Khalida Ismail, 1 Andy Sloggett, 2 and Bianca De Stavola 3

Transcription

1 American Journal of Eidemiology Coyright 2000 by The Johns Hokins University School of Hygiene and Public Health All rights reserved Vol. 52, No. 7 Printed in U.S.A. Common Mental Disorders and Cigarette Smoking Ismail et al. Do Common Mental Disorders Increase Cigarette Smoking? Results from Five Waves of a Poulation-based Panel Cohort Study Khalida Ismail, Andy Sloggett, 2 and Bianca De Stavola 3 A rosective anel cohort design was used to investigate whether mental disorders common in the general oulation increase the likelihood of increased cigarette smoking at 2 months follow-u. By 995, the last year for which data were available, a random samle of 2,057 ersons aged 6 75 years residing in rivate households in Great Britain had been recruited. At each of five annual waves, the main exosure, ast mental disorder, was derived from assessments of sychiatric morbidity as measured by the General Health Questionnaire-2. Increased cigarette smoking was derived from observations of number of cigarettes smoked and was defined by an increase of five or more er day relative to the revious calendar year. After logistic regression analysis, ersons with a common mental disorder were about 30% more likely to have increased their cigarette smoking over the revious year (odds ratio =.29, 95% confidence interval:.6,.43). The estimated effect in the youngest (6 2 years) and oldest (5 75 years) age grous was higher than that in the middle (3 50 years) age grou (odds ratios =.50,.57, and.2, resectively; test for interaction, χ 2 = 6.8 (3 df), = 0.078). These findings indirectly suort the hyothesis that common mental disorders may have an enduring effect of increasing cigarette smoking a year later. Am J Eidemiol 2000;52:65 7. logistic models; mental disorders; nicotine; rosective studies; sychiatry; smoking Desite extensive health education and taxation, the revalence of smoking in Great Britain remains high at 27 ercent (). Aroximately 3 ercent of the oulation, esecially teenagers, commence smoking annually (2, 3). Smoking is the largest reventable risk factor for mortality (4) and is also associated with sychiatric disorders (5). It has often been suggested that some smokers start and maintain smoking to self-medicate against deressive symtoms, as the sychoactive effects of nicotine hel to elevate their mood (6). Smokers resistant to quitting are more likely to be nicotine deendent (7). Major deressive disorders are twice as common in smokers than in nonsmokers (8) and are associated with nicotine deendence in young adults (9). Furthermore, cross-sectional associations between symtoms of deression and smoking have been found for women (0, ), teenagers (2), and community adult oulations (8, 3). Received for ublication Aril 20, 999, and acceted for ublication November, 999. Abbreviation: GHQ-2, General Health Questionnaire-2. Deartment of Psychological Medicine, Guy s, King s and St. Thomas School of Medicine, London, England, United Kingdom. 2 Centre for Poulation Studies, London School of Hygiene and Troical Medicine, London, England, United Kingdom. 3 Deartment of Eidemiology and Poulation Health, London School of Hygiene and Troical Medicine, London, England, United Kingdom. Corresondence to Dr. Khalida Ismail, Deartment of Psychological Medicine, Guy s, King s and St. Thomas School of Medicine, 03 Denmark Hill, London, England, United Kingdom SE5 8AZ ( khalida.ismail@io.kcl.ac.uk). Psychiatric disorders are also associated with a times increase in mortality in the general oulation, which cannot all be exlained by unnatural causes of death, such as suicide (4 6). This excess mortality due to natural causes in eole with sychiatric disorders could be artly exlained by the association between sychiatric disorders and unhealthy lifestyles, such as smoking (7). It has been suggested that sychiatric disorders increase cigarette smoking (8), but the evidence for this association has yet to be established (9). Findings from three US studies that have rosectively examined the association between smoking and deression have been inconsistent (20 22). Imortantly, none of these studies controlled for imortant sociodemograhic confounders. To the best of our knowledge, the contribution of mental disorders to smoking behaviors has not been examined rosectively in the British oulation, which may differ from US oulations regarding smoking behaviors. In the resent study, we used rosectively collected data on smoking, sychiatric morbidity, and relevant confounders from a large community cohort to examine the hyothesis that common mental disorders are associated with increased cigarette smoking at 2 months follow-u. MATERIALS AND METHODS British Household Panel Survey We used data from the British Household Panel Survey (23), a longitudinal anel survey of rivate households in Great Britain (excluding Northern Ireland) that was started 65 Downloaded from htts://academic.ou.com/aje/article-abstract/52/7/65/7594 on 29 January 208

2 652 Ismail et al. in 99. It was designed to measure social, economic, and health behaviors of ersons reresentative of the British oulation residing in rivate households. A anel is a samle of subjects for whom the same variables are measured reeatedly over several oints in time, usually saced equally. In our study, a set of social, economic, and health variables was measured at baseline (wave ) and then annually for 4 more years (waves 2 5). Hence, a series of reeated measurements of several variables was available for each erson; the number of reeated measurements deended on the number of waves in which the erson was interviewed successfully. Smoking and mental health status data from revious waves were transferred to the succeeding wave, so that changes in cigarette smoking from one year to the next could be calculated. Households were initially selected for inclusion in the anel survey by using a two-stage stratified samling rocess. The samling frame used was the Postcode Address File for Great Britain (excluding Northern Ireland). This frame is widely used in large government surveys (24). During the first stage of samling, 250 ostcode sectors were randomly selected, with the robability of selection roortional to the size of the sector. Each ostcode sector selected was called a rimary samling unit and contained on average 2,500 delivery oints (equivalent to addresses). During the second stage, an average of 33 delivery oints were randomly selected from each rimary samling unit, with equal robability for selection. If one delivery oint included more than three households, such as more than three flats in one building, then three households were randomly selected from the total number available. A rivate household was defined as one erson living alone or a grou of eole who shared either living accommodations or one meal a day. Household members had to reside there for at least 6 continuous months a year so we could avoid recruiting students who lived at their arents homes during vacations. Nonresidential addresses and institutions were excluded. All adults aged 6 75 years residing in each household selected were eligible for entry into the survey at baseline, wave, and were defined as the original samle members. The samle for wave 2 and all subsequent waves consisted of all original samle members in all households identified in wave. In addition, new eligibility for the samle could occur in wave 2 and beyond if an original samle member moved into a household with one or more new eole or one or more new eole moved in with an original samle member. All new adults identified in the original samle member s household in subsequent waves were thus recruited into the study in that wave and were followed u in successive waves as long as they were still residing with the original samle member. Each successive wave thus consisted of all original samle members lus a small roortion of new members. Measures Evidence of common mental disorder was measured at each wave of the British Household Panel Survey by using the self-administered General Health Questionnaire-2 (GHQ-2) (25). This questionnaire was devised for screening those mental disorders common in the general oulation, redominantly deressive and anxiety disorders, and scores the subject s current mental health in relation to his or her usual state. A of 3 (out of 2) is defined as the cutoff oint at which a resondent is classified as suffering from a mental disorder (or is GHQ-2 ositive). Although a secific diagnosis cannot be made by using the GHQ-2 alone, this questionnaire has a sensitivity and secificity of 89 and 80 ercent, resectively, for a sychiatric diagnosis at this cutoff oint (25). The exosure of interest in this study, ast mental disorder, was defined as being GHQ-2 ositive in the revious wave and was recorded in waves 2 5. Information on smoking status consisted of the binary indicator current smoker or nonsmoker and the number of cigarettes smoked er day. Change in cigarette smoking was calculated as the difference in the number of cigarettes smoked between successive waves and therefore could be defined for waves 2 5 only. Increased cigarette smoking, the main outcome, was defined as resent if the subject was smoking at least five cigarettes er day more than during the revious wave, and this information was generated for waves 2 5. We considered this increase the minimum associated with an increase in hysical health risk. The definition of increased cigarette smoking included both new and regular smokers, and new smokers were defined as current smokers who were nonsmokers in the revious wave. As there could not be a measure of GHQ-2 or number of cigarettes smoked before wave, there was no measure of ast mental disorder or increased cigarette smoking for wave. Hence, the main analyses concerned changes in smoking status as recorded in waves 2 5. Whenever smoking was not recorded at a articular wave, the outcome indicator for the next wave was defined as missing. Likewise, whenever mental status had not been assessed at a articular wave, the exosure of interest for the next wave was defined as missing. Other variables were investigated as otential risk factors and otential confounders of the association between ast mental disorder and increased cigarette smoking. Included were sex; age at interview, categorized into the four age bands of 6 2, 22 30, 3 50, and 5 75 years; socioeconomic status, derived from the Registrar-General classification of current or revious occuational status (26) (if unknown, the occuational status of the head of household was recorded); education, classified as having gained university qualifications usually from the age of the early twenties onward, college qualifications (Advanced level/higher National Diloma) usually at age 8 years, secondary or high school qualifications (Certificate of School Education, General Certificate of School of Education, Ordinary level) usually at age 6 years, or no academic qualifications; marital status, classified as married/cohabiting, divorced/searated, widowed, or never married; current alcohol/drug roblems, defined as resent or absent; and hysical health, classified as the number of hysical roblems reorted relating to limbs, sight, hearing, skin, chest, heart, stomach, eilesy, and diabetes. National Oinion Polls interviewers assessed all variables obtained by self-reort from study Am J Eidemiol Vol. 52, No. 7, 2000 Downloaded from htts://academic.ou.com/aje/article-abstract/52/7/65/7594 on 29 January 208

3 Common Mental Disorders and Cigarette Smoking 653 articiants. Many of these variables may have changed over time; for examle, subjects who were married when they joined the anel might have declared at later waves that they were divorced. These changes were taken into account in the analyses by using the s recorded at the wave being studied. Statistical methods The data consisted of five successive waves of observations. Stata statistical software was used (version 5; Stata Cororation, College Station, Texas). For each erson, data from waves 2 5 were aended to those for wave. First, the distribution of resondents characteristics at entry into the anel was examined, and their baseline smoking revalence was studied. Second, multivariate analyses were erformed to study the association between increased cigarette smoking and ast mental disorder, including other otential risk factors and confounders. Records for waves 2 5 were analyzed jointly by using logistic regression models. Survival analysis was not indicated because the outcome of interest was whether there was an increase in cigarette smoking after a fixed 2 months, not the amount of time that elased before cigarette smoking increased. The results were resented as odds ratios with 95 ercent confidence intervals. Likelihood ratio tests were used to evaluate the significance and linear trend of risk factors and interactions between them. The likely correlation between records obtained on the same ersons (and on the same households) at different waves was taken into account by comuting the 95 ercent confidence intervals for the odds ratios with the Huber method (27). This method gives results similar to those obtained with generalized estimating equations by using an identity correlation matrix (28). Finally, to examine the quality of the British Household Panel Survey data, we defined an indicator of articiation in the last wave, since this wave should have included all anel members. Possible resence of attrition bias was then assessed by comaring the results obtained for the subset of subjects who did or did not articiate in the last wave. RESULTS A total of 2,057 subjects were recruited between 99 and 995, 9,598 in the first wave and 2,459 in subsequent waves as a result of household changes. The distribution of risk factors at the time of recruitment is shown in table. Forty-nine ercent were men (n 5,847), the mean (standard deviation) age of the cohort was 4.7 (6.2) years, 3.7 ercent were smokers, and 25.3 ercent were GHQ-2 ositive. The results of univariate analyses of the data at recruitment showed that the odds of current smoking were significantly greater for subjects who were young, male, of lower socioeconomic status, and divorced; had hysical or alcohol/ drug roblems; and were GHQ-2 ositive (table 2). A significant linear association was found between number of cigarettes smoked and increasing concurrent GHQ-2 score (correlation coefficient 0.29, < 0.000). Increased cigarette smoking was recorded as resent if the difference in the number of cigarettes smoked between successive waves was five or more. Past mental disorder was derived from the GHQ-2 in waves 4 but was used as an exosure for subsequent waves (waves 2 5, resectively). As indicated in nearly 6 ercent (n 2,05) of the records for these last four waves, resondents cigarette smoking had increased over the revious 2 months, and about a third of these subjects (n 75 records) had been nonsmokers in the revious wave. By examining successive records for the same subjects, we found that the robability of increasing cigarette smoking increased if a subject had already increased cigarette smoking before; those whose cigarette smoking had increased between waves and 2 (5.5 ercent) were about 20 ercent more likely to further increase their cigarette smoking between subsequent waves. Results obtained from alternative logistic regression models for increased cigarette smoking are shown in table 3. The model column shows the crude odds ratio for increased cigarette smoking by ast mental disorder. Adding the demograhic variables as indicator variables (age grou, sex, and socioeconomic status) significantly imroved model but did not modify the association between ast mental disorder and increased cigarette smoking. Adding marital status and education (model 2) and the health variables, as well as alcohol/drug roblems and hysical roblems (model 3), showed that each of these variables (with the excetion of hysical roblems) significantly imroved the models and was indeendently associated with increased cigarette smoking, but none confounded the effect of ast mental disorder on increased cigarette smoking. Interactions between exosure variables were examined and showed a weak modification of the effect of ast mental disorder for different age grous (likelihood ratio test: χ (3 df), 0.078). Stratification by age grou showed that the association between ast mental disorders and increased cigarette smoking was strongest for the youngest (6 2 years) and oldest (5 75 years) age grous, less significant for the grou aged years, and not significant for the grou aged 3 50 years (table 4). Accounting for the additional clustering of observations because some of the resondents resided in the same households did not change any of the results (not shown). To examine whether attrition bias affected our results, we reeated the analyses on the two subsets of resondents who were either resent (n 8,572) or not resent (n 3,485) in wave 5. Although those who did not articiate in wave 5 were more likely to be smokers (odds ratio.25, 95 ercent confidence interval:.5,.36) and to have had a ast mental disorder (odds ratio.4, 95 ercent confidence interval:.03,.25), the estimated association between ast mental disorder and increased cigarette smoking did not substantially differ between the two subsets (likelihood ratio test: χ ( df), 0.638). DISCUSSION Cross-sectional data from the first five waves of the British Household Panel Survey showed that at every wave, Am J Eidemiol Vol. 52, No. 7, 2000 Downloaded from htts://academic.ou.com/aje/article-abstract/52/7/65/7594 on 29 January 208

4 654 Ismail et al. TABLE. Distribution of all variables by the wave during which subjects entered the British Household Panel Survey in Variable Wave (99) No. % Wave 2 (992) No. % Wave 3 (993) Wave 4 (994) No. % No. % Wave 5 (995) All waves No. % No. % Sex Male Female 4,585 5, ,847 6, Age grou (years) ,02,854 3,758 2, ,062 2,58 4,23 3, Socioeconomic status Professional/managerial Skilled manual/nonmanual Partly skilled/unskilled Armed forces/unknown 2,632 4,237 2, ,22 5,56 2, Education University College Secondary school No qualifications 706,822 2,984 3, ,482 3,879 4, Marital status Married/cohabiting Divorced/searated Widowed Never married 6, , , , Alcohol/drug roblems Mentioned Not mentioned 44 9, ,967 Physical roblems None One Two or more 4,982 2,82, ,478 3,520 2, Current smoker No Yes 6,60 2, ,50 3, Current GHQ-2* status GHQ-2 ositive GHQ-2 negative 2,227 6, ,852 8, * GHQ-2, General Health Questionnaire-2. a quarter of the subjects suffered from a common mental disorder. We found that such ersons were about 30 ercent more likely to have increased their smoking 2 months later. The General Health Questionnaire is considered almost a gold standard for screening common mental disorders in the general oulation, as it has substantial reliability and very good sensitivity and secificity when tested against standardized sychiatric interviews (25, 29). The most common mental disorders in the general oulation are mixed deressive and anxiety disorders (30, 3). Anxiety disorders may have diluted the association we found, because deression is more strongly associated with smoking than is anxiety (8, 9). To limit recall bias, the GHQ-2 measures recent change in mental health comared with a erson s usual state. There may have been some confounding by sychiatric morbidity that might have been resent between waves but had remitted at the time of interview. It is difficult to examine for this tye of confounding because the majority of mental disorders in the general oulation aear to remit (32), but about 30 ercent have a chronic course (33). Smokers tend to underreort how much they smoke by 5 0 ercent, suggesting that the associations we observed are likely to be dilutions of the true association (34). Substance misuse is strongly associated with smoking (8), but the roortion of subjects in this study who reorted alcohol and drug roblems ( ercent) was much lower than the national revalence of alcohol and drug deendence, 7 and ercent, resectively (3), suggesting a high misclassification rate in this study. It is ossible that this misclassification introduced a bias in the association of ast mental disorder with increased smoking. The direction of any misclassification bias is likely to result in underestimation of the association between ast mental disorder and increased smoking, as underreorting of alcohol and drug roblems is robably correlated with underreorting of cigarette smoking. There is evidence that after adjustment for alcohol roblems, the association between common mental disorders and smoking remains (8, 9), as was found in our study; however, adjustment for a misclassified confounder is not sufficient to remove the effects of bias regarding its measurement. The most vulnerable age grou aears to be young adults; subjects aged 6 2 years (including new smokers) were 50 ercent more likely to have increased their cigarette Am J Eidemiol Vol. 52, No. 7, 2000 Downloaded from htts://academic.ou.com/aje/article-abstract/52/7/65/7594 on 29 January 208

5 Common Mental Disorders and Cigarette Smoking 655 TABLE 2. Univariate logistic regression analysis of the current smoking status of all subjects at entry into the British Household Panel Survey in Exlanatory variable Current smokers % No. OR* 95% CI*, Wald test Current GHQ-2* status GHQ-2 negative GHQ-2 ositive Sex Female Male Age grou (years) Marital status Married/cohabiting Divorced/searated Widowed Never married Socioeconomic status Professional/managerial Skilled manual/nonmanual Partly skilled/unskilled Armed forces/unknown Education University College Secondary school No qualifications Alcohol/drug roblems Mentioned Not mentioned Physical roblems None One Two or more ,409 2,852 6,20 5,847 2,062 2,58 4,23 3,83 7, ,336 3,22 5,56 2, ,482 3,879 4,74, ,478 3,520 2, ,.64.04,.2.26,.62.09, , , , ,.5.29,.57.78, , ,.80.65, , , ,.22.0,.25 * OR, odds ratio; CI, confidence interval; GHQ-2, General Health Questionnaire-2. Comuted by using the Huber method (27) for correlated data for ersons. < <0.000 < < <0.000 <0.000 <0.000 <0.000 <0.000 <0.000 < smoking if they had had a mental disorder in the revious year. The uer confidence limit showed an estimated doubling of this risk. Subjects in the oldest age grou, 5 75 years, were 60 ercent more likely to have increased their smoking. Peole in this age grou are making transitions into retirement, bereavement, and the onset of hysical morbidity, and these may be reflected by the GHQ-2. We seculate that the smaller effect of recent common mental disorders and increased smoking in the grous of subjects aged 2 30 and 3 50 years may reflect their greater health consciousness. The advantage of our study is that we used anel data that combined cross-sectional data into a longitudinal design. Panel designs are esecially suitable for studying mental disorders, as the recise date of onset is often difficult to define. We also used a very large samle, so random error was likely to be small. Our associations did not aear to be invalidated by attrition bias. To the best of our knowledge, this is the first British rosective study demonstrating that common mental disorders (redominantly mixed anxiety and deression) affect the risk of increased smoking a year later. This finding indirectly suorts the self-medication hyothesis that nicotine, a stimulant, is used to relieve deressive symtoms (6). Other mechanisms may also be oerating. For examle, a large, female twin study found that the association between smoking and deression could be exlained by a common genetic redisosition to both (35). A third ossibility is that smoking, and other unhealthy lifestyles, can be viewed as a continuum of self-destructiveness, the extreme end of which is suicide (7). Deressed ersons have low selfesteem, are essimistic about the future, and lose interest in their health. It has been ostulated that nicotine is itself deressogenic (36). Although we could not test this theory directly, revious smoking may be associated with increased Am J Eidemiol Vol. 52, No. 7, 2000 Downloaded from htts://academic.ou.com/aje/article-abstract/52/7/65/7594 on 29 January 208

6 656 Ismail et al. TABLE 3. Multivariate logistic regression analysis of increased smoking, by ast mental disorder and other exlanatory variables, for subjects who entered the British Household Panel Survey in Exlanatory variable OR Model 95% CI, Model 2* Model 3 OR 95% CI OR 95% CI Past mental disorder GHQ-2 negative GHQ-2 ositive Sex Female Male Age grou (years) Marital status Married/cohabiting Divorced/searated Widowed Never married Socioeconomic status Professional/managerial Skilled manual/nonmanual Partly skilled/unskilled Armed forces/unknown Education University College Secondary school No qualifications Alcohol/drug roblems Not mentioned Mentioned Physical roblems None One Two or more.37.23, , , , , , , , ,.3.05,.46.20,.76.24,.80.2, , , 3.63 * Likelihood ratio test of model 2 against model : χ 2 (df) = 56. (3) < Likelihood ratio test of model 3 against model 2: χ 2 (df) = 9.3 (3) OR, odds ratio; CI, confidence interval; GHQ-2, General Health Questionnaire-2. Comuted by using the Huber method (27) for correlated data for ersons , , , , , , , ,.2.05,.46.20,.75.2,.77.2, , , , , ,.30 TABLE 4. Age-secific adjusted* odds ratios for increased smoking, by ast mental disorder, British Household Panel Survey, Past mental disorder GHQ-2 negative GHQ-2 ositive OR.49.3,.98 Age (years) % CI OR 95% CI.29.03, OR 95% CI , OR % CI.63.30, 2.03 <0.000 * Interaction between age and ast mental disorder, likelihood ratio test: χ 2 (df) = 6.8 (3), = Adjusted for age, socioeconomic status, education, marital status, and alcohol/drug roblems. OR, odds ratio; CI, confidence interval; GHQ-2, General Health Questionnaire-2. Am J Eidemiol Vol. 52, No. 7, 2000 Downloaded from htts://academic.ou.com/aje/article-abstract/52/7/65/7594 on 29 January 208

7 Common Mental Disorders and Cigarette Smoking 657 General Health Questionnaire scores, which then further increases cigarette smoking. In the general oulation, sychiatric morbidity is common and treatable. Our study suggests that a subgrou of ersons who have significant mental health roblems are more likely to increase their cigarette smoking and that this behavior is most likely to occur in the youngest age grous, who are at risk of becoming lifetime smokers. This finding suggests that smokers, esecially younger and older adults, may benefit from increased hel for their sychological distress, which may imrove both their smoking behaviors and, consequently, their hysical morbidity and mental health. ACKNOWLEDGMENTS Dr. Ismail was suorted by a Wellcome Trust Training Fellowshi in Clinical Eidemiology (grant AA27) in , when this study was started. The data used were made available through the Economic and Social Research Council Data Archive. The ESRC Research Centre originally collected the data on microsocial change at the University of Essex, England, United Kingdom. Neither the original collectors of the data nor the Archive bear any resonsibility for the analyses or interretations resented in this aer. Dr. Ismail was the rincial investigator, generated the study hyothesis and design, initiated the study, undertook data management and statistical analysis, and drafted the manuscrit. Mr. Sloggett and Dr. De Stavola contributed to the develoment of the hyothesis, design, and analysis and heled reare the manuscrit. The authors thank Drs. Jan Neeleman, Mike Farrell, and Nick Buck, who commented on an earlier draft of this manuscrit. REFERENCES. Office of Poulation Censuses and Surveys. General Household Survey 994. London, England: HMSO, Buck N, Gershuny J, Rose D, et al. Changing households: the British Household Panel Survey Colchester, England: ESRC Research Centre on Microsocial Change, Office of Poulation Censuses and Surveys. Smoking among secondary school children in England in 994. London, England: HMSO, Peto R, Loez AD, Boreham J, et al. Mortality from tobacco in develoed countries: indirect estimates from national vital statistics. Lancet 992;339: Hughes JR, Htasukami DK, Mitchell JE, et al. Prevalence of smoking amongst sychiatric outatients. Am J Psychiatry 986;43: Glass RM. Blue mood, blackened lungs: deression and smoking. JAMA 990;264: Fagerstrom KO, Kunze M, Schoberberger R, et al. Nicotine deendence versus smoking revalence: comarisons among countries and categories of smokers. Tob Control 996;5: Glassman AH, Helzer JE, Covey LS, et al. Smoking, smoking cessation, and major deression. JAMA 990;264: Breslau N, Kilbey MM, Andreski P. Nicotine deendence, major deression, and anxiety in young adults. Arch Gen Psychiatry 99;48: Lee DJ, Markides KS. Health behaviors, risk factors, and health indicators associated with cigarette use in Mexican Americans: results from the Hisanic HANES. Am J Public Health 99;8: Chetwynd J. Some characteristics of women smokers. N Z Med J 986;99: Patton GC, Hibbert M, Rosier MJ, et al. Is smoking associated with deression and anxiety in teenagers? Am J Public Health 996;86: Anda RF, Williamson DF, Escobedo LG, et al. Deression and the dynamics of smoking: a national ersective. JAMA 990; 264: Huert FA, Whittington JE. Symtoms of sychological distress redict 7-year mortality. Psychol Med 995;25: Kouzis A, Eaton WW, Leaf PJ. Psychoathology and mortality in the general oulation. Soc Psychiatry Psychiatr Eidemiol 995;30: Murhy JM, Monson RR, Olivier DC, et al. Affective disorders and mortality. A general oulation study. Arch Gen Psychiatry 987;44: Neeleman J, Wessely S, Wadsworth M. Predictors of suicide, accidental death, and remature natural death in a generaloulation birth cohort. Lancet 998;35: Hiisley-Cox J, Fielding K, Pringle MSO. Deression as a risk factor for ischaemic heart disease in men: oulation based case-control study. BMJ 998;36: Lewis G. Continuum of self-destructiveness. Lancet 998;35: Kandel DG, Davies M. Adult sequelae of adolescent deression. Arch Gen Psychiatry 986;43: Aneshensel CS, Huba GJ. Deression, alcohol use, and smoking over one year: a four-wave longitudinal causal model. J Abnormal Psychol 983;92: Breslau N, Kilbey MM, Andreski P. Nicotine deendence and major deression. New evidence from a rosective investigation. Arch Gen Psychiatry 993;50: Taylor MF, ed. British Household Panel Survey user manual. Vol A. Introduction, technical reort and aendices. Colchester, England: University of Essex, Wilson P, Elliot D. An evaluation of the Postcode Address File as a samling frame and its uses with OPCS. J R Stat Soc (A) 987;50: Goldberg D, Williams P. A user s guide to the General Health Questionnaire. Windsor, England: NFER-NELSON, Office of Poulation Censuses and Surveys. Standard occuational classification. Vol 3. Social classifications and coding methodology. London, England: HMSO, Huber PJ. Robust estimation of a location arameter. Ann Math Stat 964;35: Liang KY, Zeger S. Longitudinal analysis using generalized linear models. Biometrika 986;73: McDowell I, Newell C. Measuring health. New York, NY: Oxford University Press, Goldberg D, Huxley P. Common mental disorders: a bio-social model. London, England: Routledge, Meltzer H, Gill B, Petticrew M, et al. The revalence of sychiatric morbidity among adults living in rivate households. London, England: HMSO, Brodaty H, Andrews G. Brief sychotheray in general ractice: a controlled rosective intervention trial. Br J Psychother 983; 43: Mann A, Jenkins R, Belsey E. The 2-month outcome of atients with neurotic illnesses in general ractice. Psychol Med 98;: Office of Poulation Censuses and Surveys. Health Survey for England (995). London, England: HMSO, Kendler KS, Neale MC, Maclean CJ, et al. Smoking and major deression. A causal analysis. Arch Gen Psychiatry 993;50: Hall SM, Munoz RF, Reus VI, et al. Nicotine, negative affect, and deression. J Consult Clin Psychol 993;6:76 7. Am J Eidemiol Vol. 52, No. 7, 2000 Downloaded from htts://academic.ou.com/aje/article-abstract/52/7/65/7594 on 29 January 208