J Med Screen 2000;7:99–104
99
Predictors of attendance in the United Kingdom flexible sigmoidoscopy screening trial S Sutton, J Wardle, T Taylor, K McCaVery, S Williamson, R Edwards, J Cuzick, A Hart, J Northover, W Atkin
ICRF Health Behaviour Unit, Dept Epidemiology & Public Health, University College London S Sutton J Wardle T Taylor K McCaVery S Williamson ICRF Dept Mathematics, Statistics & Epidemiology, Lincoln’s Inn Fields, London R Edwards J Cuzick Leicester General Hospital, Leicester A Hart ICRF Colorectal Cancer Unit, St Mark’s Hospital, Northwick Park, Harrow J Northover W Atkin Correspondence to: Dr Sutton email:
[email protected] Accepted for publication 31 May 2000
Abstract Objective—To investigate predictors of attendance in the United Kingdom flexible sigmoidoscopy screening trial. Design—Prospective design in which participants completed a postal questionnaire before being sent their invitation for screening. Setting—Welwyn Garden City and Leicester, United Kingdom. Participants—A total of 2758 patients aged 55 to 64, registered with general practices in the two centres, who (a) expressed interest in having the screening test, (b) completed a postal questionnaire, and (c) were subsequently invited for screening. Main results—The attendance rate among questionnaire responders was 76.1%. Multiple logistic regression analysis yielded a final model that included nine independent predictors of attendance. Patients with the following characteristics were more likely to attend: men; home owners; non-smokers; those who had regular check ups at the dentist; those with better subjective health; those who minded less about having medical tests; those who said they would definitely rather than probably take up the oVer of sigmoidoscopy screening; and those who perceived less barriers and more benefits to having the test. Conclusions—The findings are broadly consistent with previous studies of screening participation, although subjective health emerged as an important predictor in this study. There was no evidence for “reverse targeting”: attenders were not at lower (or higher) risk for colorectal cancer compared with non-attenders. The findings relating to attitudes and beliefs could be used in eVorts to improve attendance, for example by developing information leaflets that address barriers to screening. Other findings could be used to target interventions to subgroups that have relatively low rates of screening participation. (J Med Screen 2000;7:99–104) Keyword: sigmoidoscopy
Colorectal cancer is the second most common cause of cancer death in the western world. Screening by “once only” flexible sigmoidoscopy around the age of 60 has been proposed as a way of preventing colorectal cancer by finding and removing precancerous adenomas.1 In 1995, a multicentre, randomised controlled trial was started in the
www.jmedscreen.com
United Kingdom to evaluate this screening method.2 Identifying factors associated with participation is an important part of the evaluation of a new screening procedure. Such information is relevant for interpreting the eYcacy findings. For instance, so called “reverse targeting” may occur: participants may tend to be at lower than average risk for the condition being screened.3 Such information can also be used to identify groups with low attendance that can be targeted in interventions designed to improve uptake, and to inform the development of such interventions. In the longer term, information about the factors associated with participation will need to be considered when deciding whether, and how, a national screening programme should be implemented. Identifying predictors of attendance also contributes to the wider literature on factors that influence individuals’ decisions to accept or decline invitations to participate in screening and other medical tests. This paper reports findings on predictors of attendance from the pilot centre and the first main trial centre of the United Kingdom flexible sigmoidoscopy screening trial. This study diVers from most previous studies4 in its use of a prospective design, in which potential predictors were measured before the invitation for screening was received; the large sample size; and the wide range of predictors that were investigated, which included demographic factors, health behaviours, attitudes and beliefs, recent symptoms, and personal and family history of disease. Similar variables have been shown to predict participation in faecal occult blood testing4 and breast and cervical screening.5 Methods PARTICIPANTS, DESIGN, AND PROCEDURE
Potential participants were patients aged 55 to 64, registered with general practices in Welwyn Garden City and Leicester. Lists of names and addresses of patients in the target practices were provided by the health authorities. General practitioners (GPs) were asked to exclude any patients who were inappropriate for the trial (for example, already had bowel cancer, recently had sigmoidoscopy). This resulted in the exclusion of about 1.8% of potential participants. Letters signed by the GPs were sent to the remaining patients, informing them that a trial of bowel cancer screening was due to be set up in their area, and asking whether they would take up the oVer if they were invited to have the test
100
Sutton, Wardle, Taylor, et al
A test called the “Flexi-scope” test may become available as a way of helping to prevent bowel cancer. Below is some information about the test. Preventing bowel cancer with the Flexi-scope screening test + Bowel cancer is the second most common cancer in the United Kingdom. + Screening means doing simple tests to pick up hidden problems. + The screening test being developed is called the Flexi-scope test. + The Flexi-scope screening test would be for all people in your age group, including those who feel well and have no bowel symptoms. + The test is safe and is widely available in America, but needs to be assessed in Britain. + The Flexi-scope is a thin flexible tube with a tiny camera on the end. A doctor inserts the Flexi-scope into the back passage and looks for bowel polyps. Bowel polyps are harmless growths which are not cancerous, but if left can become cancerous. Polyps can be removed quickly and painlessly with the Flexiscope. Removing these polyps helps to prevent bowel cancer. + The Flexi-scope test would be free and would take only 5 minutes. It would be done at a hospital clinic. + We think that just one test could halve your risk of bowel cancer for the next 10 years.
(screening interest question). An information sheet describing flexible sigmoidoscopy screening was included (see box), along with a prepaid reply envelope. Non-responders were sent a reminder after two weeks. Patients who responded “yes, definitely” or “yes, probably” to the screening interest question were eligible for the trial. They were randomised to screening or control in the ratio 1:2. People living at the same address were randomised to the same arm of the trial. Those in the screening group were sent an invitation for screening with a specified appointment about five weeks in advance. Patients were asked to telephone to confirm, change, or cancel their appointment. Initially, in Welwyn Garden City, an attempt was made to telephone the non-responders to encourage them to attend. However, this proved unproductive, so it was abandoned in favour of sending a postal reminder after two weeks. Those who confirmed their appointment were mailed a bowel preparation (laxative or enema) to self administer prior to attendance. Screening was performed in endoscopy units by experienced specialist gastroenterologists. After the GP exclusions, a total of 34 302 patients in the two centres were sent the GP letter and information about bowel cancer screening. Random samples of patients were sent a seven page questionnaire designed to measure factors related to subsequent screen-
www.jmedscreen.com
ing participation. This was sent either with the initial letter from the GP (in which case the screening interest question was incorporated in the questionnaire; sample A), or at a later stage after respondents who had expressed interest in having the screening test had been randomised to screening, but before they were sent their invitation (sample B; see figure 1). The latter procedure was adopted in Leicester as a more cost eVective way of collecting information on predictors of attendance—more cost eVective because the questionnaire was only sent to those who had already been randomised to screening. The main analyses in this report were based on the 2758 patients who (a) responded “yes, definitely” or “yes, probably” to the screening interest question, (b) completed the questionnaire, and (c) were subsequently invited for screening. Two people who returned the questionnaire but died before their screening appointment have been excluded from the analysis. MEASURES
The following variables were measured. All measures were obtained by questionnaire unless specified otherwise. 1) Sociodemographic factors—sex and age (obtained from patient lists); ethnicity; marital status; employment; educational qualifications; housing tenure; car access; and whether the person was invited with another member of the same household or on their own (obtained from the trial database). 2) Health behaviours—mammography; smoking; exercise; dental check ups; frequency of eating fruit. 3) Attitudes and beliefs about bowel cancer and bowel screening—perceived susceptibility to bowel cancer; worry about bowel cancer; perceived benefits and barriers to having flexible sigmoidoscopy screening; attitude to having medical tests; screening interest question. 4) Other variables—anxiety, measured using the short form of the Spielberger State Trait Anxiety Inventory6 7; optimism, assessed with the 12 item Life Orientation Test8; subjective health; number of bowel symptoms experienced in the last three months; family history of bowel cancer; personal history of bowel disease. STATISTICAL ANALYSIS
Statistical analysis was by univariate and multivariate logistic regression using STATA 6.0. Adjusted confidence intervals were calculated that took account of the clustering of the sample by GP practice and by household. However, since these diVered only slightly from the unadjusted intervals, only the latter are reported here. The population attributable fraction was calculated for each of the significant categorical variables in the multivariate analysis using the method described by Brady.9 This can be interpreted either as the estimated proportion of non-attenders who failed to attend for
101
Predictors of attendance
SAMPLE A
SAMPLE B
Sent questionnaire (including screening interest question) 5099
Sent screening interest question on its own 25 574
Returned questionnaire 3475
Responded to interest question 20 078
Interested in having test 2854
Interested in having test 15 654
Randomised to screening 870
Randomised to screening 4573
Sent questionnaire (excluding screening interest question) 2203*
Returned questionnaire 1890†
Attended for screening 655
Attended for screening 1446
Figure 1 Flow diagram showing the design of the study. * Includes six people who died before screening. † Includes two people who died before screening. The sample for the analysis of predictors of attendance (n = 2758) consisted of the two subsamples shown in bold, excluding the two who died before screening.
screening because of a given risk factor or predictor variable, or as the expected proportional reduction in the overall non-attendance rate if the risk factor were eliminated, keeping the other variables constant. Results Response rates were good for a postal questionnaire study. In sample B, 78% responded to the screening interest question, and 78% of these expressed interest in having the test. The response rate to the questionnaire in this sample was 85.8%, significantly higher than that in sample A (68.2%) where the questionnaire included the screening interest question (÷2(1) = 245.7, p < 0.001). The most plausible reason for this diVerence is that those who were sent the questionnaire in sample B were a self selected group who had previously responded to the screening interest question. Of those who returned the questionnaire in sample A, 82.1% expressed interest in having the test.
www.jmedscreen.com
Response rates were consistently higher among women than among men for both the screening interest question (sample B; 77.3% v 69.4%; ÷2(1) = 376.6, p < 0.001) and the questionnaire (77.3% v 69.4%; ÷2(1) = 59.1, p < 0.001). The attendance rate among questionnaire responders was 76.1%; it was similar in the two samples (figure 1). There were no important diVerences in predictors of attendance between the two samples, so the results are reported for the combined sample only. Table 1 shows the association between each of the potential predictor variables and attendance for screening in terms of attendance rates and univariate odds ratios. There were no significant diVerences in attendance by age, ethnicity, or whether the person was invited with a cohabitee or on their own. On the other hand, marital status (which was significantly associated with the latter variable) was predictive of attendance: married people were more likely to attend. All the other demographic factors showed significant associations with attendance. Thus, men, those in employment, those with educational qualifications, those who owned or were buying their home, and those who had access to a car, were significantly more likely to attend for screening. Housing tenure and car access had the largest odds ratios among the sociodemographic factors. With the exception of the frequency of eating fruit, health behaviours were consistently related to screening attendance. Among women, having had a mammogram in the last three years was particularly strongly related to attendance. In addition, non-smokers, those who said they took regular exercise, and those who went to the dentist for regular check ups, were significantly more likely to attend for screening. Neither a family history of bowel cancer nor a personal history of bowel disease was related to attendance. However, those who reported four or more recent bowel symptoms were significantly more likely to attend than those who reported none or one. Higher attendance rates were also found among those with better subjective health, lower anxiety, and greater optimism. In response to the item measuring perceived susceptibility to bowel cancer, the vast majority of respondents said that their risk was about the same as other men and women of their age. There was no association with screening attendance. On the worry item, those who were “a bit worried” about getting bowel cancer were most likely to attend; their attendance rate was significantly higher than among those who were “not worried at all”. The question assessing general attitude to medical tests showed a very strong association with screening attendance. Those who said that they “don’t mind at all” had odds of attending for screening that were five and half times those for the small minority who said they “mind a lot”. Of all the categorical variables, this had the strongest relationship with attendance.
102
Sutton, Wardle, Taylor, et al
Specific beliefs about having flexible sigmoidoscopy were also predictive of attendance. Those who perceived more benefits and fewer barriers were more likely to attend. Further analysis showed that each individual item was significantly and consistently related to attendance. The two items that showed the strongest associations with attendance were “having the test would take up too much time” and “the test would give me peace of mind”. Finally, the screening interest question had a strong association with attendance. Those who said
that they would definitely take up the oVer were more likely to attend than those who said they would probably do so. A multiple logistic regression analysis was conducted to identify independent predictors of attendance. All the variables listed in table 1 were entered with the exception of mammography (which applied to women only). The following variables emerged as significant independent predictors of attendance: sex, housing tenure, smoking, visiting the dentist, subjective health, attitudes to medical tests, barriers, and
Table 1: Predictors of attendance Variable
Response categories
Frequencies N %
Attendance rate (%)
Odds ratio (95% confidence interval) Univariate Multivariate
Age
61 to 65 55 to 60 White Other With cohabitee Alone Married Not married Male Female Working Not working Qualifications No qualifications Home owner Rented accommodation Car No car At least every day Less than every day Yes
1557 1201 2662 68 1266 1492 2282 453 1306 1452 1474 1243 1238 1471 2342 357 2429 294 1204 1513 1297
56.5 43.5 97.5 2.5 45.9 54.1 83.4 16.6 47.4 52.6 54.3 45.7 45.7 54.3 86.8 13.2 89.2 10.8 44.3 55.7 92.4
76.4 75.9 76.5 72.1 77 75.5 77.3 71.7 78.8 73.8 78.7 73.7 79.4 73.9 78.2 65.8 77.9 65 77.7 75.5 75.9
1.03 (0.87:1.23)
§
1.26 (0.74:2.16)
§
1.09 (0.91:1.3)
§
1.34* (1.07:1.69)
§
1.32† (1.1:1.57)
1.29* (1.03:1.61)
1.32† (1.1:1.57)
§
1.36‡ (1.14:1.49)
§
No
107
7.6
55.1
No Yes Yes No Yes No Positive Negative Positive Negative 4+ 2, 3 0, 1 Excellent Good Fair Poor (Multi-item scale) (Multi-item scale) Higher
2244 488 1846 859 1944 601 368 2390 500 2258 1071 966 721 388 1707 567 66
82.1 17.9 68.2 31.8 76.4 23.6 13.3 86.7 18.1 81.9 38.8 35 26.1 14.2 62.6 20.8 2.4
78.3 67.4 77.7 74 79.2 69.6 78.8 75.8 77.4 75.9 79.3 74.7 73.5 78.9 78.3 71.8 50
115
4.4
About the same
2173
83.3
Lower
321
12.3
76
95 302 1475 851 1558 1080 95
3.5 11.1 54.2 31.3 57 39.5 3.5
75.8 76.8 78.8 72 80.7 73.1 43.2
Ethnicity Invited with cohabitee or on their own Marital status Sex Employment Education Housing tenure Car access How many days a week do you eat fruit? (Women only) Have you had a breast screen (mammogram) in the last 3 years? Do you smoke cigarettes at all nowadays? Do you take regular exercise? Do you go to the dentist for a regular check-up? Family history of bowel cancer Personal history of bowel disease Bowel symptoms Would you say that for someone of your age your own health in general is... Anxiety score Optimism score Compared with other men and women of your age, do you think your chances of getting bowel cancer are: How worried are you about getting bowel cancer?
Very worried Quite worried A bit worried Not worried at all Do you mind having medical Don’t mind at all Mind a bit tests? Mind a lot Perceived benefits of (Multi-item scale) screening Perceived barriers to (Multi-item scale) screening If you were invited to have Definitely the bowel cancer screening test, would you take up the Probably oVer?
1.86‡ (1.46:2.36)
1.65† (1.2:2.26)
1.9‡ (1.47:2.47)
§
1.13 (0.95:1.36)
§
2.56‡ (1.71:3.82)
¶
1.75‡ (1.41:2.17)
1.66‡ (1.26:2.18)
1.22* (1.01:1.47)
§
1.66‡ (1.35:2.04)
1.38* (1.08:1.77)
1.19 (0.91:1.55)
§
1.09 (0.86:1.37)
§
1.38† (1.1:1.72) 1.07 (0.86:1.33)
§
3.73‡ (2.17:6.41) 3.61‡ (2.2:5.93) 2.54‡ (1.52:4.26)
3.08† (1.55:6.11) 3.47‡ (1.83:6.58) 2.57† (1.33:4.96)
80.9
0.94‡ (0.92:0.97) 1.04‡ (1.02:1.06) 1.33 (0.78:2.27)
§ § §
77.1
1.06 (0.81:1.4)
2015
73.1
82.2
743
26.9
59.9
1.22 (0.74:1.99) 1.29 (0.95:1.75) 1.45‡ (1.19:1.76)
§
5.5‡ (3.6:8.41) 3.59‡ (2.34:5.5)
2.08† (1.21:3.59) 2† (1.18:3.38)
1.16‡ (1.12:1.2)
1.06* (1.01:1.11)
0.83‡ (0.8:0.86)
0.9‡ (0.86:0.95)
3.09‡ (2.57:3.72)
2.27‡ (1.78:2.91)
For each categorical variable, the odds ratios refer to the odds of attending for screening in a given category relative to the final category. For the multi-item scales, the odds ratios estimate the change in odds of attending that is associated with a change of one scale point on the composite scale. * p < 0.05. † p < 0.01. ‡ p < 0.001. § Not significant at p < 0.05 in the multivariate analysis. ¶ Not included in the multivariate analysis.
www.jmedscreen.com
103
Predictors of attendance
screening interest. A backward stepwise analysis (using probabilities of 0.05 and 0.10 for entry and removal respectively) yielded a final model that included only these variables plus perceived benefits. The model was highly significant (÷2(12) = 232.0, p < 0.00005), and the proportional reduction in ÷2 compared to the constant only model (Pseudo R squared) was 9.9%. The multivariate odds ratios are given in table 1. Of the two continuous variables, barriers had a larger independent eVect than benefits. Of the categorical variables, subjective health had the largest independent predictive eVect, followed by screening interest and attitude to medical tests. Calculation of the population attributable fraction yielded a diVerent ranking of variables. Screening interest had the largest fraction (0.18), followed by sex (0.09), smoking (0.07), and visiting the dentist (0.06); attributable fractions for the remaining variables were less than 0.05. Thus, controlling for the other variables in the model, if the non-attendance rate among those who said they would probably have the test were reduced to the rate among those who said they would definitely have the test, then the overall non-attendance rate would be reduced by an estimated 18%. Discussion The present study was one component of a randomised controlled trial to evaluate the eYcacy of flexible sigmoidoscopy screening. In order to maximise statistical power for detecting an eVect of screening on mortality, only people who responded “yes, definitely” or “yes, probably” to the screening interest question were eligible for the trial. This meant that the analysis of predictors of attendance in the present study was necessarily restricted to this group. Therefore, we cannot rule out the possibility of selection bias. Because those who responded “probably not” or “definitely not” to the screening interest question and those who did not give any response to the question were not invited for screening, we have no information on predictors of attendance in these groups. It seems likely that, had these two groups been invited for screening, attendance would have been much lower than among the interested group. More importantly, the predictors of attendance might have been systematically diVerent in the not interested and nonresponder groups. The possibility of selection bias in the present study limits the generalisability of the findings and comparability with other studies of screening attendance. On the other hand, explicitly restricting the analysis of predictors of attendance to the interested group allows one to interpret the findings in terms of the barriers to attendance among people who, at least on the basis of their prior expression of interest, would be expected to attend for screening. Furthermore, the findings from the present study have implications for increasing attendance in this trial and in other trials that use a similar recruitment procedure. Although the sample was relatively homogeneous in terms of screening interest, the analy-
www.jmedscreen.com
sis revealed a number of important predictors of attendance. Men were more likely to attend for screening than women. This association was also seen in the multivariate analysis, suggesting that none of the other variables could account for the eVect. On the other hand, women had a higher response rate to the screening interest question and to the questionnaire, and male respondents were more likely than female respondents to express interest in having the screening test.10 Thus, gender eVects showed a complex pattern overall. The net eVect was that men and women in the initial sample were about equally likely to attend for screening. Consistent with previous studies of screening participation,4 5 11 higher socioeconomic status was associated with higher screening attendance. Respondents who were currently in employment, had educational qualifications, owned or were buying their home, or had access to a car, were more likely to attend. As well as serving as an indicator of socioeconomic status, having access to a car should have made it easier to attend for screening. However, the fact that it did not emerge as a significant independent predictor in the multivariate analysis suggests that it did not facilitate attendance in this way. Indeed, of the socioeconomic factors, only housing tenure emerged as a significant independent predictor. In previous research, the practice of preventive health behaviours has been shown to be fairly consistently related to screening attendance,5 12 although studies have diVered with respect to the particular behaviour found to be associated with attendance. In the present study, two of the four health behaviours measured (not smoking, and going to the dentist) were significantly associated with screening attendance in the multivariate analysis. These measures may be capturing an orientation to engage in behaviour to protect one’s health or possibly a tendency to comply with invitations and requests from health professionals. The relationship among women between prior mammography and attendance for sigmoidoscopy screening may be interpreted in the same way. Although smoking can be regarded as a marker for material deprivation,13 in this dataset it was only weakly associated with measures of socioeconomic status, and its predictive eVect on screening attendance was maintained even when controlling for these measures. A simple rating of subjective health emerged as one of the strongest predictors of attendance. This measure has been found to be a robust predictor of mortality and other health outcomes,14 but it has been used in few previous studies of screening participation. The simplest interpretation of the eVect in the present study is that those who stated on the questionnaire that their health was poor were more likely to be ill at the time they received their screening invitation and were therefore less inclined, or less able, to attend for screening. The prevalence of “poor” self rated health in the sample was very low, however, and the estimated attributable fraction for this variable was small.
104
Sutton, Wardle, Taylor, et al
The other health related measures we included were specific to bowel disease. A personal history of bowel disease and a family history of bowel cancer were both found to be unrelated to screening attendance. The number of recent bowel symptoms was weakly associated with attendance but only in the univariate analysis. To the extent that these variables represent risk factors for colorectal cancer, there was no evidence in this study that attenders were at lower risk than nonattenders; in other words, there was no suggestion that reverse targeting occurred. The trial will yield valuable information on risk factors for the development of bowel polyps and colorectal cancer. Several of the attitude and belief measures were predictive of attendance, in particular, perceived barriers to sigmoidoscopy screening and attitude to medical tests, but also perceived benefits of sigmoidoscopy screening and the screening interest question itself. The eVects of all these variables persisted in the multivariate analysis. Screening interest had the largest attributable fraction of all the categorical variables, and analyses reported elsewhere show that perceived benefits and perceived barriers are important predictors of screening interest.10 These findings can be used in eVorts to improve attendance by modifying relevant beliefs and attitudes using information leaflets and advice from GPs and other health professionals. As part of the trial, we are currently evaluating the eVect of a booklet designed to address perceived barriers to having the test, including fear of embarrassment, fear of discomfort, and dislike of medical tests. Other findings from the study could be used to target interventions to subgroups who
www.jmedscreen.com
have relatively low rates of screening participation, for example smokers or people who live in rented accommodation. Financial support for the flexible sigmoidoscopy trial was provided by the Imperial Cancer Research Fund and the Medical Research Council.
1 Atkin WS, Cuzick J, Northover JMA, et al. Prevention of colorectal cancer by once-only flexible sigmoidoscopy. Lancet 1993;341:736–40. 2 Atkin WS, Hart A, Edwards R, et al. Uptake, yield of neoplasia, and adverse eVects of flexible sigmoidoscopy screening. Gut 1998;42:560–5. 3 Woolhandler S, Himmelstein DU. Reverse targeting of preventive care due to lack of health insurance. JAMA 1988;259:2872–4. 4 Vernon SW. Participation in cancer screening: a review. J Natl Cancer Inst 1997;89:1406–22. 5 Sutton S, Bicker G, Sancho-Aldridge J, et al. Prospective study of predictors of attendance for breast screening in inner London. J Epidemiol Community Health 1994;48:65– 73. 6 Spielberger CD, Gorsuch RL, Lushene RE, et al. Manual for the State-Trait Anxiety Inventory. Palo Alto, California: Consulting Psychologist Press, 1983. 7 Marteau TM, Bekker H. The development of a six-item short-form of the state scale of the Spielberger State-Trait Anxiety Inventory (STAI). Br J Clin Psychol 1992;31:301– 6. 8 Scheier MF, Weintraub JK, Carver CS. Coping with stress: divergent strategies of optimists and pessimists. J Personality Soc Psychol 1986;51:1257–64. 9 Brady AR. Adjusted population attributable fractions from logistic regression. Stata Technical Bulletin 1998;42:8–12. 10 Wardle J, Sutton S, Williamson S, et al. Psychosocial influences on participation in bowel cancer screening. Prev Med. In press. 11 Neilson AR, Whynes DK. Determinants of persistent compliance with screening for colorectal cancer. Soc Sci Med 1995;41:365–74. 12 Vernon SW, Laville EA, Jackson GL. Participation in breast screening programs: a review. Soc Sci Med 1990;30:1107– 18. 13 Wardle J, Farrell M, Hillsdon M, et al. Smoking, drinking, physical activity and screening uptake and health inequalities. In: Gordon D, Shaw M, Dorling D, et al, eds. Inequalities in health. The evidence presented to the Independent Inquiry into Inequalities in Health, chaired by Sir Donald Acheson. Bristol: The Policy Press, 1999;213–39. 14 McGee DL, Liao Y, Cao G, et al. Self-reported health status and mortality in a multiethnic US cohort. Am J Epidemiol 1999;149:41–6.
