Smoking and the risk of first acute myocardial infarction When analyzing risk factors for first acute myocardial infarctlon in the Copenhagen City Heart Study, a large prospective population study of 20,DW men and women, smoking was found to influence risk significantly in a dose-dependent manner, the risk increaslng 2% to 3% for each gram of tobacco smoked daily. Risk was particularly associated wlth inhalation, the risk for inhalers being almost twice that of noninhalers. No difference in risk could be demonstrated between various types of tobacco (pipe, cigar/cheroots, or plain and filtered cigarettes). The rlsk seemed assoctated with current smoking only, inasmuch as the duration of the smoklng habit was not important. Ex-smokers had the same risk as those who had never smoked regardless of duration of smoking and time elapsed since quitting. Relative excess risk was significantly higher in female smokers than in male smokers, and dally alcohol intake appeared to have some protective effect on the risk of first acute myocardial infarction among heavy smokers. (AM HEART J 1991;122:438.)
J#rgen Nyboe, MSc, Gorm Jensen, MD, PhD, Merete Appleyard, Peter Schnohr, MD. Copenhagen, Denmark
In numerous epidemiologic studies smoking has consistently been identified as a major risk factor for ischemic heart disease.1-4 The pathophysiologic mechanisms responsible for the adverse effect of smoking are by no means clear. The risk seems to be dose dependent with regard to the amount of tobacco smoked.4 Most studies are concerned with cigarette smoking only, but the few dealing with other types of tobacco suggest that pipe and cigar smoking may be equally dangerous. 4-6 Some studies have shown that the risk decreases after cessation of smoking, especially in subjects with prior manifestations of ischemic heart disease,73 8 but the magnitude and rate of this decrease have not yet been determined.g In this article we present data concerning the prevalence of smoking in a sample of the population of Copenhagen and the role of tobacco smoking as a risk factor for first acute myocardial infarction (AMI). Among current smokers we have examined the risk associated with the amount and type of tobacco smoked including filtered versus plain cigarettes, duration of the smoking habit, and inhalation. Among From the Copenhagen City Heart Study, Rigshospitalet. Supported by the Danish Heart Foundation and by grants from Laegeforeningens Forskningsfond, Lily Benthine Lunds Fond, Hafnia-Haand i Haand Fond&, and Fabrikant Mads Clausens Fond. Received for publication Sept. 18, 1990; accepted Jan. 23, 1991. Reprint requests: G. Jensen, the Copenhagen City Heart Study, Department 7112, Rigshospitalet, Tagensvej 20, DK-2200 Copenhagen, Denmark. 4/l/29665
438
RHLT,
and
current nonsmokers we have examined whether cessation of smoking has reduced the risk of AMI. Furthermore, we have studied whether the effect of smoking on the risk of AM1 differs between men and women and whether it is influenced by daily alcohol consumption. METHODS
Subjects.The Copenhagen City Heart Study is a prospective cardiovascular population study comprising almost 20,000 men and women, aged 20 years or more, selected at random after age stratification among 90,000 persons living in a defined area around Rigshospitalet, University Hospital of Copenhagen. The persons selected were invited by letter to a health examination at Rigshospitalet on a specific date between March 1,1976 and March 31,1978, to collect basic data concerning possible risk factors for each person. A detailed description of the study procedures has previously been published.lO* l1 Information about new cases of AM1 was obtained from a followup examination between 1981 and 83 and through scrutiny of the National Health Service registers of death and hospitalization to the end of 1983, that is, a follow-up period of approximately 6.5 years. All caseswere verified by means of hospital records and death certificates. Analysis of risk factors is limited to persons aged 30 years or more and comprises 12,196 persons without prior AM1 and 360 cases of first AM1 (Table I).12vl3 Methodology.We have used a regression technique to describe the risk of AM1 during the follow-up period as a function of tobacco smoking as assessedby questionnaire at the initial examination. When the aim is a causative in-
Volume 122 Number 2
terpretation of the relationship, it is desirable to eliminate as far as possible the influence of all other risk factors for AM1 that may have affected the smoking pattern; this is accomplished by including variables representing these (confounding) factors in the regression model. We have included the following possibly confounding factors in the models: age, sex, family history of AMI, earlobe crease, length of school education, income, and marital status. We have also included in the analysis information about life style factors other than smoking, namely, body mass index (BMI), physical activity during work, and alcohol consumption, as well as systolic blood pressure and plasma total cholesterol concentration, to examine whether there are interactions between the effect of these factors and the effect of smoking. Collection of data. Systolic blood pressure was measured with subjects in the sitting position, after 5 minutes of rest, by means of the London School of Hygiene sphygmomanometer. Plasma cholesterol levels were measured in the nonfasting state by means of an enzymatic method. Systolic blood pressure and plasma cholesterol levels were converted to standardized deviates by subtracting the mean and dividing by the standardized deviates for the sex and lo-year age group in question. Earlobe crease was characterized by measuring the length and depth of a possible crease, and BMI was calculated in each person from measured weight and height (weightJheight2, kg/m2). The remaining information was obtained from a questionnaire. Each participant was asked whether his mother or father had had an AMI; if one of these questions was answered affirmatively, that participant was included in the “yes category” for the variable “Family history of AMI.” With regard to education participants were divided into two groups: persons with compulsory schooling only (less than 8 years in school) and persons with more education. Information about household income was grouped into three categories representing low, medium, and high incomes. Furthermore, participants stated whether they were living with a spouse or living alone. Physical activity at work was classified into four groups according to the subjects’ own assessments. Persons in groups 1 or 2, who were mainly sitting or standing and only occasionally walking, were classified as “physically inactive.” The questionnaire included a question about frequency of alcohol intake and in cases of daily alcohol consumption the average number of drinks per day, one drink containing 12 to 15 gm of alcohol. Participants were asked if they were current smokers, ex-smokers, or had never smoked. Current smokers were asked what type of tobacco they used (plain or filtered cigarettes, cheroots/cigars, pipe or mixed), the average weekly number of 50 gm pipe tobacco packages and the average daily number of other items smoked, the duration of the smoking habit, and whether or not they inhaled the smoke. The information about the amount of tobacco was converted into grams of tobacco consumed per day by using the following conversion factors: 1 cigarette = 1 gm, 1 cheroot = 3 gm, 1 cigar = 5 gm, and one 50 gm package/wk = 7 gm. Ex-smokers were asked about the duration of the smoking habit and the time elapsed since they had stopped.
Smoking
Table
and
risk
of first
AMI
439
I. Study population for risk factor analysis Study
Males
population
Invited to examination Nonresponders Responders Subjects aged 20 to 29 yr Subjects with previous AM1 Subjects with missing data Included in risk factor analysis New cases of AM1 within 7 yr
Females
9,145 2,634 6.51 I 283 305 276 5.647 260
10,184 2,472 7,712 327 144 692 6,549 loo
Total
19,329 5,106 14,223 610 449 968 12,196 360
Biochemical verification of the smoking information was not attempted. The prevalence of the covariables included in the analysis and the estimated relative risk (ERR) are shown in Table II. Statietkal analysis. The subject of the risk factor analysis is the hazard or instantaneous rate of a first AMI, and in our analysis we used the Cox regression model for proportional hazards.14 The period of observation was counted from the time of the initial examination until the earliest occurence of one of the following: the person had an AM1 (event), died of causes other than AMI (no event), or survived for 7 years or until December 31,X%3, without AM1 (no event). Some of the covariables had only two outcomes, for example, inhalation, and thus could be represented by a binary variable given the values zero (no inhalation) and one (inhalation). Covariables with more than two outcomes were sometimes included directly as a quantitative variable (e.g., BMI) and sometimes represented by means of a number of binary variables, each indicating whether or not the subject belonged to a specific category or had a variable value within a certain interval. Estimates of the regression coefficients were obtained by using the maximum likelihood method as suggested by Cox, and the question of whether the AMI hazard was significantly affected by a covariable was evaluated by means of the likelihood ratio test. The results of the analysis are presented in terms of ERR, that is, the hazard in the specified group divided by the hazard in a group selected as a basis, For BMI the ERR represents the relative increase in risk for each unit of BMI, which for individuals of normal height represents a weight increase of approximately 3 kg. Similarly the ERR for tobacco (Table III) relates to an increase of 1 gm of tobacco. On the basis of the estimated hazard in various groups of smokers relative to nonsmokers, we have calculated the so-called population attributable risk of smoking, that is, the proportion of the AM1 risk that is “explained by” or “may be attributed to” smoking in subgroups of the population. The method of calculation is described by Breslow and Day.15 RESULTS Smoking
Distribution the amount of tobacco by sex and age are shown in Fig.
habfts in the sttvdy po9ulstEon.
of smoking habits3 including
consumed &n/day)
Nyboe et al.
440
American
Table II. Distribution of covariables included effect of smoking with estimates of relative Frequency in study population (%o)
Covariable
(F)
30-39 yr 40-49 50-59 60+
(M) 30-39 yr 40-49 50-59 60+
Family disposition Earlobe crease (F) Less than 8yr (M) in school Income Partner (F) Low No Yes Middle No or high Yes (M) Low No Yes Middle No or high Yes Physical inactivity during work Daily alcohol intake None 1-2 drinks/day 3-4 drinks/day >4 drinks/day BMI per unit (kg/m2) Cholesterol St. deviate < 0.75 2
Systolic blood pressure St. deviate < - 0.5 -0.5 >0.5
to 0.5
*From a model including given in Table IV.
6.1 12.6 20.3 14.7 5.7 11.1 15.0 14.5 17.6 21.8 26.6
1.0 0.44 2.6 8.9 0.81 5.9 14 26 1.3 1.2 1.2
22.7
1.10
7.5 8.4 28.1 3.6 5.9 5.8 31.0
1.5 2.5 0.96 1.0 1.1 1.5 0.93 1.0
72.6
1.3
72.8 13.6 7.2 6.4
1.0 1.0 0.80 0.56 1.015
80.6 19.4
1.0 2.0
33.4 39.7 26.9
1.0 1.4 2.0
all above-mentioned
of
Estimated relative risk of AMZ*
9.7
0.75
in analysis AM1 risks
variables
and smoking
as
1. The pattern of smoking is different in the two sexes: among men in the 40- to 59-year age groups approximately three fourths are smokers. In the younger age groups the frequency of smokers is approximately 60%) inasmuch as a greater proportion had never smoked. A similar proportion of smokers is seen in the older age groups, among whom a larger number were ex-smokers. The average amount of tobacco smoked per smoker is fairly steady, just under 20 gm/day, with a slight decrease among the oldest age groups. Among women in the youngest age groups the fre-
Auguet 1991 Heart Journal
quency of smoking is very similar to that of men. In the 40- to 59-year age groups a slightly higher frequency is seen, although it never exceeded two thirds. In the older age groups the frequency decreases rapidly, mainly because a large proportion had never started to smoke. The average amount of tobacco smoked per female smoker is nearly the same in all age groups, just under 15 gm/day. Distributions of smokers according to type of tobacco used and inhalation are shown in Fig. 2. Almost 10% of the male smokers smoked a pipe, whereas this type of smoking is extremely rare among women. However, cigar/cheroot smoking is quite common in middle-aged women-a peculiar Danish cultural trait. Most female smokers, especially in the younger age groups, smoke filtered cigarettes, whereas the use of plain and filtered cigarettes is equally distributed in male smokers. The great majority of cigarette smokers inhale. Approximately two thirds of cigar/cheroot smokers deny inhaling, the ratio being somewhat higher in women than in men. Many pipe smokers inhale. Risk of first AMI. Some of the information about smoking habits pertains only to nonsmokers, some only to smokers, and the preliminary analyses were made within the category in question. Nonsmokers. The group of nonsmokers consisted of 4406 persons among whom 83 cases of first AM1 were found. In this group 54.8% had never smoked and the remainder were ex-smokers. A binary variable characterizing these two groups included in the model containing the 10 covariables mentioned previously was found to be of no significance (x2 = 0.48, df = l), the ERR for ex-smokers being 0.84 with 95 % confidence limits (0.52 to 1.4). Inasmuch as there was no indication that the group of previous smokers had a higher risk of AMI, there was no reason to assume that further subdivision of the ex-smokers would alter this result. In fact it was confirmed that the risk did not differ significantly among ex-smokers who had given up the habit less than 1 year previously, 1 to 5 years previously, or more than 5 years previously (x2 = 2.3, df = 2). In subjects who had stopped smoking for more than 5 years the ERR was 1.3 (confidence limits 0.63 to 2.3) relative to subjects who stopped smoking within the last 5 years. Similarly subdividing the duration of the smoking habit before quitting into three groups-less than 15 years, 15 to 29 years, or 30 years or more-yielded no significant influence on the risk of AM1 (x2 = 1.2, df = 2) The ERR in subjects having smoked for 30 years or more was 0.97 (confidence limits 0.57 to 1.6) relative to subjects who had smoked for less than 30 years. Smokers.This group comprised 7790 persons with
Volume 122 Number 2
Smoking
and
risk
n: first
AMI
441
III. Estimates of coefficients in three models showing correlation between risk of first AM1 and smoking variables (models contain the covariables given in Table II) -.-
Table
Model 1
2
3
Variable
Estimated coefficient
Standard error
Ratio
ERR I__.-
Inhalation (regardless of type) Tobacco in grams (regardless of type) Inhalation (regardless of type) Tobacco in grams Plain cigarettes Filtered cigarettes Cheroots/cigars Pipe tobacco Inhalation of Plain cigarettes Filtered cigarettes Cheroots/cigars Pipe tobacco Tobacco in grams (regardless of type)
0.666
0.158
4.2
1.96
0.024
0.0059
4.1
1.024
0.661
0.172
3.8
i .94
0.027 0.018 0.024 0.020
0.0084 0.0108 0.0628 0.0126
x2 I.7 ri.8 I .6
1.027 1.018 1.024 1.020
0.731 0.550 0.735 0.528 0.023
0.174 0.199 0.240 0.337 0.0619
4.2 2.8 :i.1 1.6 3.8
2.08 I 73 2.09 1.70 I.023
Comparison of models by likelihood ratio test: Model 1 with “full” model: x2 = 8.8, df = 13, 0.8 > p > 0.7. Model 1 with model 2: x2 = 0.99, df = 3, 0.9 > p > 0.8. Model 1 with model 3: x2 = 1.4, df = 3, 0.8 > p > 0.7.
277 cases of AMI. The number of subjects smoking more than one type of tobacco was too small to shed any light on the effect of “mixed smoking.” They were therefore excluded leaving 7286 persons with 251 cases of AMI for the preliminary analysis. In this analysis it was assumed that the relationship between the logarithm of the AM1 hazard and the amount of tobacco smoked daily is linear, that is, may be characterized by an intercept and a slope. We first fitted to the data a detailed model describing the relationship between the AM1 hazard and the amount of tobacco smoked (in grams/day) separately for each of eight groups of smokers character&d by type of tobacco used (plain cigarettes, filtered cigarettes, cheroots/cigars, or pipe tobacco) and by inhalation (yes or no). This model contained 15 “smoking variables” (eight for slopes and eight minus one for intercepts) in addition to the covariablee given in Table II. By comparing this model with simpler models for smoking, it was found that the typeof tobacco had no statistical significance, whereas the amount of tobacco and inhalation were both highly sign&ant. In fact the likelihood ratio test for comparison of the preceding “full” model with a model containing only one slope and one intercept for inhalation showed x2 = 8.8 for df = 13, that is, there was no significant loss of information with the use of the reduced model (Table III). Table III also gives results for two more detailed models, one containing
different slopes for each type of tobacco (model 2) and the other containing different inbmxpb for each type of tobacco (model 3). Thus we were unable to demonstrate significant differences between the effects of different types of tobacco, but the uneertainty of the estimates does not preclude the existence of considerable differences. Next we investigated whether the duration of the smoking habit had an influence on the risk of AMI. When a variable representing the years was added to the simple m amoung of tobacco and inhalati icant at the 5% level (x2 = 0.2, df = 1). The same conclusion was reached when the duration of smoking was represented by binary varia&s representing groups: 1 to 14 years, 15 to 29 years, and 30 or more years (x2 = 3.3, df = 2). The ERR in subjects who had smoked for 20 years or more was V,% (confidence limits 0.46 to 1.3) relative to subjects who had smoked for less than 20 years. Thus we were unable to demonstrate a relationship between the risk of AMI and the duration of smoking in the past. Population mod&s. The results of the preceding analyses were combined in a model cable to the entire population. In this model the smoking data were represented by means of six b&q variables characterizing noninhalers amok& less than 15 gm/ day, noninhalers smoking 15 to 29 g&day, noninhalers smoking 30 gm/day or more, and three similar
442
Nyboe et al.
Fig.
American
August 1991 Heart Journal
1. Smoking habits in Copenhagen City Heart Study population by sex and age.
variablesfor inhalers in relation to nonsmokers.The ERR obtained with this model are shown in Fig. 3. The risk of AM1 is lowest among nonsmokers,it increasesexponentially with the amount of tobacco smoked, and for a given amount of tobaccothe risk is nearly twice as high amonginhalers asamongnoninhalers.In Fig. 3the smoking groupsarerepresented on the horizontal scalein accordancewith their distribution in the study population; this representation showsthat the major part of the attributable risk of smoking (seebelow) comesfrom the inhalers. To examinewhetherthe effect of smoking interacts with the effect of other factors, it was necessaryto simplify the model, namely, to disregardthe amount of tobaccoamong the noninhalers.With this simplified model we found a strongereffect of smoking relative to not smoking amongwomenthan amongmen (Table IV). The estimated difference between the risks in men and women increasedconsistently with
increasingsmoking intensity, and the differencewas statistically significant at the 5% level in the three groups of smokers who inhaled. The difference between the sexesappears convincing, although the overall comparison of the models with and without interaction terms did not show a significant differenceat the 5% level. We also studied the effect of smoking separately for personswith and without daily alcohol intake. A significant interaction was found between the two factors (Table V). Whereasthe risk of AM1 increased sharply with inhalation and with increasingamounts of tobaccoamong personswithout daily intake of alcohol,the increasewasmuch lessamongdaily drinkers.Among the heaviestsmokersthe risk of AM1 was significantly lower among personsdrinking alcohol daily than among those not doing so. On the other hand, the difference in the opposite direction between the risks in the correspondinggroups of non-
Volums 122 Number 2
Smoking
arrd r&k of first AMI
443
2. Use of different types of tobacco smoked and inhalation among smokers in Copenhagen City Heart Study by sex and age. Shaded areas of columns represent subjects who inhaled.
Fig.
smokers was not statistically significant at the 5% level. We have not been able to demonstrate any interaction between the effect of smoking and the effects of BMI or physical inactivity during work. Neither could interactive effects of smoking and systolic blood pressure, respective smoking, and total plasma cholesterol concentration be demonstrated (Tables VI and VII). Attribwtable risk. Based on the ERRS given in Table IV and the distributions of the five smoking groups in each sex and lo-year age group, we have calculated the estimated attributable risks shown in Table VIII. The proportion of the AM1 risk that can be attributed to smoking is for women and men of all ages 0.59 and 0.38, respectively. Although women generally smoke less than men, their higher relative risks result in higher attributable risks in all age groups. In both
sexes the attributable risk is lower among persons more than 60 years of age than among younger persons, because there are fewer smokers and fewer inhalers in the oldest age group. DISCUSSlOW
The Copenhagen City Heart Study has confirmed the strength of the “classical” coronary risk factors,‘2,‘3 and particularly the role of tobacco smoking as a major risk factor for AMI. In the present study some new facets have been added to the picture. Previous observations that the risk depends on the amount of tobacco smoked are con&m&, and the present analysis confirm@ that the habit of inhaling the smoke also plays a very important role. In fact it appears that smoking a considerable amount of tobacco leads only to a modest increase in the risk of
444
Augurt 1991 American Heart Journal
Nyboe et al.
3. Estimated effects of tobacco smoking on risk of first AMI. Horizontal scale represents distribution of smoking groups in study population.
Fig.
IV. Estimated effect of tobacco smoking on risk of AM1 among females and males
Table
Relative risk of first AZdZ among Smoking group Nonsmokers Noninhalers Inhalers 1-14 gm/day 15-29 gmjday 130 gm/day Comparison with
model
without
Females
Males
1.0 1.5
1.0 1.2
3.6 4.6 9.4
1.6 2.1 2.9
interaction
term:
x2 = 9.3,
df = 4,
0.1 > p > 0.05.
AMI, provided that the smoke is not inhaled. On the other hand, inhalation always results in a considerably increased risk, even for a small amount of tobacco consumption. Our data suggest that the type of tobacco used is unimportant; filtered cigarettes and pipes are as dangerous as plain cigarettes and cigars/cheroots.
This finding is in agreement with those of other studies on pipe and cigar smoking,5y 6 although other studies on filtered versus plain cigarettes show small inconsistent differences.16-I8 It should be recognized that smoking habits with regard to the amount smoked and inhalation may be profoundly affected by the type of tobacco smoked and, because many more pipe and cigar smokers did not inhale, a case could still be made for persuading inveterate cigarette smokers to switch to pipe smoking. Pipe smokers tend to consume less tobacco because of the higher nicotine content of pipe tobacco.iQ On the other hand, inhalers changing to another type of tobacco are likely to continue to inhale. Inhalation seems to be the major determinant of the dose-dependent risk in smokers, but even noninhalers experience a dose-dependent risk. This may be partly explained by underreporting of inhalation, but it should be noted that “passive smokers,” that is, nonsmoking subjects exposed to tobacco smoke, are also at increased risk for AMI. The risk of AM1 in nonsmoking women is much
Volume 122 Number 2
V. Estimated effect of tobacco smoking on risk of AM1 according to alcohol consumption
Table
VII. Estimated effect of tobacco smoking on risk of AM1 according to level of total cholesterol
Table
Relative risk of AMI according to frequency of alcohol consumption
Relative risk of p’rst AM1 according to level of tota! cholesterol Standardized derxato ~I$ z 0 75
J#rgen Nyboe, MSc, Gorm Jensen, MD, PhD, Merete Appleyard, Peter Schnohr, MD. Copenhagen, Denmark
In numerous epidemiologic studies smoking has consistently been identified as a major risk factor for ischemic heart disease.1-4 The pathophysiologic mechanisms responsible for the adverse effect of smoking are by no means clear. The risk seems to be dose dependent with regard to the amount of tobacco smoked.4 Most studies are concerned with cigarette smoking only, but the few dealing with other types of tobacco suggest that pipe and cigar smoking may be equally dangerous. 4-6 Some studies have shown that the risk decreases after cessation of smoking, especially in subjects with prior manifestations of ischemic heart disease,73 8 but the magnitude and rate of this decrease have not yet been determined.g In this article we present data concerning the prevalence of smoking in a sample of the population of Copenhagen and the role of tobacco smoking as a risk factor for first acute myocardial infarction (AMI). Among current smokers we have examined the risk associated with the amount and type of tobacco smoked including filtered versus plain cigarettes, duration of the smoking habit, and inhalation. Among From the Copenhagen City Heart Study, Rigshospitalet. Supported by the Danish Heart Foundation and by grants from Laegeforeningens Forskningsfond, Lily Benthine Lunds Fond, Hafnia-Haand i Haand Fond&, and Fabrikant Mads Clausens Fond. Received for publication Sept. 18, 1990; accepted Jan. 23, 1991. Reprint requests: G. Jensen, the Copenhagen City Heart Study, Department 7112, Rigshospitalet, Tagensvej 20, DK-2200 Copenhagen, Denmark. 4/l/29665
438
RHLT,
and
current nonsmokers we have examined whether cessation of smoking has reduced the risk of AMI. Furthermore, we have studied whether the effect of smoking on the risk of AM1 differs between men and women and whether it is influenced by daily alcohol consumption. METHODS
Subjects.The Copenhagen City Heart Study is a prospective cardiovascular population study comprising almost 20,000 men and women, aged 20 years or more, selected at random after age stratification among 90,000 persons living in a defined area around Rigshospitalet, University Hospital of Copenhagen. The persons selected were invited by letter to a health examination at Rigshospitalet on a specific date between March 1,1976 and March 31,1978, to collect basic data concerning possible risk factors for each person. A detailed description of the study procedures has previously been published.lO* l1 Information about new cases of AM1 was obtained from a followup examination between 1981 and 83 and through scrutiny of the National Health Service registers of death and hospitalization to the end of 1983, that is, a follow-up period of approximately 6.5 years. All caseswere verified by means of hospital records and death certificates. Analysis of risk factors is limited to persons aged 30 years or more and comprises 12,196 persons without prior AM1 and 360 cases of first AM1 (Table I).12vl3 Methodology.We have used a regression technique to describe the risk of AM1 during the follow-up period as a function of tobacco smoking as assessedby questionnaire at the initial examination. When the aim is a causative in-
Volume 122 Number 2
terpretation of the relationship, it is desirable to eliminate as far as possible the influence of all other risk factors for AM1 that may have affected the smoking pattern; this is accomplished by including variables representing these (confounding) factors in the regression model. We have included the following possibly confounding factors in the models: age, sex, family history of AMI, earlobe crease, length of school education, income, and marital status. We have also included in the analysis information about life style factors other than smoking, namely, body mass index (BMI), physical activity during work, and alcohol consumption, as well as systolic blood pressure and plasma total cholesterol concentration, to examine whether there are interactions between the effect of these factors and the effect of smoking. Collection of data. Systolic blood pressure was measured with subjects in the sitting position, after 5 minutes of rest, by means of the London School of Hygiene sphygmomanometer. Plasma cholesterol levels were measured in the nonfasting state by means of an enzymatic method. Systolic blood pressure and plasma cholesterol levels were converted to standardized deviates by subtracting the mean and dividing by the standardized deviates for the sex and lo-year age group in question. Earlobe crease was characterized by measuring the length and depth of a possible crease, and BMI was calculated in each person from measured weight and height (weightJheight2, kg/m2). The remaining information was obtained from a questionnaire. Each participant was asked whether his mother or father had had an AMI; if one of these questions was answered affirmatively, that participant was included in the “yes category” for the variable “Family history of AMI.” With regard to education participants were divided into two groups: persons with compulsory schooling only (less than 8 years in school) and persons with more education. Information about household income was grouped into three categories representing low, medium, and high incomes. Furthermore, participants stated whether they were living with a spouse or living alone. Physical activity at work was classified into four groups according to the subjects’ own assessments. Persons in groups 1 or 2, who were mainly sitting or standing and only occasionally walking, were classified as “physically inactive.” The questionnaire included a question about frequency of alcohol intake and in cases of daily alcohol consumption the average number of drinks per day, one drink containing 12 to 15 gm of alcohol. Participants were asked if they were current smokers, ex-smokers, or had never smoked. Current smokers were asked what type of tobacco they used (plain or filtered cigarettes, cheroots/cigars, pipe or mixed), the average weekly number of 50 gm pipe tobacco packages and the average daily number of other items smoked, the duration of the smoking habit, and whether or not they inhaled the smoke. The information about the amount of tobacco was converted into grams of tobacco consumed per day by using the following conversion factors: 1 cigarette = 1 gm, 1 cheroot = 3 gm, 1 cigar = 5 gm, and one 50 gm package/wk = 7 gm. Ex-smokers were asked about the duration of the smoking habit and the time elapsed since they had stopped.
Smoking
Table
and
risk
of first
AMI
439
I. Study population for risk factor analysis Study
Males
population
Invited to examination Nonresponders Responders Subjects aged 20 to 29 yr Subjects with previous AM1 Subjects with missing data Included in risk factor analysis New cases of AM1 within 7 yr
Females
9,145 2,634 6.51 I 283 305 276 5.647 260
10,184 2,472 7,712 327 144 692 6,549 loo
Total
19,329 5,106 14,223 610 449 968 12,196 360
Biochemical verification of the smoking information was not attempted. The prevalence of the covariables included in the analysis and the estimated relative risk (ERR) are shown in Table II. Statietkal analysis. The subject of the risk factor analysis is the hazard or instantaneous rate of a first AMI, and in our analysis we used the Cox regression model for proportional hazards.14 The period of observation was counted from the time of the initial examination until the earliest occurence of one of the following: the person had an AM1 (event), died of causes other than AMI (no event), or survived for 7 years or until December 31,X%3, without AM1 (no event). Some of the covariables had only two outcomes, for example, inhalation, and thus could be represented by a binary variable given the values zero (no inhalation) and one (inhalation). Covariables with more than two outcomes were sometimes included directly as a quantitative variable (e.g., BMI) and sometimes represented by means of a number of binary variables, each indicating whether or not the subject belonged to a specific category or had a variable value within a certain interval. Estimates of the regression coefficients were obtained by using the maximum likelihood method as suggested by Cox, and the question of whether the AMI hazard was significantly affected by a covariable was evaluated by means of the likelihood ratio test. The results of the analysis are presented in terms of ERR, that is, the hazard in the specified group divided by the hazard in a group selected as a basis, For BMI the ERR represents the relative increase in risk for each unit of BMI, which for individuals of normal height represents a weight increase of approximately 3 kg. Similarly the ERR for tobacco (Table III) relates to an increase of 1 gm of tobacco. On the basis of the estimated hazard in various groups of smokers relative to nonsmokers, we have calculated the so-called population attributable risk of smoking, that is, the proportion of the AM1 risk that is “explained by” or “may be attributed to” smoking in subgroups of the population. The method of calculation is described by Breslow and Day.15 RESULTS Smoking
Distribution the amount of tobacco by sex and age are shown in Fig.
habfts in the sttvdy po9ulstEon.
of smoking habits3 including
consumed &n/day)
Nyboe et al.
440
American
Table II. Distribution of covariables included effect of smoking with estimates of relative Frequency in study population (%o)
Covariable
(F)
30-39 yr 40-49 50-59 60+
(M) 30-39 yr 40-49 50-59 60+
Family disposition Earlobe crease (F) Less than 8yr (M) in school Income Partner (F) Low No Yes Middle No or high Yes (M) Low No Yes Middle No or high Yes Physical inactivity during work Daily alcohol intake None 1-2 drinks/day 3-4 drinks/day >4 drinks/day BMI per unit (kg/m2) Cholesterol St. deviate < 0.75 2
Systolic blood pressure St. deviate < - 0.5 -0.5 >0.5
to 0.5
*From a model including given in Table IV.
6.1 12.6 20.3 14.7 5.7 11.1 15.0 14.5 17.6 21.8 26.6
1.0 0.44 2.6 8.9 0.81 5.9 14 26 1.3 1.2 1.2
22.7
1.10
7.5 8.4 28.1 3.6 5.9 5.8 31.0
1.5 2.5 0.96 1.0 1.1 1.5 0.93 1.0
72.6
1.3
72.8 13.6 7.2 6.4
1.0 1.0 0.80 0.56 1.015
80.6 19.4
1.0 2.0
33.4 39.7 26.9
1.0 1.4 2.0
all above-mentioned
of
Estimated relative risk of AMZ*
9.7
0.75
in analysis AM1 risks
variables
and smoking
as
1. The pattern of smoking is different in the two sexes: among men in the 40- to 59-year age groups approximately three fourths are smokers. In the younger age groups the frequency of smokers is approximately 60%) inasmuch as a greater proportion had never smoked. A similar proportion of smokers is seen in the older age groups, among whom a larger number were ex-smokers. The average amount of tobacco smoked per smoker is fairly steady, just under 20 gm/day, with a slight decrease among the oldest age groups. Among women in the youngest age groups the fre-
Auguet 1991 Heart Journal
quency of smoking is very similar to that of men. In the 40- to 59-year age groups a slightly higher frequency is seen, although it never exceeded two thirds. In the older age groups the frequency decreases rapidly, mainly because a large proportion had never started to smoke. The average amount of tobacco smoked per female smoker is nearly the same in all age groups, just under 15 gm/day. Distributions of smokers according to type of tobacco used and inhalation are shown in Fig. 2. Almost 10% of the male smokers smoked a pipe, whereas this type of smoking is extremely rare among women. However, cigar/cheroot smoking is quite common in middle-aged women-a peculiar Danish cultural trait. Most female smokers, especially in the younger age groups, smoke filtered cigarettes, whereas the use of plain and filtered cigarettes is equally distributed in male smokers. The great majority of cigarette smokers inhale. Approximately two thirds of cigar/cheroot smokers deny inhaling, the ratio being somewhat higher in women than in men. Many pipe smokers inhale. Risk of first AMI. Some of the information about smoking habits pertains only to nonsmokers, some only to smokers, and the preliminary analyses were made within the category in question. Nonsmokers. The group of nonsmokers consisted of 4406 persons among whom 83 cases of first AM1 were found. In this group 54.8% had never smoked and the remainder were ex-smokers. A binary variable characterizing these two groups included in the model containing the 10 covariables mentioned previously was found to be of no significance (x2 = 0.48, df = l), the ERR for ex-smokers being 0.84 with 95 % confidence limits (0.52 to 1.4). Inasmuch as there was no indication that the group of previous smokers had a higher risk of AMI, there was no reason to assume that further subdivision of the ex-smokers would alter this result. In fact it was confirmed that the risk did not differ significantly among ex-smokers who had given up the habit less than 1 year previously, 1 to 5 years previously, or more than 5 years previously (x2 = 2.3, df = 2). In subjects who had stopped smoking for more than 5 years the ERR was 1.3 (confidence limits 0.63 to 2.3) relative to subjects who stopped smoking within the last 5 years. Similarly subdividing the duration of the smoking habit before quitting into three groups-less than 15 years, 15 to 29 years, or 30 years or more-yielded no significant influence on the risk of AM1 (x2 = 1.2, df = 2) The ERR in subjects having smoked for 30 years or more was 0.97 (confidence limits 0.57 to 1.6) relative to subjects who had smoked for less than 30 years. Smokers.This group comprised 7790 persons with
Volume 122 Number 2
Smoking
and
risk
n: first
AMI
441
III. Estimates of coefficients in three models showing correlation between risk of first AM1 and smoking variables (models contain the covariables given in Table II) -.-
Table
Model 1
2
3
Variable
Estimated coefficient
Standard error
Ratio
ERR I__.-
Inhalation (regardless of type) Tobacco in grams (regardless of type) Inhalation (regardless of type) Tobacco in grams Plain cigarettes Filtered cigarettes Cheroots/cigars Pipe tobacco Inhalation of Plain cigarettes Filtered cigarettes Cheroots/cigars Pipe tobacco Tobacco in grams (regardless of type)
0.666
0.158
4.2
1.96
0.024
0.0059
4.1
1.024
0.661
0.172
3.8
i .94
0.027 0.018 0.024 0.020
0.0084 0.0108 0.0628 0.0126
x2 I.7 ri.8 I .6
1.027 1.018 1.024 1.020
0.731 0.550 0.735 0.528 0.023
0.174 0.199 0.240 0.337 0.0619
4.2 2.8 :i.1 1.6 3.8
2.08 I 73 2.09 1.70 I.023
Comparison of models by likelihood ratio test: Model 1 with “full” model: x2 = 8.8, df = 13, 0.8 > p > 0.7. Model 1 with model 2: x2 = 0.99, df = 3, 0.9 > p > 0.8. Model 1 with model 3: x2 = 1.4, df = 3, 0.8 > p > 0.7.
277 cases of AMI. The number of subjects smoking more than one type of tobacco was too small to shed any light on the effect of “mixed smoking.” They were therefore excluded leaving 7286 persons with 251 cases of AMI for the preliminary analysis. In this analysis it was assumed that the relationship between the logarithm of the AM1 hazard and the amount of tobacco smoked daily is linear, that is, may be characterized by an intercept and a slope. We first fitted to the data a detailed model describing the relationship between the AM1 hazard and the amount of tobacco smoked (in grams/day) separately for each of eight groups of smokers character&d by type of tobacco used (plain cigarettes, filtered cigarettes, cheroots/cigars, or pipe tobacco) and by inhalation (yes or no). This model contained 15 “smoking variables” (eight for slopes and eight minus one for intercepts) in addition to the covariablee given in Table II. By comparing this model with simpler models for smoking, it was found that the typeof tobacco had no statistical significance, whereas the amount of tobacco and inhalation were both highly sign&ant. In fact the likelihood ratio test for comparison of the preceding “full” model with a model containing only one slope and one intercept for inhalation showed x2 = 8.8 for df = 13, that is, there was no significant loss of information with the use of the reduced model (Table III). Table III also gives results for two more detailed models, one containing
different slopes for each type of tobacco (model 2) and the other containing different inbmxpb for each type of tobacco (model 3). Thus we were unable to demonstrate significant differences between the effects of different types of tobacco, but the uneertainty of the estimates does not preclude the existence of considerable differences. Next we investigated whether the duration of the smoking habit had an influence on the risk of AMI. When a variable representing the years was added to the simple m amoung of tobacco and inhalati icant at the 5% level (x2 = 0.2, df = 1). The same conclusion was reached when the duration of smoking was represented by binary varia&s representing groups: 1 to 14 years, 15 to 29 years, and 30 or more years (x2 = 3.3, df = 2). The ERR in subjects who had smoked for 20 years or more was V,% (confidence limits 0.46 to 1.3) relative to subjects who had smoked for less than 20 years. Thus we were unable to demonstrate a relationship between the risk of AMI and the duration of smoking in the past. Population mod&s. The results of the preceding analyses were combined in a model cable to the entire population. In this model the smoking data were represented by means of six b&q variables characterizing noninhalers amok& less than 15 gm/ day, noninhalers smoking 15 to 29 g&day, noninhalers smoking 30 gm/day or more, and three similar
442
Nyboe et al.
Fig.
American
August 1991 Heart Journal
1. Smoking habits in Copenhagen City Heart Study population by sex and age.
variablesfor inhalers in relation to nonsmokers.The ERR obtained with this model are shown in Fig. 3. The risk of AM1 is lowest among nonsmokers,it increasesexponentially with the amount of tobacco smoked, and for a given amount of tobaccothe risk is nearly twice as high amonginhalers asamongnoninhalers.In Fig. 3the smoking groupsarerepresented on the horizontal scalein accordancewith their distribution in the study population; this representation showsthat the major part of the attributable risk of smoking (seebelow) comesfrom the inhalers. To examinewhetherthe effect of smoking interacts with the effect of other factors, it was necessaryto simplify the model, namely, to disregardthe amount of tobaccoamong the noninhalers.With this simplified model we found a strongereffect of smoking relative to not smoking amongwomenthan amongmen (Table IV). The estimated difference between the risks in men and women increasedconsistently with
increasingsmoking intensity, and the differencewas statistically significant at the 5% level in the three groups of smokers who inhaled. The difference between the sexesappears convincing, although the overall comparison of the models with and without interaction terms did not show a significant differenceat the 5% level. We also studied the effect of smoking separately for personswith and without daily alcohol intake. A significant interaction was found between the two factors (Table V). Whereasthe risk of AM1 increased sharply with inhalation and with increasingamounts of tobaccoamong personswithout daily intake of alcohol,the increasewasmuch lessamongdaily drinkers.Among the heaviestsmokersthe risk of AM1 was significantly lower among personsdrinking alcohol daily than among those not doing so. On the other hand, the difference in the opposite direction between the risks in the correspondinggroups of non-
Volums 122 Number 2
Smoking
arrd r&k of first AMI
443
2. Use of different types of tobacco smoked and inhalation among smokers in Copenhagen City Heart Study by sex and age. Shaded areas of columns represent subjects who inhaled.
Fig.
smokers was not statistically significant at the 5% level. We have not been able to demonstrate any interaction between the effect of smoking and the effects of BMI or physical inactivity during work. Neither could interactive effects of smoking and systolic blood pressure, respective smoking, and total plasma cholesterol concentration be demonstrated (Tables VI and VII). Attribwtable risk. Based on the ERRS given in Table IV and the distributions of the five smoking groups in each sex and lo-year age group, we have calculated the estimated attributable risks shown in Table VIII. The proportion of the AM1 risk that can be attributed to smoking is for women and men of all ages 0.59 and 0.38, respectively. Although women generally smoke less than men, their higher relative risks result in higher attributable risks in all age groups. In both
sexes the attributable risk is lower among persons more than 60 years of age than among younger persons, because there are fewer smokers and fewer inhalers in the oldest age group. DISCUSSlOW
The Copenhagen City Heart Study has confirmed the strength of the “classical” coronary risk factors,‘2,‘3 and particularly the role of tobacco smoking as a major risk factor for AMI. In the present study some new facets have been added to the picture. Previous observations that the risk depends on the amount of tobacco smoked are con&m&, and the present analysis confirm@ that the habit of inhaling the smoke also plays a very important role. In fact it appears that smoking a considerable amount of tobacco leads only to a modest increase in the risk of
444
Augurt 1991 American Heart Journal
Nyboe et al.
3. Estimated effects of tobacco smoking on risk of first AMI. Horizontal scale represents distribution of smoking groups in study population.
Fig.
IV. Estimated effect of tobacco smoking on risk of AM1 among females and males
Table
Relative risk of first AZdZ among Smoking group Nonsmokers Noninhalers Inhalers 1-14 gm/day 15-29 gmjday 130 gm/day Comparison with
model
without
Females
Males
1.0 1.5
1.0 1.2
3.6 4.6 9.4
1.6 2.1 2.9
interaction
term:
x2 = 9.3,
df = 4,
0.1 > p > 0.05.
AMI, provided that the smoke is not inhaled. On the other hand, inhalation always results in a considerably increased risk, even for a small amount of tobacco consumption. Our data suggest that the type of tobacco used is unimportant; filtered cigarettes and pipes are as dangerous as plain cigarettes and cigars/cheroots.
This finding is in agreement with those of other studies on pipe and cigar smoking,5y 6 although other studies on filtered versus plain cigarettes show small inconsistent differences.16-I8 It should be recognized that smoking habits with regard to the amount smoked and inhalation may be profoundly affected by the type of tobacco smoked and, because many more pipe and cigar smokers did not inhale, a case could still be made for persuading inveterate cigarette smokers to switch to pipe smoking. Pipe smokers tend to consume less tobacco because of the higher nicotine content of pipe tobacco.iQ On the other hand, inhalers changing to another type of tobacco are likely to continue to inhale. Inhalation seems to be the major determinant of the dose-dependent risk in smokers, but even noninhalers experience a dose-dependent risk. This may be partly explained by underreporting of inhalation, but it should be noted that “passive smokers,” that is, nonsmoking subjects exposed to tobacco smoke, are also at increased risk for AMI. The risk of AM1 in nonsmoking women is much
Volume 122 Number 2
V. Estimated effect of tobacco smoking on risk of AM1 according to alcohol consumption
Table
VII. Estimated effect of tobacco smoking on risk of AM1 according to level of total cholesterol
Table
Relative risk of AMI according to frequency of alcohol consumption
Relative risk of p’rst AM1 according to level of tota! cholesterol Standardized derxato ~I$ z 0 75
