Effect of Smoking Cessation on Mortality After Myocardial Infarction:  Meta-analysis of Cohort Studies FREE

SMOKING IS the leading preventable cause of premature death in the United States and is responsible for more than 400,000 deaths annually.¹ Many of these smoking-related deaths occur in patients with cardiac disease. Individuals with ischemic heart disease who smoke are at particular risk for increased mortality due to the adverse effects of cigarettes on coronary blood flow, myocardial oxygen demand, and risk of thrombosis.^2- 7 These effects seem to be reversible, and significant reductions in mortality have been observed in patients with established cardiovascular disease who quit smoking.⁸

Rigorous randomized trials^9- 11 have focused on the effect of smoking cessation programs on cessation rates in patients after myocardial infarction (MI) and after coronary bypass surgery. One trial¹² of nicotine replacement therapy vs placebo in patients with stable coronary artery disease examined the effect on cardiac events (death, MI, cardiac arrest, hospital admission for angina, arrhythmia, or congestive heart failure) and found no significant difference.

A randomized trial to specifically examine the question of whether individuals who quit smoking after MI have a lower mortality rate than those who continue smoking would require allocating patients to a continuing smoking arm. Because this is not practical, investigators have instead relied on cohort studies to obtain evidence on this issue. These cohort studies^13- 26 have consistently demonstrated improvements in mortality associated with smoking cessation, although the benefits have been variable. We performed a meta-analysis of these cohort studies to determine a more precise estimate of the mortality benefit associated with smoking cessation after MI. We also examined whether this effect was present in several subgroups.

We conducted a search of English- and non–English-language articles in the MEDLINE database between January 1966 and September 1996 using the following strategy. Exp smoking or exp smoking cessation was cross-referenced with MI and the following prognosis terms: incidence or exp mortality or mortality(sh [subject heading]) or exp follow-up studies or prognos:(tw [text word]) or predict(tw) or cohort studies or course:(tw) or prognosis or natural(tw) and history(tw). We also performed a similar search of the EMBASE database. We chose keywords for prognosis studies that have been demonstrated to maximize sensitivity for detecting sound clinical studies in this area.²⁷ In addition, references from review articles were searched and letters were sent to authors of selected studies requesting information about other relevant unpublished material.

Two of us (K.W. and N.G.) independently evaluated the MEDLINE and EMBASE abstracts of each article and the retrieved articles (foreign-language articles were reviewed by only one individual). Articles selected for the meta-analysis had to meet the following criteria: (1) study of patients after MI, (2) determination of smoking status at the time of MI and any time thereafter, (3) at least 1 year of follow-up, (4) reporting of mortality rates in those who quit and those who continued smoking, and (5) enrollment of at least 100 patients. For selection of MEDLINE and EMBASE abstracts, both reviewers were masked to the author, journal, and year of publication. All disagreement was resolved by consensus, and κ statistics were used to evaluate chance-corrected agreement.

We examined the articles selected for review to determine their quality. Articles were assigned scores using the following criteria (fully meets criteria, partially meets criteria, does not meet criteria): (1) Was a representative, well-defined sample of patients identified (2, 1, 0)? (2) Were patients enrolled consecutively (2, 1, 0)? (3) Were there clear definitions of MI (1, 0.5, 1) and smoking status at time of enrollment (1, 0.5, 1)? (4) Was there an objective, valid assessment of smoking status at follow-up (biological verification) (2, 1, 0)? (5) Was follow-up complete (≥90%) (2, 1, 0)? (6) Was there adjustment for important prognostic factors (2, 1, 0)? The maximum score for an article is 10.

In duplicate, we independently extracted data from the relevant articles. Data regarding mortality in smokers and nonsmokers were abstracted as the primary outcome. Further information was obtained regarding location and year of trial, sex and age of the study population, time of assessment of smoking status, duration of follow-up, and effect of age and infarct size on mortality.

Before combining the results, we performed a test for homogeneity. We compared mortality rates in continuing smokers and the cessation group using odds ratios. We calculated odds ratios for the individual studies and a combined odds ratio using a DerSimonion and Laird random effects model.²⁸ We also ordered the studies according to quality score to examine if quality was associated with magnitude of effect.

We examined the mortality benefit associated with smoking cessation in the following prespecified subgroups: sex, study period (enrollment before or after 1980), duration of follow-up (10 years or <10 years), and country of study. These analyses were conducted to determine whether the benefit was present in different populations (sex) or was affected by temporal or geographical differences in management or differences in medical therapies (study period and country of study) and whether the effect persisted over time (duration of follow-up). We performed a study-level regression analysis to determine whether sex, duration of follow-up, study period, or quality of study affected the effect size.²⁸ We also examined the distribution of certain prognostic factors between continuing smokers and those who quit.

The MEDLINE and EMBASE search strategy yielded 605 citations that included 13 potentially relevant articles.^{13- 18,20- 24,26,29} One additional article¹⁹ was obtained from a reference of a review. Of these 14 articles, 2 were excluded: an early follow-up of a previously identified article²⁶ and a duplicate report.²⁹ Kappa scores for agreement on abstracts and final articles were 0.79 and 0.65, respectively, suggesting a substantial level of agreement.³⁰

Twelve studies involving 5878 patients were included in the final analysis (Table 1). The studies were conducted in 6 countries: Ireland, United States, Finland, Sweden, United Kingdom, and Norway. The earliest study¹⁴ enrolled patients beginning in 1949, and the latest study²⁴ accrued patients from 1986 to 1988. Four studies examined men only,^13,15,17,20 1 examined women only,¹⁹ and 7 examined a mixed population.^{14,16,18,21- 24} Time of assessment of smoking status ranged from 1 month to 1 year after MI, with 2 studies having ongoing assessment of smoking status.^14,18 Duration of follow-up ranged from 2 to 10 years. Smoking cessation rates in these studies ranged from 29% to 74%.

A lower mortality rate was associated with smoking cessation in all studies (Table 2). Odds ratios ranged from 0.29 to 0.84, and the combined odds ratio was 0.54 (95% confidence interval [CI], 0.46-0.62). Four studies had 95% CIs that included 1.00. The Zelen test for homogeneity was not significant (P=.61), suggesting that the differences in the results of the individual studies were compatible with chance. The estimate of between-study variance was negative. Consequently, the combined odds ratio and 95% CI using a random effects model were identical to those of a fixed effects model (odds ratio, 0.54; 95% CI, 0.46-0.62) (Figure 1).

Mortality rates in the cessation group ranged from 4% to 37% and in continuing smokers from 8% to 54%. Relative risk reductions ranged from 15% to 61%, with 10 of 12 studies having values greater than 30%. Absolute risk reductions ranged from 1.2% to 27.5%. Based on an odds ratio of 0.54 and an estimated mortality rate of 20% in continuing smokers, the number needed to quit smoking to save one life is approximately 13.³¹ Mean duration of follow-up adjusted for study size was 4.8 years. The number needed to treat compares favorably with those for other therapeutic interventions.³²

Quality scores for the primary studies ranged from 6.5 to 8 (Table 3).

Three of the included studies^16,20,22 were subgroups of randomized controlled trials. All of the studies had a well-defined sample of patients. Five studies^{13,15,17,19- 20} restricted patients by sex; most had age restrictions (usually enrolling patients <65 years old). All studies^13- 24 either attempted to enroll consecutive patients or had a systematic method for inclusion (eg, including every other male <65 years old). One study²⁴ relied on response to a questionnaire for smoking cessation data and had a 45% response rate.

The definition of an MI was consistent across studies, requiring some combination of chest pain, electrocardiographic changes, and increase in cardiac enzyme levels. The definition of smoking status was more variable; however, most studies required smoking at least 1 cigarette per day. Smoking status depended on self-report in all but one study,¹⁷ which biochemically verified smoking status on a subgroup of patients. Three studies^17- 19 described some form of smoking cessation counseling.

Nine of 12 studies^{14- 15,17- 21,23- 24} examined the differences in age and infarct size in continuing smokers vs those who quit. In all 10 studies^{13- 14,16- 21,23- 24} in which information on follow-up could be obtained, follow-up was greater than 90%.

The mortality benefit was noted in all of the subgroups (Table 4). The odds ratio for mortality in women was 0.36 (95% CI, 0.23-0.54) and in men was 0.52 (95% CI, 0.45-0.58). However, several studies^{14,16,22- 24} did not present data separately for men and women. Eight studies enrolled patients before 1980^13- 19,21 and 4 enrolled patients after 1980.^20,22- 24 The respective odds ratios (95% CIs) were 0.55 (0.45-0.66) and 0.52 (0.41-0.65). Two studies reported extended follow-up of 10 years.^17,21 The combined odds ratio for these studies was 0.54 (95% CI, 0.41-0.70). The odds ratio for the shorter follow-up studies was 0.53 (95% CI, 0.45-0.64). Odds ratios for country of study ranged from 0.49 (United States, 2 studies with 897 patients)^14,22 to 0.84 (Norway, 1 study with 918 patients).¹⁶

The study-level regression analysis illustrated that none of the following factors affected the combined odds ratio: sex (P=.16), year of study (P=.85), duration of follow-up (P=.87), and quality of study (P=.92).

Nine studies^{14- 15,17- 21,23- 24} considered the effect of age on mortality rates. Only one of these studies²³ noted that continuing smokers were significantly older. In the study by Greenwood et al,²⁴ adjustment for age increased the observed mortality benefit associated with smoking cessation (reduction in the odds ratio from 0.71 to 0.58). Four studies^15,17,19,23 considered the effect of infarct size. These studies primarily used peak enzyme rise or evidence of congestive heart failure to evaluate the size of the infarct. No study relied on left ventricular ejection fraction to quantitate infarct size. Of these 4 studies, 3 of them^15,17,19 suggested that the patients who quit smoking had larger infarcts, and the other study²³ showed no differences between the groups (Table 5).

Several studies also reported the distribution of cardiac risk factors between those who continued smoking and those who quit. Two studies^14,17 noted that ex-smokers were more likely to be diabetic than were those who continued smoking, and one study¹⁵ observed no difference. Three studies^14,17- 18 observed that continuing smokers had lower mean blood pressure, one²³ observed that they had a higher mean blood pressure, and another²¹ showed no difference. One study¹⁴ found that continuing smokers had lower mean cholesterol levels. Two studies^14,18 found that continuing smokers had lower mean weights. In no instance were the differences statistically significant.

The mortality benefit of smoking cessation on mortality after MI observed in our meta-analysis is of moderate to large magnitude and is consistent across study location, patient sex, time periods, and different durations of follow-up. The results are consistent with the mortality benefits observed in smoking cessation studies^8,33- 34 of patients after coronary artery bypass surgery, after angioplasty, and with stable coronary artery disease. The odds ratio of 0.54 in this study compares favorably with odds ratios from meta-analyses of other MI interventions. In comparison, the odds ratio for reduction in mortality after MI for thrombolytic therapies is 0.75 (95% CI, 0.71-0.79), for aspirin after MI is 0.77 (95% CI, 0.70-0.84), and for β-blockers after MI is 0.88 (95% CI, 0.80-0.98).³⁵ These odds ratios are from meta-analyses of randomized controlled trials, which generally have higher validity than do cohort studies in determining treatment effects and tend to show smaller effect sizes.^36- 37

The degree of heterogeneity in this meta-analysis was surprisingly low. This might be explained by the fact that the primary studies were all of approximately the same validity and used consistent methods. The lack of heterogeneity adds further support to the observed mortality benefit associated with smoking cessation after MI. The results of the study-level regression analysis were not surprising as there was no heterogeneity to explain.

Caution is appropriate in interpreting the results of this meta-analysis because the results are based on data from cohort studies. Lack of randomization in these studies increases the likelihood of there being unequal distribution of important prognostic variables. We examined the effect of age and infarct size as potential confounding variables in the primary studies and found that in studies that took these factors into account, they did not seem to bias the results. Mechanisms for evaluating infarct size were not optimal, however. We cannot exclude the possibility that other significant variables might be unequally distributed across smoking and nonsmoking groups, biasing the results of this meta-analysis. Individuals who quit smoking might have other changes in behavior that could provide them with a survival advantage. It is also possible that these individuals received more effective MI management than did those who continued smoking. The latter issue is made less likely by the observation that the mortality benefit was present before 1980 when use of most modern post-MI therapies was not routine. Former smokers might also have smoked fewer cigarettes, on average, before the infarct than continuing smokers, conferring on them a survival advantage.

A major methodological concern was the accuracy of self-report of smoking status because only one study¹⁷ used biological confirmation. Misrepresentation of smoking status rates based on self-report have been found to be as high as 26% after MI.³⁸ For this potential bias to inflate the effect size, the assumption would have to be made that those who misrepresented themselves as nonsmokers have a better chance of surviving. Only 2 of 12 studies^14,18 had ongoing assessment of smoking status. This could potentially bias the results against the benefit of smoking cessation if individuals changed their smoking status after the time of assessment. Previous reviews of smoking cessation after MI have found that although most relapses occur within the first 3 months, relapses do occur after 1 year.³⁹

Publication bias also could potentially affect the results of this meta-analysis. Randomized trials of MI therapies, in particular, may have examined smoking subgroups and chosen not to publish the subgroup results if no important interaction was noted. We attempted to identify any relevant unpublished material by contacting authors in the field. Two published abstracts^40- 41 that were not associated with full manuscripts were found, and both reported large reductions in mortality associated with cessation of smoking that were consistent with the results of our meta-analysis. The large fail-safe N of 241 (the number of null result studies that would need to be added to the meta-analysis to produce a nonsignificant result) makes it unlikely that there would be enough unpublished data to counter the effect found in this meta-analysis.⁴²

This meta-analysis provides compelling evidence for the benefit of smoking cessation on mortality in patients after MI. Given the magnitude of this mortality benefit, even modest reductions in smoking rates in this patient population might translate into a significant decrease in the number of deaths. The results of this meta-analysis and other studies^11,43- 44 suggest that the development of cardiac disease itself is a strong stimulus to induce smoking cessation. Attempts have been made to increase cessation rates using smoking cessation programs. However, randomized trials^9- 11 examining the benefits of these programs in patients with cardiovascular disease have yielded mixed results. Use of nicotine replacement therapy in patients with stable cardiac disease has also been studied^12,45 and, although seeming not to have any adverse effects, has not been demonstrated to induce long-term cessation. More randomized trials of smoking cessation interventions in patients with cardiovascular disease are warranted. Ideally, these trials will have extended follow-up and examine effects on cardiovascular events, mortality, and cessation rates.

Accepted for publication June 29, 1999.

We thank John Attia, Roman Jaeschke, and Phillip Tschopp for their assistance with translation of foreign-language articles.

Reprints: Kumanan Wilson, MD, MSc, FRCPC, Civic Parkdale Clinic, 737 Parkdale Ave, Suite 414, Ottawa, Ontario, Canada K1Y 1J8 (e-mail: kwilson@lri.ca).