Design Population-based retrospective cohort study.

Mortality Following Inpatient

Addictions Treatment

Role of Tobacco Use in a Community–Based Cohort

Leigh Gomez-Dahl; Thomas E. Kottke, MD; Robert M. Morse, MS; L. Joseph Melton III, MD

Objective –To determine the impact of tobacco- and alcohol-related deaths on overall mortality following inpatient treatment for alcoholism and other non nicotine drugs of dependence.

Design –Population-based retrospective cohort study.

Setting –Olmsted County, Minnesota (The Rochester Epidemiology Project), and the inpatient addiction Program (IAP) at Mayo Clinic, Rochester.

Patients –All 485Olmsted County residents admitted to an inpatient addiction program for treatment of alcoholism and other non nicotine drugs of dependence during the period 1972 through 1983.

Methods –Patients were followed up through the medical record linkage system of the Rochester Epidemiology Project through December 1994 to obtain vital status, and death certificates were obtained for those who died. The underlying cause of death was classified as alcohol related, tobacco related, both, or neither based on the classification from the Center for Disease Control and Prevention. The observed number of deaths by underlying cause was compared with the expected number using cause-specific 1987 death rates for the white population of the United States. All-cause mortality was also compared with that expected for persons in the West North Central Region of the United States of like age, sex, and year of birth. University and multivariate assessments were made to identify predictors of all-cause morality from baseline demographic information.

Results –At admission, the mean (SD) age of the 845 patients was 41.4 (14.5) years, and 35% were women. Altogether, 78% had alcohol as their only non nicotine drug of dependence and 18% had alcohol and other non nicotine drugs of dependence, while 4% were classified as having a nonalcohol, non nicotine drug dependence alone. At admission, 75% were current and 8% were former cigarette smokers, 3% were current cigar or pipe

smokers, and 2% were current users of smokeless tobacco. Follow-up after the index IAP admission totaled 8913 person-years (mean [SD] of 10.5 [5.6] years per patient). Death certificates were obtained for 96% (214) of the 222 patients who died. Of these 214 deaths, 50.9% (109) had a

tobacco-related and 34.1% (73) had an alcohol-related underlying cause (p<.001) and male sex (P<.001).

Conclusions –Patients previously treated for alcoholism and/or other non nicotine drug dependence had an increased cumulative mortality that was due more to tobacco-related than to alcohol-related causes. Nicotine dependence treatment is imperative in such high-risk patients.

(JAMA 1996;275:1097-1103)

From the Nicotine Dependence Center (Drs Hurt and Croghan), Section of Biostatistics (Mr Offord and Ms Gomez-Dahl), Department of Health Sciences Research (Drs Kottke and Melton), and Department of Psychiarty and Psychology (Dr Morse), Mayo Clinic, Rochester, Minn.

Reprints: Rchiard D. Hurt, MD, Mayo clinic, 200 First St SW, Rochester, MN 55905

ADDICTIVE drug death account for one fourth to one third of all deaths in the United States, with the number of tobacco-related deaths being several times greater than those caused by alcohol. There is a well-known association between cigarette smoking and alcohol consumption, with the heaviest smokers also being the heaviest drinkers and vise versa. The prevalence of smoking among substance abusers is two to three times that of the general population and alcoholics may constitute a quarter of all smokers. Indeed, the relationship is so strong that intractable heavy smoking is a predictor of unrecognized alcohol abuse. Despite this, efforts at smoking cessation receive little attention in most alcohol treatment programs. If smoking contributes substantially to the morality following treatment for alcoholism, then interventions aimed at nicotine dependence should be considered by the

treatment community.

A key outcome of alcoholism treatment is morality, but most assessments of death following treatment for alcoholism have limited sample sizes or short follow-up, and few have addressed the impact of tobacco-related diseases. For example, in a 20-year follow-up study of 99 men treated for alcoholism, 44 men had died, 29% due to circulatory disease and 25% due to lung cancer, for an observed/expected mortality ratio of 3.6. Smoking and alcohol abuse may be independently associated with coronary death, but controlling for smoking status may account for the increased risk of lung cancer among alcoholic veterans.

However, most studies have not considered the contribution that smoking makes to mortality in alcoholics. Even in more recent years, tobacco use and the diseases it causes have been conspicuously absent in the assessment of predictors of mortality after alcoholism treatment despite the inclusion of almost every other possible factor. Thus, there is a need for more definitive information on whether there is an excess of tobacco-related mortality in patients treated for alcoholism and/or other drug

deaths to overall mortality in cohort of Olmsted County, Minnesota, residents admitted for inpatient treatment of alcoholism and/or other non nicotine drug dependencies. The report is intended to heighten awareness of the effect of nicotine dependence on the outcomes of treatment programs for alcohol and other drug dependence.

There currently is considerable controversy in the addictions treatment field about this issue, and there is resistance to initiating nicotine dependence treatment for individuals in treatment for alcoholism or other drug dependence. We believe the sheer magnitude of the problem as reported herein will sensitize the field to the critical need for some type of intervention for nicotine dependence in patients with other addictive disorders. Information on morbidity will be the subject of a separate report.

Methods

The inpatient Addiction Program (IAP) opened in late 1972 as the main source of inpatient treatment for alcoholism and other drug dependence for patients at the Mayo Clinic in Rochester, Minn. The program, directed by a psychiatrist and staffed by resident physicians, psychologists, counselors, nurses, social workers, and other support staff, is based on the “Minnesota model” for intensive treatment. On admission each patient undergoes a comprehensive psychiatric and medical examination, a psychological assessment, and an interview by an addictions counselor. As with other Mayo programs, the patient census reflects the regional nature of the practice, with about 25% of patients from the remainder of Minnesota, 45% from other states, mostly in the upper Midwest, and 2% or fewer from foreign countries.

Following approval by the Mayo Institutional Review Board, we identified all 845 Olmsted County residents admitted to the IAP for the first time from 1972 through 1983 and obtained vital status follow-up through December 1994. We obtained basic

demographic information as of the first IAP admission for each subject and medical and psychiatric diagnoses prior to, at the time of, and subsequent to this admission. The type and pattern of tobacco and other substance use were also recorded.

We were able to abstract medical findings for these patients because Mayo Clinic maintains a unit medical record system jointly with its two affiliated hospitals (Saint Mary’s and Rochester Methodist) and because the Roadsters Epidemiology Project provides a linkage with the inpatient and outpatient medical records of the other

providers in the community. These include the Olmsted Medical Group and its affiliated Olmsted Community Hospital, the Rochester State Hospital, the University of Minnesota and Department of Veterans Affairs hospitals in Minneapolis, the small hospitals in surrounding counties, and the few solo medical practitioners in Rochester.

The result is the linkage of medical records from essentially all sources of medical care available to and utilized by the Olmsted County population. Information on smoking and disease occurrence was not restricted to that obtained at the time of admission to the IAP

but included the entire medical care experience of all community health care providers as recorded in the medical records. These records included inpatient, outpatient, surgical, and emergency care.

Because the Rochester Epidemiology Project routinely obtains death certificates for residents of Olmsted County, they were already available for many subjects who had died. In addition, we ordered death certificates for those who had died elsewhere. The underlying cause of death and all other findings on the death certificates were coded by an expert coder external to our investigative group and entered into the data set for analysis using the coding system of the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9) In addition, a physician (R.D.H.) verified all the death certificate information. Specific causes of death may be related to tobacco, alcohol, both, or neither. To avoid any potential bias in this assignment, we used the schema developed by the Centers for Disease Control and Prevention (CDC) to categorize deaths as tobacco related or alcohol related McNewmar’s test for correlated proportions was used to test whether alcohol-related diseases accounted for more deaths.

One set of analyses compared the observed and expected number of deaths by the IDC-9-coded underlying cause. The expected number of deaths were based on a person-years analysis using cause-specific death rates for the US white population in 1987. For each observed cause of death, we computed the expected number of deaths based on the U.S. age and sex specific rates for 1987, treating the ICD-9-codes alone or appropriately grouped. Relative risk (RR) estimates and 95% confidence intervals (CIs) were

computed. If the lower limit of a 95% CI exceeded 1, this reflected excess mortality from the underlying cause, whereas an upper limit below 1 reflected reduced mortality. The relatedness to alcohol or smoking was based on published reports from the CDC using ICD-9-codes.

In another set of analyses, we estimated the all-cause cumulative mortality in this cohort and compared it with that expected for persons in the general population of the West North Central Region of the United States of like age, sex, and year of birth using the one-sample log rank test. To identify predictors of mortality, we employed multivariate assessments using the Cox proportional haard model. For this, we started with the set of all potential risk factors and eliminated those that did not contribute significantly to the prediction. The candidate predictors included age at admission, sex, years of education (<12, 12, 13 to 15, or >16) or type of non nicotine addiction (alcohol only, alcohol and another drug, another drug only), cigarette smoking status at admission (current, former, or never smoker), and year of admission.

University evaluations were made using the log rank test. In addition (alcohol only, alcohol and another drug, another drug only), cigarette smoking status at admission (current, former, or never smoker), and year of admission. Unvaried evaluations were made using the log rank test. In addition, for multivariate assessment of predictions of mortality, each patient’s expected 0-year survival was used as a covariate to adjust for age and sex differences in mortality that would be expected because of their known

influence on mortality in the general population. All reported P values are two-tailed, with values <.05 used to indicate findings not attributable to chance.

Results

Altogether, 845 Olmsted County residents were admitted for the first time to the IAP from 1972 through 1983. Table 1 shows the demographic features of this cohort at the time of admission. Their mean (SD) age was 41.4 (14.5) years (range 17 to 82 years); 65% were men and 35% were women. There were nine minority patients (1%), with the balance being white.

At admission, most of the patients (78%) were diagnosed with alcoholism as their only non nicotine drug of dependence, while 18% were characterized as having alcoholism and another non nicotine drug dependence; only 4% were classified as having a non alcohol, non nicotine drug dependence alone. Table 1 also shows the cigarette smoking status at admission, with 75% current and 8% former smokers.

For those 647 patients with sufficient information to calculate pack-years of cigarette smoking (attributing 0 pack=years to never smokers), the median pack-years of smoking was 20; the median was 25 pack-years for the 559 ever smokers. Only 3% were current cigar or pipe smokers, while 2% were current users of smokeless tobacco. The entire cohort was followed up after the index IAP admission for a total of 8913 person-years (mean [SD], 10.5 [5.6] years, range 0.1 to 21.8 years per patient). During this period of observation, 222 patients died; the mean (SD) time from admission to death was 8.3 (5.3) years.

We were able to obtain death certificates for 96% (214) of these 222 deaths. The average (SD) follow-up in the 623 survivors was 11.2 (5.5 years). Only 6% (37) of the survivors were followed up for less than 1 year; 11% (69) were followed up for 1 to 5 years, 18% (112) for 6 to 10 years, and 65% (405) for 10 years or more. Table 2 shows the summary cause of death information according to alcohol and/or tobacco relatedness. In addition to the observed number of deaths from specific causes, the expected number of deaths is presented among with accompanying RR estimates and the associated 95% CIs. The relatedness to alcohol or tobacco is based on published reports using ICD-9 codes from the CDC. Tobacco-related causes of death accounted for 50.9% of deaths (109) and were twice as frequent as expected for the general population. Alcohol-related conditions accounted for 34.1% of deaths (73) and were more than four times the number expected. The proportion of deaths from tobacco-related causes significantly exceeded that from alcohol-related conditions (P<.001).

When the deaths were sub-classified into four groups based on whether the cause of death was alcohol or tobacco related, there was 63 deaths from alcohol-related but not tobacco-related causes, more than four times the number expected. Conversely, there were 81 tobacco- but not alcohol-related deaths, for an RR of 1.8. Of note is that 78 deaths were due to neither alcohol-nor tobacco-related causes, more than four times the number

expected. Table 3 further delineates the causes of death for all major ICD-9 groupings and for selected sub-categories. For the 214 patients for whom we had death certificates, the observed numbers of death by causes are ranked in order from most to least frequent. Some individual causes of death that are related to alcohol or tobacco are not shown if no deaths were observed for these categories.

Within the major groupings, it was not surprising to find that cardiovascular diseases and chronic liver diseases were so frequent a cause of death but of particular note are the 17 patients who died of chronic obstructive pulmonary disease, five of who died of

pancreatic cancer, and 13 who died from suicide. Though 13 died of lung cancer, this was not significantly greater than the expected number of eight (P>.05).

The frequency of all death certificate findings, including not only the underlying cause of death but also the other diagnosis listed, was the presence of coronary artery disease and alcoholic liver disease as the top two findings on the death certificates – this is not surprising given the previously cited literature.

However, it is of note that chronic obstructive pulmonary disease and lung cancer were frequently mentioned. A total of 16 patients had lung cancer mentioned somewhere on their death certificates. The cumulative mortality in this cohort compared with that expected among West North Central US residents of like age, sex, and year of birth. The cumulative mortality experience significantly exceeded that expected (P<.001). The observed mortality was 8.7% vs an expected 3.3% at 5 years, 18.5% vs 7.4% at 10 years, 32.0% vs 12.5% at 15 years, and 48.1% vs 18.5% at 20 years.

Among the variables assessed invariantly, mortality differed significantly (P=.005) according to type of addiction to a non alcohol drug only, the RR (95% CI) for those with an addiction to both alcohol and another drug was 1.45 (0.49 to 4.24). For those with only addiction to alcohol, the RR was 2.66 (95% CI, 0.99 to 7.16). We further compared the subgroup of 659 patients with alcohol as their only non nicotine addiction against the subgroup of 154 patients with both alcohol and another non nicotine drug dependence. Those with alcohol only had a significantly worse mortality (RR, 1.84; 95% CI, 1.16 to 2.92) than those with both (P=008). Male sex (RR, 1.39; 95% CI, 1.04 to 1.86) was also related to higher mortality (P=.03).

Mortality also differed significantly across age group (P<.001). When we used as a reference group age below 40 uears, the RR for those 40 to 59 years old was 3.93 (95% CI, 2.62 to 5.19). For those aged 60 years and older at admission, the RR was 12.57 (95% CI, 9.19 to 19.29). Mortality also differed (P=02) across calendar period. When we used as the reference group those admitted most recently, in years 1980 through 1983, those admitted in the period through 1976 through 1979 had worse mortality (RR, 1.70; 95% CI, 1.17 to 2.47), while for those admitted from 1972 through 1975 it was intermediate (RR, 1.36; 95% CI, 0.92 to 2.00). Years of education was also related to mortality

(P=.01). When we used as a reference group those with at least 4-year college degree, the group with some college had an RR of 0.79 (95% CI, 0.48 to 1.28); high school graduates

had an RR of 1.24 (95% CI, 0.79 to 1.96), while for those who did not graduate from high school, the RR was .41 95% CI, 0.89 to 2.26).

The 659 patients with alcohol only as their primary non nicotine addiction had

significantly worse (P<.001) cumulative mortality (5-and 10-year cumulative mortality of 9.2% and 20.7%) than expected (5- and 10-year of 3.8% and 8.6%). This was also

evident for the 154 patients with both alcohol and another non nicotine drug of

dependence (observed 5- and 10- year of 3.1% and 3.1% vs 5- and 10-year of 2.24% and 4.9%, P = .105).

We failed to detect significant mortality differences according to cigarette smoking status at admission. There was due in part to the limited number of never smokers and the general predominance of smoking in this cohort. Furthermore, compared with expected mortality, the observed mortality was significantly higher in never smokers (observed 5- and 10-year cumulative mortality of 1.2% and 9.8% vs expected 2.9% and 6.5%, P=.01), former smokers (observed 5- and 10-year of 10.8% and 22.4% vs expected 6.8 and 14.8%, P=.007), and current smokers (observed 5- and 10-year of 7.6% and 16.8% vs expected 2.8% and 6.4% P<.001).

COMMENT

The major finding in this study is that tobacco-related diseases are the leading cause of death in patients previously treated for alcoholism and/or other nonnicotine drug dependence. This has compelling implications for the addictions treatment community. This information should alert professionals in the filed that the treatment of nicotine dependence is imperative in these high-risk patients. Although nicotine is increasingly considered a drug of dependence, most practice settings allow patients undergoing treatment for alcoholism or other nonnicotine drug dependencies to continue smoking. This continues despite the demonstration that such patients can be treated for nicotine dependence while in treatment for other addictive disorders without interfering with subsequent abstinence from alcohol and other nonnicotine drugs of dependence. Recently, there has been a heightened awareness of the need to consider nicotine as a drug of dependence in the addictive disorders treatment setting.

As a result, a growing number of addiction treatment programs now offer nicotine dependence treatment as either an optional or a required part of their program, and a systematic methodology for treatment programs to follow has been developed. The best treatment model for nicotine dependence in patients with other addictive disorders has not been completely defined as yet, but the incorporation of nicotine dependence in the addiction treatment model is one viable option. Other options include addressing nicotine dependence after patients have been treated for their chemical dependencies.

This might be acceptable for some patients, but only if it can be ensured that their nicotine dependence treatment will not be postponed indefinitely. While some have contended that from a therapeutic standpoint, social support is the most alterable risk

factor for preventing premature mortality in patients previously treated for alcoholism,24 we believe that nicotine dependence is by far the more important alterable risk factor and should be incorporated into the treatment milieu.

Some of the deaths that we observed were due both to alcohol- and tobacco-related causes, with related caused being predominant(P<.001). The extent of tobacco-related deaths among such patients has not been reported, principally due to the

limitations of previous studies with respect to sample size, selection bas, follow-up, or neglect of tobacco use as a factor. While previous studies have not focused on tobacco, there have been hints of the underlying tobacco-related disease burden, such as the observation that cardiac disease accounted for 34% and lung cancer 6% of deaths in a 20-year follow-up of 133 treated alcoholics.18 That study was limited by the small sample size and the loss of 30% of the subjects to follow-up. In a larger study with only 6 years of follow-up, the overall mortality for patients treated for alcoholism was four times higher than that of the general population, and cardiac diseases accounted for 25% of the deaths.16 Thus, our finding that coronary artery disease was the most frequent underlying specific cause of death is not new, but does provide more definitive results because of the larger sample size and longer follow-up.

Further evidence that smoking has an important influence on survival is our observation that chronic obstructive pulmonary disease was the underlying cause of death in 8% of patients and mentioned in the death certificate in 18%. This has not been previously reported and is particularly noteworthy in light of the young average age of cohort at admission. It is also of note that lung cancer was the underlying cause of death in 6% of patients and was mentioned on the death certificate in additional three patients. In some respects this could have been predicted, since the smoking rate among this alcoholic cohort was nearly 80%. However, we were surprised that the ER for lung cancer was not significantly elevated, through our findings are consistent with an RR as high as 2.7, and may not have reached significance e because of the sample size. There may be other explanations for this observation.

The cohort is young (the mean age at admission was 41 years), and we would expet more lung cancers to develop as these subjects are followed up to older ages. For example, the median age at diagnosis of lung cancer in Minnesota is 68.0 years for men and 67.0 years for women. Furthermore, there are other competing causes of death, so that the risk of dying of lung cancer may not be realized if patients die earlier from other conditions such as coronary artery disease, accidents, or suicide. The observed 6% of deaths from lung cancer in this cohort, with a mean follow-up of 10.5 years, was lower than the 25% reported in a 20-year follow-up of 99% alcoholic men. Finally, there may be individual differences in susceptibility to smoking-related diseases on which we have no

information. Although it is known to be a smoking-related cancer, the observation that carcinoma of the pancreas was such a frequent underlying cause of death in this cohort is also interest given the rarity of this cancer and the poor outcome associated with its occurrence.

As expected, alcohol-related diseases and conditions such as alcoholic liver disease, accidents, injury, and trauma were frequently listed as the underlying cause of death and also frequently mentioned on the death certificate as an additional finding. We also confirmed the substantial risk for suicide in patients with addictive disorders, as this accounted for 13 (5.9%) of the deaths in our cohort.

Addictions professionals are aware of this substantial risk factor for premature mortality in patients previously treated for alcoholism, but others may not be as alert to this possibility. Between 15% and 26% of suicides are thought to be in alcoholics, and 34% of men and 15% of women who attempt suicide have alcohol-related problems. We are unaware of any other reports of the frequency of suicide in a population-based study for patients treated for alcoholism and other nonnicotine drugs of dependence, though earlier case series did report high suicide rates in alcoholics. Though suicide is a rare event in the general population, with an incidence rate of only 12.5 per 100,000 person-years for residents of Olmsted County, the patients in our cohort are clearly at extremely high risk of suicide. This indicates the need for more research to assess factors related to suicide (e.g., concurrent psychiatric illnesses, state of abstinence) that might be important to recognize and address in this high-risk group.

Given the high proportion of cigarette smokers and alcohol and other drug dependence in this cohort, we expected to find a higher cumulative mortality compared with the general population. The fact that there was a 48.1% cumulative mortality at 20 years compared with an expected 18.5% is not only statistically significant (P<.001) but is of substantial clinical importance. The observation that the RR estimates for deaths from causes unrelated to alcohol and tobacco are also high suggests that there are other possible causes that are not yet confirmed (e.g., lack of social support, poor diet). The small number of deaths in the categories unrelated to alcohol and tobacco make it difficult to asses this with precision.

Our study shows a higher mortality for alcoholic men, which differs from the previous observation that mortality in alcoholic women is similar to that of their male counterparts. The reasons for this difference are not clear, but since our cohort is extensive and has a large number of women, we believe the finding is credible.

This study has many strengths. First, it is population based, and the patients who are treated at the IAP are representative of middle-class patients in the upper Midwest. The study is also based on a treatment program that has been consistent of over the years in the provision of services to the population served. Indeed, the mortality rates reported herein are probably the best case scenario, given that these patients are from a small Midwestern community and not self-selected for higher mortality characteristics such as might be found in patients from inner-city populations, veterans, or referral series. The inner-city and veteran alcoholic patients my have higher mortality rates and the

distribution of diseases that cause death might be different, though substantial tobacco-related mortality would still be expected. Data from these groups, however, are not likely to be generalizable to the majority of the US population.

While the small number of minorities in this cohort is a limitation, this is one of the few studies that report mortality in a large number of women with addictive disorders. We believe that our data, though not representative of all addictions programs, can be generalized to other US whites and specifically to those in the upper Midwest during the period 1972 through 1983. Having death certificates for 96.4% of the 222 subjects who died is also a strength, as is the long mean follow-up in survivors. Concerted effort was made to locate patients with incomplete follow-up through telephone contacts and searches of state and national death indexes.

Thus, only a small percentage did not have a lengthy follow-up. The large sample size, 8913 person-years of follow-up, allows for many events, thus supporting meaningful statistical analyses not possible in smaller samples with shorter follow-up. Another consideration in interpreting these findings is the use of the CDC’s Morbidity and Mortality Weekly Reports as a basis to determine the relatedness of deaths to alcohol and/or tobacco. We recognize that literature is continuously accumulating that adds to the lists of causes of deaths related to alcohol and/or tobacco, but we chose to include

disorders whose association is sufficiently certain to be published in the reports from the CDC. Thus, our results may underestimate the number of deaths related to tobacco and/or alcohol.

A major weakness is the lack of a good assessment of the outcome of the addictions treatment for this cohort. Thus, we cannot make comparisons of mortality rates in those who were and were not abstinent from alcohol and other drugs, as was done by others. Another limitation, as noted above, is that our sample is not representative of patients with alcoholism and/or other addictions. We have few minorities and relatively few patients who were treated for nonalcohol (nonnicotine) dependence. However, we are unaware of a single population-based cohort that is representative of the entire US population and would encourage others to replicate our work in other populations. We also do not know the use patterns of tobacco alcohol, and/or other drugs throughout the follow-up period and how this might affect disease development. Another limitation of this and all death certificate-based studies is that they are retrospective in nature and rely on abstraction of medical records and death certificates for information. We also

recognize that many important questions remain, such as how abstinence from smoking affects mortality in alcohol- and other drug-dependent patients and how abstinence from alcohol and other nonnicotine drugs affects smoking behavior.

In conclusion, several observations are obvious from our findings:

1. Patients previously treated for alcohol and/or other nonnicotine drug dependence are at high risk for premature mortality.

2. Tobacco-related causes are the leading cause of death in such patients and are significantly more frequent than alcohol-related causes of death.

3. Nicotine dependence treatment is imperative in this high-risk group.

This work was supported in part by research grants DA 08039 from the National Institute on Drug Abuse and AR30582 from the National Institutes of Health, US Public Health Service.

We appreciate the careful work of our nurse abstrators Marilyn Kochheiser, RN, and Judy Trautman, RN, and the diligence of Nancy Pilger in preparing the manuscript.

References

1. Ravenholt RT. Addiction mortality in the United States. Popul Dev Rev. 1984;4:697-725. 2. McGinnis JM, Foege WH. Actual causes of death in the United States. JAMA.

1993;2207-2212.

3. Craig TJ, Vannatta PA. The association of smoking and drinking habits in a community sample. J Stud Alcohol. 1977;38:1434-1439.