

0.142

(0.197)

0.059

(0.203)

0.350^

(0.164)

0.414*

(0.131)

0.295*

(0.146)

-0.063

seeking

info

Internet

Adjusted a

(0.143)

-0.119

(0.158)

0.054

(0.331)

0.303

(0.2)

0.345^

(0.164)

0.003

(0.187)

0.254

beta (SE)

seeking

other info

Adjusted a

80

Table 4.5 – Cervical cancer screening and cancer information seeking

^ p < 0.10

Controlling for patient-physician communication, family history of cancer, self-reported

* p < 0.05

a

** p < 0.01

*** p < 0.001 (2007)

51-65

(2007)

36-50

(2007)

18-35

(2003)

51-65

(2003)

36-50

(2003)

1462

1141

783

493

870

792

0.780

0.877

0.789

0.860

0.838

0.818

rate

analysis a

group

18-35

Screening

n used in

Age

0.480

0.656

0.429

0.725

0.681

0.688

0.329

0.354

0.414

NA

NA

-0.066

NA c

(0.115)

0.248*

(0.181)

0.056 (0.137)

(0.134)

(0.254)

(0.193) 0.118

0.396

0.367^

(0.21)

(0.266) (0.162) 0.525**

0.449*

(0.202)

0.077

(0.122)

0.319**

(0.156)

-0.011

seeking

info

Internet

Adjusted b

0.497^

NA

NA

NA

(SE)

seeking beta

other info

Unadjusted

0.541**

(0.179)

0.075

(0.117)

-0.260*

(0.140)

(SE)

seeking beta

Internet info

Unadjusted

rate

seeking

seeking rate

info

Other

info

Internet

(0.161)

0.040

(0.234)

0.435^

(0.31)

0.542^

NA

NA

NA

beta (SE)

seeking

other info

Adjusted b

81

Table 4.6 – Cervical cancer screening and health information seeking

^ p < 0.10

In this group, n’s are different because of missing values on the Internet health info seeking variable (question was only asked of those who reported using the Internet) b Controlling for patient-physician communication, family history of cancer, self-reported health, health insurance status, and income c In 2003, the question about health information seeking is specifically about Internet health information

* p < 0.05

a

in analysis

group (survey

** p < 0.01

*** p < 0.001

reported health, health insurance status, and income (2007)

0.731

0.199

0.279

0.191

0.367***

(0.108)

0.111

1879

0.181

(0.186)

61-75

0.559

0.092

(0.088)

1858

0.071

(2007)

50-60

(2003)

0.532

-0.166

966

61-75

-0.162

seeking

info

Internet

Unadjusted

(0.130)

0.091

rate

seeking

info

Other

(2003)

0.247

50-60

seeking

info

Internet

rate 0.579

rate

Screening

year) 1030

n used

Age

(0.104)

0.237*

(0.088)

0.256**

(0.183)

0.226

(0.171)

0.068

beta (SE)

seeking

other info

Unadjusted

(0.106)

-0.012

(0.1)

0.231*

(0.207)

-0.214

(0.140)

-0.055

seeking

info

(0.114)

0.202^

(0.101)

0.198*

(0.212)

0.178

(0.180)

0.153

seeking

info

other

a

a

Internet

Adjusted

Adjusted

82

Table 4.7 – Colorectal cancer screening and cancer information seeking

^ p < 0.10

Controlling for sex, patient-physician communication, family history of cancer, self-

83

* p < 0.05 a

(0.110) (0.113) (0.098)

(0.097)

0.310** 0.233* 0.344***

(2007)

61-75

(2007)

1879

0.731

0.183

0.346

0.313**

(0.111) (0.126) (0.111)

(0.099)

0.191^ 0.096 0.559 1858 50-60

(2003)

374 61-75

(2003)

*** p < 0.001

0.217^ 0.605

(0.159)

NA 0.560

0.305

NA 0.517 0.558 676 50-60

year)

** p < 0.01

0.389***

0.290**

(0.164)

NA -0.161 0.074

(0.119)

NA

(0.121)

NA b t (SE) -0.066 c

0.105

NA

beta (SE) seeking (SE) (SE) rate rate

seeking info seeking beta seeking beta seeking seeking (survey

other info Internet other info Internet info info rate analysis a group

info

Screening n used in Age

Internet

Other

Unadjusted

Unadjusted

Adjusted b

Adjusted a

Table 4.8 – Colorectal cancer screening and health information seeking

^ p < 0.10

In this group for 2003, n’s are different because of missing values on the Internet health info seeking variable (question was only asked of those who reported using the Internet) b Controlling for sex, patient-physician communication, family history of cancer, selfreported health, health insurance status, and income c In 2003, the question about health information seeking is specifically about Internet health information

84

*** p < 0.001

(0.429)

NA -0.427

(0.374)

NA 0.270 (health)

61+

(health)

143

0.884

0.832 0.696 278 45-60

(cancer)

-0.152

(0.163)

NA

(0.187)

NA 0.382* NA NA c

0.361*

(0.325) (0.444) (0.426)

(0.282)

0.345 -0.134 0.004 0.406 0.076 0.865 318 61+

(cancer)

363 45-60

** p < 0.01

0.371

(0.199) (0.207) (0.207)

(0.207)

0.182 b t (SE) 0.254 0.669

0.344

0.396

0.293

0.246

beta (SE) seeking (SE) (SE) rate info)

analysis (type of

* p < 0.05 a

rate

seeking info seeking beta seeking beta seeking seeking

other info Internet other info Internet info info info in group

rate

n a used Age

Screening

Internet

Other

Unadjusted

Unadjusted

Adjusted b

Adjusted b

Table 4.9 – Prostate cancer screening and cancer/health information seeking

^ p < 0.10

In this group, n’s are different because of missing values on the Internet health info seeking variable (question was only asked of those who reported using the Internet) b Controlling for patient-physician communication, family history of cancer, self-reported health, health insurance status, and income c In 2003, the question about health information seeking is specifically about Internet health information

* p < 0.05

a

** p < 0.01

*** p < 0.001

health, health insurance status, and income

seeking

info

Health

seeking

info

Cancer

150

250

analysis

info

seeking

n used in

Type of

0.619

0.617

g rate

Screenin

0.531

0.340

rate

seeking

info

Internet

NA

0.313

rate

seeking

info

Other

(0.202)

NA

(0.262)

(0.244)

-0.448*

-0.095

beta (SE)

seeking -0.086

seeking

info

ed other

Unadjust

info

Internet

ed

Unadjust

(0.273)

-0.429

(0.281)

-0.037

beta (SE)

seeking

info

NA

(0.264)

-0.106

beta (SE)

seeking

info

other

a

a

Internet

Adjusted

Adjusted

85

Table 4.10 – Prostate cancer screening and cancer/health information seeking in

participants who did not get a recommendation from their physicians

^ p < 0.10

Controlling for patient-physician communication, family history of cancer, self-reported

86

Table 4.11 – Breast cancer screening and health information seeking – partial mediation model Age range

Independent

Mediator

Outcome

variable

Path B

Beta

coefficient

(standardized)

(SE) 40-50

Internet info

PP comm

0.081

seeking

(0.056) PP comm

* p < 0.05

Screening

0.423*

adherence

(0.173)

Internet info

Screening

-0.205

seeking

adherence

(0.183)

** p < 0.01

0.077

*** p < 0.001

^ p < 0.10

0.210

-0.097

87

Table 4.12 – Cervical cancer screening and information seeking – partial mediation model Age range

Independent

(survey

variable

Mediator

Outcome

year)

Path B

Beta

coefficient

(standardized)

(SE)

36-50

Internet info

(2003)

seeking

Cancer

Non-Internet

info

info seeking

PP comm

-0.011 (0.056)

PP comm

0.042

Screening

0.168

adherence

(0.114)

Internet info

Screening

0.296*

seeking

adherence

(0.131)

Non-Internet

Screening

-0.004

info seeking

adherence

(0.164)

Internet info

(2003)

seeking

Cancer

Non-Internet

info

info seeking

0.028

(0.057) PP comm

51-65

-0.009

PP comm

-0.087

0.089

0.127

-0.002

-0.063

(0.060) PP comm

-0.014

-0.009

(0.082) PP comm

Screening

-0.097

adherence

(0.151)

Internet info

Screening

0.405*

seeking

adherence

(0.163)

-0.051

0.154

88

Non-Internet

Screening

0.344^

info seeking

adherence

(0.200)

36-50

Internet info

(2003)

seeking

Health

PP comm

PP comm

Screening

0.203^

adherence

(0.107)

Internet info

Screening

0.334**

seeking

adherence

(0.123)

18-35

Internet info

(2007)

seeking

Health

Non-Internet

info

info seeking

PP comm

0.079

0.106

0.158

0.055

(0.059) PP comm

0.089

0.041

(0.145) PP comm

Screening

0.095

adherence

(0.176)

Internet info

Screening

0.348^

seeking

adherence

(0.203)

Non-Internet

Screening

0.294

info seeking

adherence

(0.328)

** p < 0.01

-0.069

(0.133)

info

* p < 0.05

-0.076

0.110

*** p < 0.001

^ p < 0.10

0.056

0.142

0.081

89

Table 4.13 – Colorectal cancer screening and information seeking – partial mediation model Age range

Independent

(type of

variable

Mediator

Outcome

info)

Path B

Beta

coefficient

(standardized)

(SE)

50-60

Internet info

(cancer)

seeking Non-Internet

PP comm

-0.112** (0.038)

PP comm

-0.117*

info seeking Screening

0.218**

adherence

(0.082)

Internet info

Screening

0.256*

seeking

adherence

(0.101)

Non-Internet

Screening

0.224*

info seeking

adherence

(0.098)

Internet info

(cancer)

seeking Non-Internet

-0.085

(0.049) PP comm

61-75

-0.089

PP comm

-0.030

0.115

0.106

0.086

-0.020

(0.059) PP comm

-0.014

info seeking

-0.011

(0.049) PP comm

Screening

0.272**

adherence

(0.083)

Internet info

Screening

-0.004

seeking

adherence

(0.107)

0.146

-0.001

90

Non-Internet

Screening

0.205^

info seeking

adherence

(0.113)

50-60

Internet info

(health)

seeking Non-Internet

PP comm

-0.068

PP comm

-0.123^

Screening

0.212*

adherence

(0.083)

Internet info

Screening

0.221^

seeking

adherence

(0.126)

Non-Internet

Screening

0.199^

info seeking

adherence

(0.111)

(health)

seeking Non-Internet

PP comm

-0.017

PP comm

0.104

0.085

-0.053

-0.002

0.008

(0.015) PP comm

Screening

1.061**

adherence

(0.310)

Internet info

Screening

0.251*

seeking

adherence

(0.113)

Non-Internet

Screening

0.313**

info seeking

adherence

(0.110)

** p < 0.01

0.112

(0.016)

info seeking

* p < 0.05

-0.061

(0.066) PP comm

Internet info

-0.099

(0.064)

info seeking

61-75

0.088

*** p < 0.001

^ p < 0.10

0.154

0.112

0.146

91

Table 4.14 – Prostate cancer screening and health information seeking – partial mediation model Age range

Independent

Mediator

Outcome

Path B

Beta

coefficient (SE)

(standardized)

-0.055 (0.090)

-0.047

Screening

0.058 (0.226)

0.032

Screening

0.385* (0.186)

0.182

0.019 (0.079)

0.022

variable 45-60

Internet info

PP comm

seeking PP comm Internet info seeking 61+

Internet info

PP comm

seeking PP comm Internet info

Screening

-1.086** (0.333) -0.387

Screening

-0.407 (0.441)

seeking * p < 0.05

** p < 0.01

*** p < 0.001

^ p < 0.10

-0.169

92

Table 4.15 - Cervical cancer screening and perceived physician reaction to health information Age group

n used in the

Screening rate

analysis a

Unadjusted B

Adjusted b B

(SE)

(SE)

18-35

151

0.913

-0.631 (0.47)

-1.284^ (0.719)

36-50

256

0.935

0.007 (0.243)

0.029 (0.162)

51-65

283

0.869

-0.059 (0.167)

-0.054 (0.162)

* p < 0.05 a

** p < 0.01

*** p < 0.001

^ p < 0.10

This sub-sample consisted of participants who reported bringing Internet health

information to their physicians b

Controlled for patient-physician communication, income, health insurance, self-reported

health, and history of cancer in the family

93

Table 4.16 – Colorectal cancer screening and perceived physician reaction to health information Age group

n used in the

Screening rate

analysis a

Unadjusted B

Adjusted b B

(SE)

(SE)

50-60

356

0.650

0.110 (0.116)

0.160 (0.134)

61-75

248

0.825

0.263^ (0.140)

0.283 (0.179)

* p < 0.05 a

** p < 0.01

*** p < 0.001

^ p < 0.10

This sub-sample consisted of participants who reported bringing Internet health

information to their physicians b

Controlled for patient-physician communication, income, health insurance, self-reported

health, and history of cancer in the family

94

Table 4.17 – Partial mediation test of the relationship between patient-physician communication and cervical cancer screening Age range

Independent

Mediator

Outcome

variable

Path B

Beta

coefficient

(standardized)

(SE) 18-35

Patient-

Physician

0.154

physician

reaction to

(0.168)

comm

IHI Reaction to

Cancer

-0.755***

IHI

screening

(0.169)

Patient-

Cancer

-0.061

physician

screening

(0.715)

0.124

-0.561

-0.036

comm 36-50

Patient-

Reaction to

0.618**

physician

IHI

(0.224)

0.368

comm Reaction to

Cancer

-0.001

IHI

screening

(0.182)

Patient-

Cancer

0.133

physician

screening

(0.203)

-0.001

0.060

comm 51-65

Patient-

Reaction to

0.393**

physician

IHI

(0.133)

0.281

95

comm Reaction to

Cancer

-0.052

IHI

screening

(0.176)

Patient-

Cancer

-0.005

physician

screening

(0.227)

comm * p < 0.05

** p < 0.01

*** p < 0.001

^ p < 0.10

-0.042

-0.003

96

Table 4.18 – Partial mediation test of the relationship between patient-physician communication and colorectal cancer screening Age range

Independent

Mediator

Outcome

variable

Path B

Beta

coefficient

(standardized)

(SE) 50-60

Patient-

Reaction to

0.517***

physician

IHI

(0.134)

0.333

comm Reaction to

Cancer

0.039

IHI

screening

(0.130)

Patient-

Cancer

0.368*

physician

screening

(0.179)

0.031

0.186

comm 61-75

Patient-

Reaction to

2.833***

physician

IHI

(0.724)

0.560

comm Reaction to

Cancer

0.055

IHI

screening

(0.159)

Patient-

Cancer

0.885

physician

screening

(0.745)

comm * p < 0.05

** p < 0.01

*** p < 0.001

^ p < 0.10

0.048

0.153

97

Figure 1.1. Model tested in Specific Aim 2.

Internet cancer information seeking

Patientprovider communication Cancer screening adherence: • Breast • Cervical • Colorectal Ever cancer screening: • Prostate

Internet health information seeking

Figure 1.2. Model tested in Specific Aim 3.

Physician reaction to Internet health information Patientprovider communication

Cancer screening adherence: • Cervical • Colorectal

98

Figure 3.1. Alternative model for Specific Aim 2.

Internet cancer and health information seeking Patientprovider communication

Cancer screening adherence: • Breast • Cervical • Colorectal Ever cancer screening: • Prostate

Figure 3.2. Alternative model for Specific Aim 3.

Physician reaction to Internet health information

Patientprovider communication

Cancer screening adherence: • Cervical • Colorectal

References Ahmad, F., Hudak, P. L., Bercovitz, K., Hollenberg, E., & Levinson, W. (2006). Are physicians ready for patients with Internet-based health information? J Med Internet Res, 8(3), e22. doi: 10.2196/jmir.8.3.e22 Ajzen, I. (1991). The Theory of Planned Behavior. Organizational Behavior and Human Decision Processes, 50(2), 179-211. American Cancer Society. (2011a). Breast Cancer: Early Detection. American Cancer Society. (2011b). Cancer Facts & Figures 2011. Atlanta: American Cancer Society. Andriole, G. L. (2005). Prostate specific antigen based prostate cancer screening: accumulating evidence of efficacy but persistent uncertainty. J Urol, 174(2), 413414. doi: 10.1097/01.ju.0000171051.35187.d0 Andriole, G. L., Crawford, E. D., Grubb, R. L., 3rd, Buys, S. S., Chia, D., Church, T. R., . . . Berg, C. D. (2009). Mortality results from a randomized prostate-cancer screening trial. N Engl J Med, 360(13), 1310-1319. doi: 10.1056/NEJMoa0810696 Andriole, G. L., Levin, D. L., Crawford, E. D., Gelmann, E. P., Pinsky, P. F., Chia, D., . . . Gohagan, J. K. (2005). Prostate Cancer Screening in the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial: findings from the initial screening round of a randomized trial. J Natl Cancer Inst, 97(6), 433-438. doi: 10.1093/jnci/dji065 Baile, W. F., & Aaron, J. (2005). Patient-physician communication in oncology: past, present, and future. Curr Opin Oncol, 17(4), 331-335. Baker, L., Wagner, T. H., Singer, S., & Bundorf, M. K. (2003). Use of the Internet and email for health care information: results from a national survey. JAMA, 289(18), 2400-2406. doi: 10.1001/jama.289.18.2400 Beck, R. S., Daughtridge, R., & Sloane, P. D. (2002). Physician-patient communication in the primary care office: a systematic review. J Am Board Fam Pract, 15(1), 2538. Becker, M. H. (1974). The health belief model and personal health behavior. Health Education Monographs, 2, 324-508.

99

100

Berkowitz, Z., Hawkins, N. A., Peipins, L. A., White, M. C., & Nadel, M. R. (2008). Beliefs, risk perceptions, and gaps in knowledge as barriers to colorectal cancer screening in older adults. J Am Geriatr Soc, 56(2), 307-314. doi: 10.1111/j.15325415.2007.01547.x Berland, G. K., Elliott, M. N., Morales, L. S., Algazy, J. I., Kravitz, R. L., Broder, M. S., . . . McGlynn, E. A. (2001). Health information on the Internet - Accessibility, quality, and readability in English and Spanish. Jama-Journal of the American Medical Association, 285(20), 2612-2621. Bernstam, E. V., Walji, M. F., Sagaram, S., Sagaram, D., Johnson, C. W., & MericBernstam, F. (2008). Commonly cited website quality criteria are not effective at identifying inaccurate online information about breast cancer. Cancer, 112(6), 1206-1213. doi: 10.1002/cncr.23308 Black, A. T., McCulloch, A., Martin, R. E., & Kan, L. (2011). Young women and cervical cancer screening: what barriers persist? Can J Nurs Res, 43(1), 8-21. Blumberg, S. J., Luke, J. V., & Cynamon, M. L. (2006). Telephone coverage and health survey estimates: evaluating the need for concern about wireless substitution. Am J Public Health, 96(5), 926-931. doi: 10.2105/AJPH.2004.057885 Breast-cancer awareness: too much of a good thing? (2007). Lancet Oncol, 8(12), 1041. doi: 10.1016/S1470-2045(07)70347-9 Breen, N., K, A. C., Meissner, H. I., Taplin, S. H., Tangka, F. K., Tiro, J. A., & McNeel, T. S. (2007). Reported drop in mammography : is this cause for concern? Cancer, 109(12), 2405-2409. doi: 10.1002/cncr.22723 Briss, P., Rimer, B., Reilley, B., Coates, R. C., Lee, N. C., Mullen, P., . . . Lawrence, R. (2004). Promoting informed decisions about cancer screening in communities and healthcare systems. Am J Prev Med, 26(1), 67-80. Broom, A. (2005). Virtually he@lthy: the impact of internet use on disease experience and the doctor-patient relationship. Qualitative Health Research, 15(3), 325-345. doi: 10.1177/1049732304272916 Bylund, C. L., Gueguen, J. A., D'Agostino, T. A., Li, Y. L., & Sonet, E. (2010). DoctorPatient Communication About Cancer-Related Internet Information. Journal of Psychosocial Oncology, 28(2), 127-142. doi: 10.1080/07347330903570495 Bylund, C. L., Gueguen, J. A., Sabee, C. M., Imes, R. S., Li, Y. L., & Sanford, A. A. (2007). Provider-patient dialogue about internet health information: An exploration of strategies to improve the provider-patient relationship. Patient Educ Couns, 66(3), 346-352. doi: 10.1016/j.pec.2007.01.009

101

Byrne, B. M. (2012). Structural equation modeling with Mplus. New York, NY: Taylor & Francis Group, LLC. Cantor, D., Coa, K., Crystal-Mansour, S., Davis, T., Dipko, S., & Sigman, R. (2009). HINTS 2007 Final Report. Rockville, MD: National Cancer Institute. Cantor, D., Covell, J., Davis, T., Park, I., & Rizzo, L. (2005). HINTS 2005 Final Report. Rockville, MD: National Cancer Institute. Cantor, D., & McBride, D. (2007). Analyzing HINTS 2007: Considering differences between the mail and RDD frames. Cline, R. J. W., & Haynes, K. M. (2001). Consumer health information seeking on the Internet: the state of the art. Health Educ Res, 16(6), 671-692. Committee on Quality Health Care in America. (2001). Crossing the quality chasm: A new health system for the 21st century: Institute of Medicine. Cooper-Patrick, L., Gallo, J. J., Gonzales, J. J., Vu, H. T., Powe, N. R., Nelson, C., & Ford, D. E. (1999). Race, gender, and partnership in the patient-physician relationship. JAMA, 282(6), 583-589. Cooper, L. A., Roter, D. L., Johnson, R. L., Ford, D. E., Steinwachs, D. M., & Powe, N. R. (2003). Patient-centered communication, ratings of care, and concordance of patient and physician race. Ann Intern Med, 139(11), 907-915. Crocco, A. G., Villasis-Keever, M., & Jadad, A. R. (2002). Analysis of cases of harm associated with use of health information on the internet. JAMA, 287(21), 28692871. Diaz, J. A., Griffith, R. A., Ng, J. J., Reinert, S. E., Friedmann, P. D., & Moulton, A. W. (2002). Patients' use of the Internet for medical information. J Gen Intern Med, 17(3), 180-185. Dobson, A. J. (2002). Binary variables and logistic regression An Introduction to Generalized Linear Models (2nd edition ed., pp. 115-131): Chapman & Hall/CRC. Eddy, D. M. (1986). Secondary prevention of cancer: an overview. Bull World Health Organ, 64(3), 421-429. Epstein, R. M., Alper, B. S., & Quill, T. E. (2004). Communicating evidence for participatory decision making. [Review]. JAMA, 291(19), 2359-2366. doi: 10.1001/jama.291.19.2359

102

Fahimi, M., Link, M., Mokdad, A., Schwartz, D. A., & Levy, P. (2008). Tracking chronic disease and risk behavior prevalence as survey participation declines: statistics from the behavioral risk factor surveillance system and other national surveys. [Comparative Study]. Prev Chronic Dis, 5(3), A80. Ferguson, T. (2007). E-patients: How they can help us heal health care., from http://epatients.net/ Finney Rutten, L., Nelson, D. E., & Meissner, H. (2004). Examination of populationwide ternds in barriers to cancer screening from a diffusion of innovation perspective (1987-2000). Preventive Medicine, 38, 258-268. Finney Rutten, L. J., Meissner, H. I., Breen, N., Vernon, S. W., & Rimer, B. K. (2005). Factors associated with men's use of prostate-specific antigen screening: evidence from Health Information National Trends Survey. Prev Med, 40(4), 461-468. doi: 10.1016/j.ypmed.2004.07.011 Food and Drug Administration. (2006). Gardasil (human pappilomavirus vaccine) questions and answers - Gardasil, June 8, 2006, from http://www.fda.gov/BiologicsBloodVaccines/Vaccines/QuestionsaboutVaccines/u cm096052.htm Fox, S. (2011a). Health Topics (pp. 1-33): Pew Research Center. Fox, S. (2011b). The social life of health information, 2011 (pp. 1-45): Pew Research Center. Fox, S., & Fallows, D. (2003). Internet health resources: Health searches and email have become more commonplace, but there is room for improvement in searches and overall Internet access: Pew Internet & American Life Project. Gentry, E. M., Kalsbeek, W. D., Hogelin, G. C., Jones, J. T., Gaines, K. L., Forman, M. R., . . . Trowbridge, F. L. (1985). The behavioral risk factor surveys: II. Design, methods, and estimates from combined state data. Am J Prev Med, 1(6), 9-14. Gierisch, J. M., & Bastian, L. A. (2010). Cancer Screening Adherence Improving Patient Treatment Adherence. In H. Bosworth (Ed.), (pp. 123-150): Springer New York. Glynn, R. W., Kelly, J. C., Coffey, N., Sweeney, K. J., & Kerin, M. J. (2011). The effect of breast cancer awareness month on internet search activity--a comparison with awareness campaigns for lung and prostate cancer. BMC Cancer, 11, 442. doi: 10.1186/1471-2407-11-442

103

Hawkins, N. A., Berkowitz, Z., & Peipins, L. A. (2010). What does the public know about preventing cancer? Results from the Health Information National Trends Survey (HINTS). Health Educ Behav, 37(4), 490-503. doi: 10.1177/1090198106296770 Helft, P. R., Hlubocky, F., & Daugherty, C. K. (2003). American oncologists' views of internet use by cancer patients: a mail survey of American Society of Clinical Oncology members. J Clin Oncol, 21(5), 942-947. Hesse, B. W., Moser, R. P., Rutten, L. J., & Kreps, G. L. (2006). The health information national trends survey: research from the baseline. J Health Commun, 11 Suppl 1, vii-xvi. doi: 10.1080/10810730600692553 Hesse, B. W., Nelson, D. E., Kreps, G. L., Croyle, R. T., Arora, N. K., Rimer, B. K., & Viswanath, K. (2005). Trust and sources of health information: the impact of the Internet and its implications for health care providers: findings from the first Health Information National Trends Survey. Arch Intern Med, 165(22), 26182624. doi: 10.1001/archinte.165.22.2618 Hewitson, P., Glasziou, P., Watson, E., Towler, B., & Irwig, L. (2008). Cochrane systematic review of colorectal cancer screening using the fecal occult blood test (hemoccult): an update. Am J Gastroenterol, 103(6), 1541-1549. doi: 10.1111/j.1572-0241.2008.01875.x HINTS 2003 Final Report. (2003): National Cancer Institute. Holmbeck, G. N. (2002). Post-hoc probing of significant moderational and mediational effects in studies of pediatric populations. J Pediatr Psychol, 27(1), 87-96. Huang, G. J., & Penson, D. F. (2008). Internet health resources and the cancer patient. Cancer Invest, 26(2), 202-207. doi: 10.1080/07357900701566197 Ilic, D. (2010). The role of the internet on patient knowledge management, education, and decision-making. Telemed J E Health, 16(6), 664-669. doi: 0.1089/tmj.2010.0003 Jadad, A. R., & Gagliardi, A. (1998). Rating health information on the Internet: navigating to knowledge or to Babel? JAMA, 279(8), 611-614. Jennings-Dozier, K., & Lawrence, D. (2000). Sociodemographic predictors of adherence to annual cervical cancer screening in minority women. Cancer Nurs, 23(5), 350356; quiz 357-358. King, S. (2010). Pink diplomacy: on the uses and abuses of breast cancer awareness. Health Communication, 25(3), 286-289. doi: 10.1080/10410231003698960

104

Kline, R. B. (2005). Principles and practice of structural equation modeling (Second ed.). New York, NY: The Guilford Press. Koch-Weser, S., Bradshaw, Y. S., Gualtieri, L., & Gallagher, S. S. (2010). The Internet as a Health Information Source: Findings from the 2007 Health Information National Trends Survey and Implications for Health Communication. J Health Commun, 15(sup3), 279-293. doi: 10.1080/10810730.2010.522700 Kohler, B. A., Ward, E., McCarthy, B. J., Schymura, M. J., Ries, L. A., Eheman, C., . . . Edwards, B. K. (2011). Annual report to the nation on the status of cancer, 19752007, featuring tumors of the brain and other nervous system. J Natl Cancer Inst, 103(9), 714-736. doi: 10.1093/jnci/djr077 Lin, K., Lipsitz, R., Miller, T., & Janakiraman, S. (2008). Benefits and harms of prostatespecific antigen screening for prostate cancer: an evidence update for the U.S. Preventive Services Task Force. Ann Intern Med, 149(3), 192-199. Ling, B. S., Klein, W. M., & Dang, Q. (2006). Relationship of communication and information measures to colorectal cancer screening utilization: results from HINTS. J Health Commun, 11 Suppl 1, 181-190. doi: 0.1080/10810730600639190 MacKinnon, D. P. (2008). Introduction to Statistical Mediation Analysis. New York, NY: Taylor & Francis Group, LLC. MacKinnon, D. P., & Fritz, M. S. (2007). Distribution of the product confidence limits for the indirect effect: Program PRODCLIN. Behavior Research Methods, 39(3), 384-389. Malone, M., Harris, R., Hooker, R., Tucker, T., Tanna, N., & Honnor, S. (2004). Health and the Internet--changing boundaries in primary care. Fam Pract, 21(2), 189191. Mandelblatt, J. S., Cronin, K. A., Bailey, S., Berry, D. A., de Koning, H. J., Draisma, G., . . . Feuer, E. J. (2009). Effects of mammography screening under different screening schedules: model estimates of potential benefits and harms. Ann Intern Med, 151(10), 738-747. doi: 10.1059/0003-4819-151-10-200911170-00010 Marcus, A. C., & Crane, L. A. (1998). A review of cervical cancer screening intervention research: implications for public health programs and future research. Prev Med, 27(1), 13-31. doi: 10.1006/pmed.1997.0251 Massad, L. S., Evans, C. T., Weber, K. M., Goderre, J. L., Hessol, N. A., Henry, D., . . . Wilson, T. E. (2010). Changes in knowledge of cervical cancer prevention and human papillomavirus among women with human immunodeficiency virus. Obstet Gynecol, 116(4), 941-947. doi: 10.1097/AOG.0b013e3181f2dbae

105

McMullan, M. (2006). Patients using the Internet to obtain health information: how this affects the patient-health professional relationship. Patient Educ Couns, 63(1-2), 24-28. doi: 10.1016/j.pec.2005.10.006 Meissner, H. I., Smith, R. A., Rimer, B. K., Wilson, K. M., Rakowski, W., Vernon, S. W., & Briss, P. A. (2004). Promoting cancer screening: Learning from experience. Cancer, 101(5 Suppl), 1107-1117. doi: 10.1002/cncr.20507 Murray, E., Lo, B., Pollack, L., Donelan, K., Catania, J., White, M., . . . Turner, R. (2003). The impact of health information on the internet on the physician-patient relationship: patient perceptions. Arch Intern Med, 163(14), 1727-1734. doi: 10.1001/archinte.163.14.1727 Muthen, L. K., & Muthen, B. O. (2010). Mplus (Version 6.11). Los Angeles. National Cancer Institute. HINTS - GEM Measures, from http://secure.mmgct.com/hintsgem/itemdetails.aspx?itemid=261 National Cancer Institute. (2010). Fact Sheet: Mammograms. In National Institutes of Health (Ed.). National Center for Chronic Disease Prevention and Health Promotion. (2011). Colorectal cancer. Atlanta, GA: Centers for Disease Control and Prevention. Nelson, D. E., Kreps, G. L., Hesse, B. W., Croyle, R. T., Willis, G., Arora, N. K., . . . Alden, S. (2004). The Health Information National Trends Survey (HINTS): development, design, and dissemination. J Health Commun, 9(5), 443-460; discussion 481-444. doi: 10.1080/10810730490504233 Nelson, D. E., Powell-Griner, E., Town, M., & Kovar, M. G. (2003). A comparison of national estimates from the National Health Interview Survey and the Behavioral Risk Factor Surveillance System.. Am J Public Health, 93(8), 1335-1341. Nelson, H. D., Tyne, K., Naik, A., Bougatsos, C., Chan, B. K., & Humphrey, L. (2009). Screening for breast cancer: an update for the U.S. Preventive Services Task Force. Ann Intern Med, 151(10), 727-737, W237-742. doi: 10.1059/0003-4819151-10-200911170-00009 Nelson, W., Moser, R. P., Gaffey, A., & Waldron, W. (2009). Adherence to cervical cancer screening guidelines for U.S. women aged 25-64: data from the 2005 Health Information National Trends Survey (HINTS). J Womens Health (Larchmt), 18(11), 1759-1768. doi: 10.1089/jwh.2009.1430 O'Malley, A. S., Forrest, C. B., & Mandelblatt, J. (2002). Adherence of low-income women to cancer screening recommendations. J Gen Intern Med, 17(2), 144-154.

106

O'Malley, M. S., Earp, J. A., Hawley, S. T., Schell, M. J., Mathews, H. F., & Mitchell, J. (2001). The association of race/ethnicity, socioeconomic status, and physician recommendation for mammography: who gets the message about breast cancer screening. Am J Public Health, 91(1), 49-54. Office on Women's Health. (2010). Pap test: Frequently Asked Questions. Retrieved from http://www.womenshealth.gov/publications/our-publications/fact-sheet/paptest.pdf. Ong, L. M., de Haes, J. C., Hoos, A. M., & Lammes, F. B. (1995). Doctor-patient communication: a review of the literature. Soc Sci Med, 40(7), 903-918. Powers, D. A., & Xie, Y. (2008). Statistical Models for Categorical Data Analysis (2nd edition ed.). Bingley, UK: Emerald Group Publishing Limited. Reisig, V., & Wildner, M. (2008). Primary Prevention. In W. Kirch (Ed.), Encylcopedia of Public Health (Vol. 1, pp. 1141-1143). New York, NY: Springer Science+Business Media, LLC. Rice, R. E. (2006). Influences, usage, and outcomes of Internet health information searching: multivariate results from the Pew surveys. International Journal of Medical Informatics, 75(1), 8-28. doi: 10.1016/j.ijmedinf.2005.07.032 Rigal, L., Saurel-Cubizolles, M. J., Falcoff, H., Bouyer, J., & Ringa, V. (2011). Do social inequalities in cervical cancer screening persist among patients who use primary care? The Paris Prevention in General Practice survey. Prev Med. doi: 10.1016/j.ypmed.2011.06.016 Rim, S. H., Joseph, D. A., Steele, C. B., Thompson, T. D., & Seeff, L. C. (2011). Colorectal Cancer Screening - United States, 2002, 2004, 2006, 2008 Mortality & Morbidity Weekly Report (Vol. 60, pp. 42-46): Centers for Disease Control and Prevention. Sadler, G. R., Ko, C. M., Cohn, J. A., White, M., Weldon, R. N., & Wu, P. (2007). Breast cancer knowledge, attitudes, and screening behaviors among African American women: the Black cosmetologists promoting health program. BMC Public Health, 7, 57. doi: 10.1186/1471-2458-7-57 Sasieni, P., Castanon, A., & Cuzick, J. (2009). Effectiveness of cervical screening with age: population based case-control study of prospectively recorded data.BMJ, 339, b2968. doi: 10.1136/bmj.b2968 Schroder, F. H. (2009). PSA screening--a review of recent studies. Eur J Cancer, 45 Suppl 1, 402-404. doi: 10.1016/S0959-8049(09)70065-6

107

Schroder, F. H., Hugosson, J., Roobol, M. J., Tammela, T. L., Ciatto, S., Nelen, V., . . . Auvinen, A. (2009). Screening and prostate-cancer mortality in a randomized European study. N Engl J Med, 360(13), 1320-1328. doi: 10.1056/NEJMoa0810084 Shi, L., Lebrun, L. A., Zhu, J., & Tsai, J. (2011). Cancer screening among racial/ethnic and insurance groups in the United States: a comparison of disparities in 2000 and 2008. J Health Care Poor Underserved, 22(3), 945-961. doi: 0.1353/hpu.2011.0079 Shim, M., Kelly, B., & Hornik, R. (2006). Cancer information scanning and seeking behavior is associated with knowledge, lifestyle choices, and screening. J Health Commun, 11 Suppl 1, 157-172. doi: 10.1080/10810730600637475 Shokar, N. K., Vernon, S. W., & Weller, S. C. (2005). Cancer and colorectal cancer: knowledge, beliefs, and screening preferences of a diverse patient population. Fam Med, 37(5), 341-347. Smith, R. A., Cokkinides, V., Brooks, D., Saslow, D., & Brawley, O. W. (2010). Cancer screening in the United States, 2010: a review of current American Cancer Society guidelines and issues in cancer screening. CA Cancer J Clin, 60(2), 99119. doi: 10.3322/caac.20063 Smith, S. W., Nazione, S., Laplante, C., Kotowski, M. R., Atkin, C., Skubisz, C. M., & Stohl, C. (2009). Topics and sources of memorable breast cancer messages and their impact on prevention and detection behaviors. J Health Commun, 14(3), 293-307. doi: 10.1080/10810730902805903 Solomon, D., Breen, N., & McNeel, T. (2007). Cervical cancer screening rates in the United States and the potential impact of implementation of screening guidelines. CA Cancer J Clin, 57(2), 105-111. Stevenson, F. A., Kerr, C., Murray, E., & Nazareth, I. (2007). Information from the Internet and the doctor-patient relationship: the patient perspective--a qualitative study. BMC Fam Pract, 8, 47. doi: 10.1186/1471-2296-8-47 Stewart, M. A. (1995). Effective physician-patient communication and health outcomes: a review. CMAJ, 152(9), 1423-1433. Subramanian, S., Klosterman, M., Amonkar, M. M., & Hunt, T. L. (2004). Adherence with colorectal cancer screening guidelines: a review. Prev Med, 38(5), 536-550. doi: 10.1016/j.ypmed.2003.12.011 . Susan G. Komen for the Cure. (2012) Retrieved March 5, 2012, from http://ww5.komen.org/AboutUs/OurWork.html

108

Taylor, V., Lessler, D., Mertens, K., Tu, S. P., Hart, A., Chan, N., . . . Thompson, B. (2003). Colorectal cancer screening among African Americans: the importance of physician recommendation. J Natl Med Assoc, 95(9), 806-812. U.S. Cancer Statistics Working Group. (2010). United States cancer statistics: 1999-2007 incidence and mortality web-based report Retrieved September 13, 2011, from http://apps.nccd.cdc.gov/uscs/ U.S. Preventive Services Task Force. (2002). Screening for Breast Cancer (2002), from http://www.uspreventiveservicestaskforce.org/uspstf/uspsbrca2002.htm U.S. Preventive Services Task Force. (2003). Screening for Cervical Cancer, from http://www.uspreventiveservicestaskforce.org/uspstf/uspscerv.htm U.S. Preventive Services Task Force. (2008a). Screening for Colorectal Cancer, from http://www.uspreventiveservicestaskforce.org/uspstf/uspscolo.htm U.S. Preventive Services Task Force. (2008b). Screening for colorectal cancer: U.S. Preventive Services Task Force recommendation statement.Ann Intern Med, 149(9), 627-637. U.S. Preventive Services Task Force. (2008c). Screening for Prostate Cancer, from http://www.uspreventiveservicestaskforce.org/uspstf/uspsprca.htm U.S. Preventive Services Task Force. (2008d). Screening for prostate cancer: U.S. Preventive Services Task Force recommendation statement. Ann Intern Med, 149(3), 185-191. U.S. Preventive Services Task Force. (2009). Screening for breast cancer: U.S. Preventive Services Task Force recommendation statement. Ann Intern Med, 151(10), 716-726, W-236. doi: 10.1059/0003-4819-151-10-200911170-00008 U.S. Preventive Services Task Force. (2011a). Draft recommendation statement. Screening for Prostate Cancer: U.S. Preventive Services Task Force Recommendation Statement Retrieved February 2012, from http://www.uspreventiveservicestaskforce.org/uspstf12/prostate/draftrecprostate.h tm U.S. Preventive Services Task Force. (2011b). Methods and Processes Retrieved February, 2012, from http://www.uspreventiveservicestaskforce.org/methods.htm U.S. Preventive Services Task Force. (2012). Screening for Cervical Cancer.

109

Vandenberg, R. J., & Grelle, D. M. (2009). Alternative model specifications in structural equation modeling: Facts, fictions, and truth. In C. E. Lance & R. J. Vandenberg (Eds.), Statistical and Methodological Myths and Urban Legends (pp. 165-191). New York, NY: Taylor & Francis Group, LLC. Vesco, K. K., Whitlock, E. P., Eder, M., Burda, B. U., Senger, C. A., & Lutz, K. (2011). Risk factors and other epidemiologic considerations for cervical cancer screening: a narrative review for the U.S. Preventive Services Task Force. Ann Intern Med, 155(10), 698-705, W216. doi: 10.1059/0003-4819-155-10-201111150-00377 Viswanath, K., & Kreuter, M. W. (2007). Health disparities, communication inequalities, and eHealth. Am J Prev Med, 32(5 Suppl), S131-133. doi: 10.1016/j.amepre.2007.02.012 Wald, H. S., Dube, C. E., & Anthony, D. C. (2007). Untangling the Web--the impact of Internet use on health care and the physician-patient relationship. Patient Educ Couns, 68(3), 218-224. doi: 10.1016/j.pec.2007.05.016 Walton, R. J., Blanchet, M., Boyes, D. A., Carmichael, J. A., Marshall, K. G., Miller, A. B., . . . Hill, G. B. (1976). Cervical cancer screening programs. Canadian Medical Association Journal, 114(11), 1003-1033. Waxman, A. G. (2005). Guidelines for cervical cancer screening: history and scientific rationale. Clin Obstet Gynecol, 48(1), 77-97. Weaver, J. B., 3rd, Thompson, N. J., Weaver, S. S., & Hopkins, G. L. (2009). Healthcare non-adherence decisions and internet health information. Computers in Human Behavior, 25, 1373-1380. Wildner, M., & Nennstiel-Ratzel, U. (2008). Secondary Prevention. In W. Kirch (Ed.), Encyclopedia of Public Health (Vol. 1, pp. 1143-1145). New York, NY: Springer Science+Business Media, LLC. Winterich, J. A., Grzywacz, J. G., Quandt, S. A., Clark, P. E., Miller, D. P., Acuna, J., . . . Arcury, T. A. (2009). Men's knowledge and beliefs about prostate cancer: education, race, and screening status. Ethn Dis, 19(2), 199-203. Yu, C. (2002). Evaluating cutoff criteria of model fit indices for latent variable models with binary and continuous outcomes. Ph.D., University of California. Retrieved from http://statmodel2.com/download/Yudissertation.pdf Zachariae, R., Pedersen, C. G., Jensen, A. B., Ehrnrooth, E., Rossen, P. B., & von der Maase, H. (2003). Association of perceived physician communication style with patient satisfaction, distress, cancer-related self-efficacy, and perceived control over the disease. Br J Cancer, 88(5), 658-665. doi: 10.1038/sj.bjc.6600798

110

Zolnierek, K. B., & Dimatteo, M. R. (2009). Physician communication and patient adherence to treatment: a meta-analysis. Med Care, 47(8), 826-834. doi: 10.1097/MLR.0b013e31819a5acc

Appendix 1 HINTS conceptual framework

D. E. Nelson et al. (2004)

111