Smoking Cessation in Relation to Total Mortality Rates in Women

Smoking Cessation in Relation to Total Mortality Rates in Women A Prospective Cohort Study Ichiro Kawachi, MB ChB; Graham A. Colditz, MB BS; Meir J. Stampfer, MD; Walter C. Willett, MD; JoAnn E. Manson, MD; Bernard Rosner, PhD; David J. Hunter, MB BS; Charles H. Hennekens, MD; and Frank E. Speizer, MD

• Objective: To examine the temporal relationship between stopping smoking and total mortality rates among middle-aged women. • Design: Prospective cohort study with 12 years of follow-up. • Setting: Registered nurses residing in the United States. • Participants: 117 001 female registered nurses, ages 30 to 55 years, who were free of manifest coronary heart disease, stroke, and cancer (except nonmelanoma skin cancer) in 1976. • Main Outcome Measures: Total mortality, further categorized into deaths from cardiovascular diseases, cancers, and violent deaths. • Results: A total of 2847 deaths (933 among "never smokers," 799 among former smokers, and 1115 among current smokers) occurred during 1.37 million person-years of follow-up. The multivariate relative risks for total mortality compared with never smokers were 1.87 (95% CI, 1.65 to 2.13) for current smokers and 1.29 (CI, 1.14 to 1.46) for former smokers. Participants who started smoking before the age of 15 years had the highest risks for total mortality (multivariate relative risk, 3.15; CI, 2.16 to 4.59), cardiovascular disease mortality (relative risk, 9.94; CI, 5.15 to 19.19), and deaths from external causes of injury (relative risk, 5.39; CI, 1.84 to 15.78). Compared with continuing smokers, former smokers had a 24% reduction in risk for cardiovascular disease mortality within 2 years of quitting. The excess risks for total mortality and both cardiovascular disease and total cancer mortality among former smokers approached the level of that for never smokers after 10 to 14 years of abstinence. The health benefits of cessation were clearly present regardless of the age at starting and daily number of cigarettes smoked. • Conclusions: The risk of cigarette smoking on total mortality among former smokers decreases nearly to that of never smokers 10 to 14 years after cessation.

Ann Intern Med. 1993;119:992-1000. From Harvard Medical School, Brigham and Women's Hospital, and Harvard School of Public Health, Boston, Massachusetts. For current author addresses, see end of text. 992

£* ormer smokers are at lower risk for total mortality compared with continuing smokers (1-8). However, the time required for the risk for total mortality among quitters to reach the level of never smokers differs among studies, ranging from 6 to 10 years in one study (9) to 15 or more years in other studies (7, 10). Few previous studies have been able to examine the temporal relationship between smoking cessation and decline in mortality risk while adjusting for other smoking-related factors, such as age at starting and number of cigarettes smoked. Further, most large prospective studies with data on smoking cessation have been based solely on men (5, 8, 10). The British female doctors study (11) had insufficient data to characterize the relationship between time since quitting and decrease in mortality. We analyzed the health benefits of smoking cessation in the Nurses' Health Study, a large prospective cohort of women in the United States. Our results describe the timing and extent of decrease in risks for total and cause-specific mortality for women who stopped smoking. Methods The Nurses' Health Study Cohort The Nurses' Health Study cohort was established in 1976, when 121 700 female registered nurses ages 30 to 55 years completed a mailed questionnaire requesting information about risk factors for cancer and coronary heart disease, including current and past smoking habits, past history of myocardial infarction, angina, cancer, diabetes, hypertension, high serum cholesterol levels, menopause, and parental history of myocardial infarction. Questions were also included on height, weight, postmenopausal use of hormones, and history of oral contraceptive use. Since 1976, follow-up questionnaires have been mailed every 2 years to update information on smoking behavior, other cardiovascular risk factors, and the development of major illness. If no questionnaire was returned during the follow-up years, the most recent record of exposure status was used for the subsequent follow-up interval (for smoking-related variables, this occurred in less than 0.5% of the cohort). Further details of the Nurses' Health Study have been published previously (12, 13). Exposure Data Women were categorized according to their smoking status as "never smokers," current smokers, or former smokers. Current smokers were further classified as smoking 1 to 4, 5 to 14, 15 to 24, 25 to 34, 35 to 44, or 45 or more cigarettes per day. On the 1976 questionnaire, smokers were asked the age at which they started to smoke. In our analysis, current and former smokers were classified as starting smoking at less than

© 1993 American College of Physicians

Downloaded From: http://annals.org/ by a Washington University User on 12/07/2013

15 years, between 15 and 17 years, 18 and 21 years, 22 and 25 years, or at age 26 years or older. For time since stopping, former smokers were categorized as having stopped for fewer than 2 years, between 2 and 4 years, 5 and 9 years, 10 and 14 years, or for more than 15 years. Alcohol intake may potentially confound the association between cigarette smoking and risk for total mortality because smokers tend to drink more, and alcohol intake is associated with reduced risk for cardiovascular disease mortality (14, 15). In 1980, we assessed the average frequency of alcohol intake during the preceding year. The levels of alcohol intake were categorized into 0 g/d, 0.01 to 4.9 g/d, 5.0 to 14.9 g/d, 15.0 to 24.9 g/d, and 25.0 to 49.9 g/d. The validity of self-reported alcohol intake is high, as reported in detail previously (16, 17). Vigorous physical activity may also potentially confound the association between smoking and total mortality because smokers tend to exercise less, whereas exercise reduces the risk for cardiovascular disease (18). We identified women taking part in regular vigorous exercise by asking the following question in the 1980 questionnaire: "At least once a week, do you engage in any regular activity similar to brisk walking, jogging, bicycling, et cetera, long enough to work up a sweat?" The validity of this item as a measure of physical activity has been discussed previously (19). End Points The end points in our study comprised deaths from all causes occurring after the 1976 questionnaire was returned but before 1 June 1988. The deaths were further grouped into four broad categories: total cardiovascular diseases (ICD 8th revision codes 410 to 440 and 795); total cancers (ICD 8th revision codes 140 to 207); total cancers excluding lung cancer (ICD 8th revision codes 140 to 161 and 163 to 207); and external causes of injury (all ICD "E" codes), which included accidents and suicides. The mortality surveillance included systematic searches of the vital records of the states and the National Death Index to discover deaths among women who did not respond during each questionnaire cycle. This search was supplemented by reports from relatives and postal authorities. We estimate that more than 98% of the deaths in the cohort were ascertained by these methods (20). Physicians reviewed death certificates to classify individual causes of death. Deaths caused by cancer, cardiovascular disease, and external injury were classified as confirmed if they were listed as the underlying causes on the death certificate. Statistical Analysis The primary analysis of follow-up data from 1976 to 1988 used incidence rates with person-years of follow-up as the denominator. For each participant, person-years were allocated according to the 1976 exposure variables and were updated according to information on biennial follow-up questionnaires. For women who died, person-years were assigned according to the covariate status reported in the most recently completed questionnaire; follow-up was terminated at the date of death. Relative risks were calculated as the rate of death in each smoking category divided by the corresponding rate in the reference category. For analyses evaluating the relative risk for mortality among current smokers, we used the never smokers as the reference category. To assess the effect on mortality risk among former smokers by time since stopping, we followed the suggestion of the 1990 Surgeon General's report (7) and used current smokers as the reference category. All relative risks were age-adjusted by 5-year intervals, and 95% CIs were calculated (21). The attributable risk for mortality (that is, the excess number of deaths attributable to smoking per 100 000 person-years) was calculated as the difference between the mortality rates between current (or former) smokers and never smokers. When appropriate, we performed the Mantel test for linear trend across categories of smoking variables and reported the two-tailed P values (22). We also used proportional hazards models to control simultaneously for age, cigarette smoking, and other risk factors for cardiovascular disease and cancer.

Previous reports have indicated that former smokers tend to have smoked fewer cigarettes per day (7) and to have started smoking at an older age than continuing smokers (23). Thus, at any age of quitting, former smokers have less cumulative exposure to cigarettes, on average, than continuing smokers. Failure to adjust for differences in cumulative exposure between former and current smokers may therefore exaggerate the benefits of cessation. Wherever appropriate, we adjusted the relative risks for death among former smokers by the daily number of cigarettes smoked and the age at which they started. We excluded from analysis all women who had reported angina, myocardial infarction, stroke, and cancer (other than nonmelanoma skin cancer) at baseline. These exclusions left a total cohort of 117 001 women available for follow-up. Confounding by Intermediate Variables In a prospective study involving repeated measurement of smoking status, nonfatal disease may act simultaneously as a confounding factor and an intermediate variable in the pathway between smoking and mortality (24, 25). For example, in analyzing the relation between smoking and coronary heart disease mortality, intervening morbid events (such as nonfatal myocardial infarction or angina) may act both as a determinant of subsequent exposure to smoking and as an independent risk factor for subsequent death from coronary heart disease. We assessed the extent of such confounding by performing the G-computational algorithm, as described by Robins (24, 25), on the relation between cigarette smoking and fatal coronary heart disease, using nonfatal myocardial infarction and angina in the analysis as the time-dependent covariates. Results Characteristics of Smokers In 1976, 33.2% of the cohort members were current smokers; 43.3%, never smokers; and 23.5%, former smokers. By 1988, the corresponding proportions were 22.1%, current; 42.6%, never; and 35.3%, former smokers. Some, but not all, health-related habits were distributed similarly between never, current, and former smokers (Table 1). Compared with never smokers, current smokers tended to engage less frequently in vigorous exercise, to have slightly lower body mass index, and to be more likely to drink more than 15 g/d of alcohol. Causes of Death During the 12 years of observation, 2847 deaths occurred in 1 374 556 person-years: Five hundred sixty-six persons died of cardiovascular diseases, 247 of lung cancer, 1209 of cancers excluding lung cancer; 261 from external causes of injury, and 564 of other causes. The category of external causes of injury covers the 8th revision International Classification of Diseases codes E800 to E999 and includes all deaths from accident, poisoning, suicide, and other trauma (26). Risks of Current Smokers Total Mortality Compared with women who had never smoked, current smokers had higher rates of total mortality (ageadjusted relative risk, 1.86; CI, 1.71 to 2.03) (Table 2). The age-adjusted relative risk increased with the number of cigarettes smoked per day (P < 0.0001). Women who smoked 35 to 44 cigarettes per day had a relative

15 November 1993 • Annals of Internal Medicine • Volume 119 • Number 10


993

Table 1. Distribution of Various Potential Self-reported Risk Factors According to Smoking Status among 117 001 Nurses9 Health Study Participants* Variable Never Smoked

Smoking Status Current Smoker

Former Smoker

43.1 591 634 0 24.2 16.3 2.2 5.0 13.1 18.3 43.1 7.5

30.1 339 618 33.8 23.3 14.9 2.0 5.5 15.1 20.2 37.9 21.2

26.3 375 844 17.1 24.0 16.9 2.3 5.8 14.5 19.6 47.3 16.7

Participants in 1980, % Total person-years of follow-up (1976-1988), v Pack-years smoked, v Body mass index, kg/m2 Hypertension, % Diabetes, % High cholesterol, % Parental myocardial infarction before age 60 years, % Postmenopausal hormone use, % Vigorous exercise at least once per week, % Alcohol intake > 15 g/d, %

* Percentage prevalence of risk factors have been directly age-standardized by 5-year age categories to the distribution of ages in the whole cohort in 1980.

risk for total mortality of 2.42 (CI, 2.01 to 2.90), whereas women who smoked 45 or more cigarettes per day had a relative risk of 3.57 (CI, 2.52 to 5.04). The attributable risk for total mortality in current smokers compared with never smokers was 120 per 100 000 person-years. Approximately 37% of all deaths among current smokers in this cohort were attributable to cigarette smoking. Cause-specific Mortality The age-adjusted relative risk for cardiovascular disease mortality among current smokers was 3.47 (CI,

2.85 to 4.22), and the risk increased with the number of cigarettes smoked daily (P = 0.002) (Table 2). Women who smoked only one to four cigarettes per day had an elevated risk for death from cardiovascular disease (relative risk, 2.13; CI, 1.20 to 3.79), and those who smoked 45 or more cigarettes per day had a relative risk of 6.35 (CI, 3.26 to 12.34). For total cancer mortality including lung cancer, the overall age-adjusted relative risk among current smokers was 1.51 (CI, 1.33 to 1.70), and risk increased with the daily number of cigarettes smoked (P < 0.001).

Table 2. Total and Cause-specific Mortality by Daily Number of Cigarettes Smoked by Current Smokers* Event

Cigarettes Smoked per Day by Current Smokers* 25 to 34 15 to 24 >35 1 to 14

"Never Smoker"

Former Smoker

Current Smoker

Total mortality Cases, n Relative riskt Relative riskt

933 1.00 1.00

799 1.28 1.29 (1.14-1.46)

1115 1.86 1.87 (1.65-2.13)

234 1.41 1.51 (1.26-1.81)

480 1.99 2.02 (1.74-2.35)

215 2.06 2.09 (1.71-2.55)

153 2.57 2.63 (2.12-3.27)

Total cardiovascular diseases Cases, n Relative riskt Relative riskt

131 1.00 1.00

151 1.69 1.57 (1.17-2.12)

284 3.47 3.74 (2.86-4.89)

56 2.48 2.69 (1.82-3.97)

124 3.78 4.25 (3.14-5.77)

57 4.11 4.28 (2.86-6.41)

37 4.73 5.64 (3.52-9.04)

Total cancer (including lung cancer) Cases, n Relative riskt Relative riskt

516 1.00 1.00

438 1.26 1.28 (1.08-1.52)

502 1.51 1.42 (1.20-1.67)

95 1.03 1.05 (0.82-1.35)

233 1.73 1.68 (1.29-2.17)

93 1.60 1.62 (1.26-2.08)

67 1.99 1.95 (1.40-2.70)

Total cancer (excluding lung cancer) Cases, n Relative riskt Relative riskt

492 1.00 1.00

366 1.11 1.11 (0.91-1.34)

351 1.10 1.19 (0.99-1.44)

83 0.94 1.02 (0.77-1.36)

161 1.25 1.23 (0.97-1.56)

61 1.10 1.09 (0.78-1.53)

36 1.12 1.15 (0.77-1.74)

External causes of injury Cases, n Relative riskt Relative riskt

90 1.00 1.00

68 1.16 1.26 (0.83-1.89)

103 1.69 1.73 (1.25-2.41)

27 1.60 1.83 (1.07-3.13)

28 1.12 1.25 (0.73-2.16)

20 1.83 1.95 (1.03-3.69)

24 3.92 4.19 (2.33-7.55)

* Cigarettes smoked per day were unknown for 33 cases, including 10 cardiovascular disease deaths, 4 lung cancer deaths, 10 cancers other than lung cancer, 4 accidents or suicide, and 5 deaths from all other causes. t Age-adjusted relative risk. Ninety-five percent confidence intervals are shown in parentheses. t Multivariate relative risk, adjusted for age in 5-year intervals, follow-up period (1976 to 1978, 1978 to 1980, 1980 to 1982, 1982 to 1984, 1984 to 1986, 1986 to 1988), body mass index, history of hypertension, high cholesterol levels, diabetes, parental history of myocardial infarction before age 60 years, postmenopausal estrogen therapy, menopausal status, previous use of oral contraceptives, and age at starting smoking.

994



When lung cancers were excluded, however, the relative risk for cancer mortality among current smokers was not statistically significantly elevated (relative risk, 1.10; CI, 0.96 to 1.26), and no dose-response relationship with the number of cigarettes smoked daily was found (P > 0.2) (see Table 2). This finding is explained by the fact that breast and colorectal cancer, two of the most common cancers in this cohort, had no association with cigarette smoking. The relative risk for deaths caused by breast cancer among current smokers was 0.87 (CI, 0.68 to 1.12) and it was 1.01 for colorectal cancer (CI, 0.69 to 1.46). On the other hand, the weak association between current smoking and total cancer after excluding lung cancer did not preclude the presence of strong associations between smoking and mortality from the less common cancers of individual sites, such as cancers of the buccal cavity and pharynx (six cases among current smokers; age-adjusted relative risk, 5.0; CI, 1.2 to 20.7), esophagus (seven cases among current smokers; age-adjusted relative risk, 11.1; CI, 2.1 to 58.9), and pancreas (26 cases among current smokers; age-adjusted relative risk, 1.9; CI, 1.1 to 3.4). The age-adjusted relative risk for death from external causes of injury among current smokers was 1.69 (CI, 1.28 to 2.24), and risk increased with the daily number of cigarettes smoked (P = 0.001). Multivariate Models The associations of cigarette smoking with total and cause-specific mortality changed only slightly after we controlled for potential confounders (including history of hypertension, diabetes, high serum cholesterol, relative body weight, parental history of myocardial infarction before the age of 60 years, past use of oral contraceptives, postmenopausal estrogen therapy, and age at starting smoking) in multivariate proportional-hazards models (see Table 2).

who had quit for 10 to 14 years. This finding was compatible with chance, however, because the 95% confidence intervals of the relative risks beyond 10 years after cessation excluded the point estimates of the risks among former smokers who had quit for shorter durations, but we cannot exclude the possibility that a small excess in risk extends beyond 15 years of abstinence. Cause-specific Mortality A 24% reduction in the risk for mortality from cardiovascular disease was apparent within 2 years of giving up cigarettes; however, the excess risk did not approach the level of never smokers until 10 to 14 years after cessation. For deaths from external causes of injury, the age-adjusted relative risks among former smokers decreased to the level of never smokers 5 to 9 years after cessation (Table 3). A greater risk for total cancer was observed among former smokers compared with current smokers in the first 2 years after cessation (age-adjusted relative risk, 1.42; CI, 1.12 to 1.81). The excess risk decreased to the level of never smokers after 10 to 14 years of abstinence. However, when lung cancer was excluded from the analysis of total cancer mortality, an excess risk was apparent only within the first 2 years after quitting (age-adjusted relative risk, 1.35; CI, 1.00 to 1.81) (Table 3). The observations regarding the relationship of time since quitting with the risks of total and cause-specific mortality remained unchanged after controlling for potential confounding factors, including history of hypertension, diabetes, high serum cholesterol, relative body weight, parental history of myocardial infarction before age 60 years, past use of oral contraceptives, postmeno-

Decrease in Risk among Former Smokers Total Mortality Compared with never smokers, the age-adjusted relative risk for total mortality among former smokers was 1.28 (CI, 1.16 to 1.40). When we examined the relation between time since quitting and the relative risk for total mortality, the risk among former smokers approached the level of never smokers 10 to 14 years after cessation (Table 3; Figure 1). In terms of absolute risk, if 1000 women ages 40 to 44 years quit smoking today, approximately 17 deaths would occur in that group during the next 14 years, compared with 30 deaths if they had continued smoking; that is, a net saving of 13 lives per 1000 women during a 14-year period. The absolute benefits of cessation increase with age. If 1000 women 60 years or older quit smoking today, approximately 103 deaths would occur during a period of 14 years, compared with 138 deaths if they had continued smoking; that is, a net saving of 35 lives per 1000 women. A slight and nonsignificant increase in relative risk was apparent in the category of former smokers who had quit for more than 15 years compared with those

Figure 1. Risk of total mortality by time since quitting. Multivariate relative risk for total mortality by time since quitting (reference category: current smokers). Error bars represent 95% confidence intervals. Nonfatal coronary heart disease, stroke, and cancer (except nonmelanoma skin cancer) were excluded at baseline and at the beginning of each successive 2-year follow-up period. Variables in model include age in 5-year categories, follow-up period (1976 to 1978, 1978 to 1980. . .1986 to 1988), body mass index, history of hypertension, high cholesterol, diabetes, parental history of myocardial infarction before age 60 years, postmenopausal estrogen therapy, menopausal status, previous use of oral contraceptives, age at starting smoking, and daily number of cigarettes smoked during the period before cessation.



995

pausal estrogen therapy, the daily number of cigarettes smoked, and age at starting smoking (Table 3). Age at Starting To Smoke Current Smokers We examined the relation between age at starting smoking and risk for total and cause-specific mortality among current smokers (Table 4). The relative risk for total mortality among current smokers who started smoking before age 15 years was 2.80 (CI, 2.05 to 3.82), whereas for those who started after age 26 years, the relative risk was 1.59 (CI, 1.28 to 1.97) (P = 0.01). Starting to smoke before the age of 15 years was associated with the highest risks for death from cardiovascular disease (age-adjusted relative risk, 8.72; CI, 5.58 to 13.65), and external causes of injury (age-adjusted relative risk, 3.22; CI, 1.37 to 7.55). Adjusting for multiple risk factors, including the daily number of cigarettes smoked, did not substantially affect these observations (Table 4). Former Smokers Among former smokers the relation between age at starting smoking and the risks for total and cause-specific mortality were not statistically significant. These observations remained unchanged after adjusting for multiple risk factors, including the number of cigarettes smoked per day and the time since quitting.

The "III Quitter" Effect Previous studies have suggested that recent quitters include a disproportionate number of those who have quit because they are ill (7). The effect of this is to cause a spuriously elevated risk for death among former smokers during the early years after cessation. To address this problem, we did analyses in which women who reported nonfatal cardiovascular disease (myocardial infarction, angina, stroke) or cancer (excluding nonmelanoma skin cancer) at each biennial questionnaire were excluded from further follow-up. These exclusions left a total of 1980 deaths for analysis (Appendix Tables 1 and 2). Among current smokers, the strength of the association between the daily number of cigarettes smoked and the risk for total mortality increased after excluding diseases at the beginning of each 2-year interval (Appendix Table 1). When we examined the relation of time since quitting with risk for cancer mortality (including and excluding lung cancer), an excess risk was no longer apparent among former smokers within the first 2 years after quitting, suggesting an association between cancer diagnosis and stopping smoking. Confounding by Intermediate Variables As we explained in the Methods section, in a prospective study involving repeated measurement of smoking status, nonfatal disease may act simultaneously

Table 3. Total and Cause-specific Mortality by Time since Quitting* Event

Years since Quitting among Former Smokersf 10 to 14 2 to 4 5 to 9

Current Smoker

"Never Smoker"

15

* Reference category consists of current smokers. t Time since quitting was missing for 138 cases, including 27 cardiovascular disease deaths, 9 lung cancer deaths, 66 deaths from cancers other than lung, 11 accidents or suicide, and 25 deaths from all other causes. $ Age-adjusted relative risk. Ninety-five percent confidence intervals are shown in parentheses. § Multivariate relative risk, adjusted for age in 5-year intervals, follow-up period (1976 to 1978, 1978 to 1980, 1980 to 1982, 1982 to 1984, 1984 to 1986, or 1986 to 1988), body mass index, history of hypertension, diabetes, high cholesterol levels, postmenopausal estrogen therapy, menopausal status, previous use of oral contraceptives, parental history of myocardial infarction before age 60 years, and daily number of cigarettes smoked during the period before stopping smoking.

996



Table 4. Total and Cause-specific Mortality by Age at Starting Smoking among Current Smokers Event

"Never Smoker"

Age at Starting to Smoke among Current Smokers * 22 to 25 18 to 21 26

Total mortality Cases, n Relative riskf Relative risk$

933 1.00 1.00

38 2.80 3.15 (2.16-4.59)

175 1.86 1.85 (1.51-2.28)

657 1.88 1.77 (1.54-2.03)

149 1.95 1.86 (1.50-2.32)

88 1.59 1.45 (1.15-1.92)

Cardiovascular disease Cases, n Relative riskf Relative risk$

131 1.00 1.00

15 8.72 9.94 (5.15-19.19)

41 3.43 3.55 (2.27-5.56)

157 3.32 3.18 (2.36-4.28)

47 4.18 4.06 (2.65-6.21)

24 2.99 2.88 (1.704.87)

Cancer (including lung cancer) Cases, n Relative riskf Relative riskt

516 1.00 1.00

11 1.46 1.64 (0.84-3.20)

77 1.48 1.43 (1.06-1.93)

300 1.54 1.43 (1.18-1.74)

73 1.70 1.69 (1.24-2.29)

38 1.23 1.10 (0.75-1.62)

Cancer (excluding lung cancer) Cases, n Relative riskf Relative risk$

492 1.00 1.00

6 0.84 0.88 (0.37-2.11)

52 1.05 1.07 (0.75-1.52)

201 1.08 1.00 (0.80-1.25)

56 1.36 1.35 (0.96-1.90)

33 1.11 1.00 (0.66-1.51)

External causes of death Cases, n Relative riskf Relative riskf

90 1.00 1.00

5 3.22 5.39 (1.84-15.78)

21 2.00 2.01 (1.11-3.65)

61 1.69 1.69 (1.09-2.61)

8 1.22 1.27 (0.54-2.99)

7 1.39 1.73 (0.75-4.03)

* Age at starting smoking was unknown for 8 cases, including 3 deaths from cancers other than lung, 1 accident or suicide, and 4 deaths from all other causes. f Age-adjusted relative risk (RR). Ninety-five percent confidence interval are shown in parentheses. f Adjusted for age in 5-year intervals, follow-up period (1976 to 1978, 1978 to 1980, 1980 to 1982, 1982 to 1984, 1984 to 1986, or 1986 to 1988) history of hypertension, diabetes, high cholesterol levels, body mass index, previous use of oral contraceptives, postmenopausal estrogen therapy, menopausal status, parental history of myocardial infraction before age 60 years, and daily number of cigarettes smoked.

as a confounding factor and an intermediate variable in the pathway between smoking and death (24, 25). We assessed the extent of such confounding by applying a method previously described (24, 25) to the association between cigarette smoking and fatal coronary heart disease, with nonfatal myocardial infarction as the intermediate variable. When we did this analysis, the risk estimate for coronary heart disease mortality was identical to the crude estimate of risk. Therefore we concluded that confounding by intermediate variables was unlikely to be of major concern in our study. Confounding by Alcohol Intake and Vigorous Exercise To examine the possible confounding effects of alcohol consumption and vigorous exercise on cigarette smoking, we analyzed data from the 1980 to 1988 follow-up interval. Although the analysis was limited to 2356 deaths, the age-adjusted association between daily number of cigarettes smoked and the risk for total mortality remained unchanged. The relative risk for total mortality increased from 1.56 (CI, 1.28 to 1.91) among women smoking 1 to 14 cigarettes per day to 2.53 (CI, 1.95 to 3.24) among women smoking 35 or more cigarettes per day. After adjusting for alcohol consumption and vigorous exercise, the risk for total mortality among former smokers still declined to the level of never smokers 10 to 14 years after cessation. Adjusting for alcohol intake and vigorous exercise resulted in a slight strengthening of the association be-

tween current smoking and cardiovascular disease mortality. For example, the multivariate relative risk for cardiovascular disease mortality increased from 2.69 to 3.08 among women smoking 1 to 14 cigarettes per day and from 4.28 to 4.93 among women smoking 25 to 34 cigarettes per day. However, the excess risk for death from cardiovascular disease among former smokers still took 10 to 14 years to decline to the level of never smokers. The risk for death from external causes of injury among current smokers compared with never smokers declined from 1.69 (CI, 1.28 to 2.24) to 1.54 (CI, 1.10 to 2.17) after adjusting for alcohol intake and vigorous exercise. The excess risk among former smokers approached the level of never smokers 5 to 9 years after quitting.

Discussion Our data suggest that previous studies of the relation between smoking and risk for total mortality in women have underestimated the strength of the association. Studies done among women during the 1950s and 1960s reported relative risks for total mortality ranging from 1.3 to 1.4 (1-6), whereas smokers in the Nurses' Health Study were at nearly 1.9 times the risk compared with never smokers (see Table 2). This increase in the magnitude of relative risk across time is probably caused by the increasing proportion of women in more recent birth



997

cohorts who started smoking at a young age. Data from National Health Interview Surveys indicate that the proportion of women starting to smoke before age 16 years increased from 7.2% among women born between 1910 and 1914 to 20.2% among those born between 1950 and 1954 (27). In the Nurses' Health Study, women who started smoking before age 15 years had the highest risk (multivariate relative risk, 3.15) for total mortality (see Table 4). Further, the observed deaths in this study were premature because they all occurred among women who were younger than 67 years during the 12 years of follow-up. A recent study of smoking-attributable deaths in developed countries estimated that those who die as a consequence of tobacco use between ages 35 to 69 years lose an average of about 23 years of life (28). The time required for the risk for total mortality among quitters to reach the level of never smokers differs across studies. The Nurses' Health Study data indicate that the risk among former smokers decreases to the level of never smokers 10 to 14 years after cessation. This estimate of the time required is somewhat shorter than that of several previous studies. For example, the American Cancer Society Cancer Prevention Study I (ACS CPS-I) found that among former smokers of 20 or more cigarettes per day, the risk for total mortality was still higher than that of never smokers even after 10 years of abstinence (6). In the more recent ACS CPS-II study, involving 4 years of follow-up among 521 555 men and 658 748 women, the risk for total mortality among female former smokers decreased to the level of never smokers 16 or more years after cessation (7); and in the U.S. Veterans Study (10), the overall mortality risk among male smokers remained elevated 15 years or more after cessation (among smokers of 10 to 20 cigarettes per day, relative risk, 1.47; among smokers of 21 to 39 cigarettes per day, relative risk, 1.22). The differences among studies in estimates of the duration needed for a former smoker to have the same overall mortality risk as a never smoker may be caused partly by factors such as the lack of ascertainment of smoking status after enrollment in earlier studies (6, 7, 10). In these studies, persons who smoked at enrollment but subsequently quit remained assigned to the current smoker category. This misclassification tends to obscure the benefits of cessation in comparison with continued smoking (7). In contrast, another study (9), which updated smoking status at yearly intervals, reported that the relative risk for total mortality among women who formerly smoked returned to the level of never smokers 6 to 10 years after cessation. A potential limitation of our study is that our cohort consists of predominantly white middle-aged female nurses. Although our findings might not be generalizable to older women or other ethnic groups, the qualitatively similar effects of smoking across population subgroups defined by age, sex, and race suggest that the biological effects of smoking cessation are also not likely to differ in major ways across demographic groups. Smokers who quit may be unrepresentative in ways that could not be controlled in our analysis, although we adjusted for a broad range of potential con998

founding variables in our multivariate analyses. Finally, during the 12-year follow-up period, current smokers were marginally less likely to respond compared with former smokers. This could have potentially resulted in under-estimation of the benefits of cessation. The difference in response rates, however, never exceeded 0.5 percentage points, making this an unlikely source of major bias. The observation of an excess cancer mortality risk within the first 2 years of giving up smoking in our study (see Table 3) and in previous studies (7) has been attributed to the "ill quitter" effect. When analyses excluded women with cardiovascular disease and cancer at the beginning of each 2-year follow-up interval, the excess risk among recent quitters was removed. The data in Appendix Tables 1 and 2 provide a comparison between the effect of stopping smoking before developing disease and the situation in which a proportion of smokers stop after the onset of disease. The benefits of smoking cessation, in terms of a reduction in all-cause as well as cause-specific mortality, occur sooner in the former case. If a smoker stops smoking before the onset of disease, she will experience a 24% reduction in the risk for total mortality within 2 years of quitting (including a 37% reduction in cardiovascular disease mortality) as well as a rapid decline of the risk for cancer mortality to the level of a never smoker (Appendix Table 2). These benefits are more substantial and occur sooner than in the case of delaying the cessation of smoking until the onset of disease (Table 3). The finding of an association between cigarette smoking and suicide and accidents has been reported in previous studies, including the British Doctors stu^y (8) and the MRFIT study (29). To our knowledge, our study is the first to report the relationships of age at starting smoking and time since quitting with risk for deaths from external causes of injury. Although a recent study dismissed the association as causal because of lack of biological plausibility (29), several studies have shown an association between cigarette smoking and depression (30-33). The Nurses' Health Study collected no data on the mental health of participants before 1992. On the other hand, the association between smoking and external causes of injury persisted after controlling for alcohol intake in multivariate analysis. Whether this association is causal or whether smoking is merely correlated with one or more factors (as yet unidentified) predisposing to accidents and suicide deserves further study. Weight gain after smoking cessation is thought to be a factor contributing to continuing smoking by women (7). In the Nurses' Health Study, women who quit smoking had on average 1.4 to 2.8 kg greater weight gain during an 8-year follow-up period compared with continuing smokers (34). Our analyses of total mortality, which balance any adverse as well as desirable effects of smoking cessation, clearly indicate that the health benefits of smoking cessation far exceed the risks posed by this amount of weight gain. The best health advice remains not to start smoking at all, particularly at a young age. Benefits of smoking cessation are substantial, however, and begin to accrue almost immediately after quitting.



Appendix Table 1. Total and Cause-specific Mortality by Daily Number of Cigarettes Smoked: Comparison of Analyses with and without 2-Year Exclusion of Disease at the Start of Each Period Event

Cigarettes Smoked per Day by Current Smokers *35 25 to 34 ltol4 15 to 24

"Never Smoker"

Former Smoker

Total mortality Cases,* n Relative risk Cases,t n Relative risk

933 1.00 632 1.00

799 1.29 410 1.15 (1.01-1.29)

234 1.51 176 1.56 (1.26-1.94)

480 2.02 381 2.17 (1.82-2.59)

215 2.09 175 2.23 (1.77-2.80)

153 2.63 130 3.16 (2.49-4.02)

Cardiovascular disease Cases,* n Relative risk Cases,t n Relative risk

131 1.00 111 1.00

151 1.57 106 1.48 (1.13-1.94)

56 2.69 47 2.48 (1.61-3.82)

124 4.25 114 3.99 (2.84-5.59)

57 4.28 52 4.35 (2.83-6.69)

37 5.64 33 5.82 (3.50-9.69)

Total cancer (including lung cancer) Cases,* n Relative risk Cases,t n Relative risk

516 1.00 262 1.00

438 1.28 172 1.03 (0.85-1.26)

95 1.05 51 1.11 (0.76-1.62)

233 1.68 168 2.13 (1.62-2.81)

93 1.62 64 1.82 (1.24-2.65)

67 1.95 50 2.96 (2.03-4.34)

Total cancer (excluding lung cancer) Cases,* n Relative risk Cases,t n Relative risk

492 1.00 244 1.00

366 1.11 140 0.90 (0.73-1.11)

83 1.02 43 1.03 (0.68-1.55)

161 1.23 100 1.44 (1.05-1.99)

61 1.09 33 0.95 (0.58-1.57)

36 1.15 22 1.41 (0.82-2.40)

* Cases and multivariate relative risks after baseline exclusion of coronary heart disease, stroke, and cancer except nonmelanoma skin cancer. Multivariate relative risks were adjusted for age in 5-year intervals, follow-up period (1976 to 1978, 1978 to 1980, 1980 to 1982, 1982 to 1984, 1984 to 1986, 1986 to 1988), body mass index, history of hypertension, high cholesterol, diabetes, parental history of myocardial infarction before age 60 years, postmenopausal estrogen therapy, menopausal status, previous use of oral contraceptives, and age at starting smoking. t Cases and multivariate relative risks after exclusion of coronary heart disease, stroke, and cancer (except nonmelanoma skin cancer) at the beginning of each 2-year follow-up interval. Ninety-five percent confidence intervals are shown in parentheses.

Appendix Table 2. Total and Cause-specific Mortality by Time since Quitting: Comparison of Analyses with and without 2-Year Exclusion of Diseases at the Start of Each Follow-up Period Event

Yean > since Quitting among Former Smokers 10 to 14 2 to 4 5 to 9

Current Smoker

"Never Smoker"