Behavior-related health risk factors, mental disorders and mortality after 20 years in a working aged general population sample

Mortality is predicted by the sum of behavior-related health risk factors (BRFs: tobacco smoking, alcohol drinking, body overweight, and physical inactivity). We analyzed degrees and combinations of BRFs in their relation to mortality and adjusted for mental disorders. In a random sample of the general population in northern Germany aged 18–64, BRFs and mental disorders had been assessed in 1996–1997 by the Munich Composite International Diagnostic Interview. A sum score including eight ranks of the behavior-related health risk factors was built. Death and its causes were ascertained 2017–2018 using residents’ registration files and death certificates. Relations of the sum score and combinations of the BRFs at baseline with all-cause, cancer, and cardiovascular mortality 20 years later were analyzed. The sum score and combinations predicted all-cause, cardiovascular and cancer mortality. The odds ratio of the sum score was 1.38 (95% confidence interval 1.31–1.46) after adjustment for age, gender, and mental disorder. In addition to the BRFs, mood, anxiety or somatoform disorders were not related to mortality. We concluded that the sum score and combinations of behavior-related health risk factors predicted mortality, even after adjustment for mental disorders.


Sample
A random adult population sample aged 18 to 64 years was used 33 .Among the 5,829 individuals eligible for the baseline study, 4,093 (70.2%) interviews had been completed July 1996 to March 1997, and 4,075 were analyzed 33 .A mortality follow-up was conducted from April 2017 until April 2018.Among the 4,075 baseline study participants, vital statistics data could not be proved for 47 34 .For 4,028 (98.8% of 4,075) study participants, we received the data.This is our final sample for the reported data analysis.

Assessments
Baseline Four BRFs were assessed by self-statements: tobacco smoking, alcohol drinking, body overweight, and leisure time physical inactivity.Tobacco smoking was part of a standardized interview.Alcohol drinking, body overweight and physical inactivity were assessed using standardized questionnaires which were filled in by the study participants as part of the interview.Tobacco smoking included smoking status (never or ever less than daily smoker, former daily smoker, current daily smoker less than 20 cigarettes per day, current daily smoker 20 or more cigarettes per day).Never smokers were those who answered "No" to the question whether they ever had smoked tobacco by cigarette, cigar or pipe in their life before.Ever less than daily smokers reported a history of smoking but not daily over a time period longer than 4 weeks in their life.Former daily smokers were smokers who had disclosed daily tobacco smoking over a time period of more than 4 weeks but not during the last 12 months prior to the baseline interview.Current daily smokers had smoked daily during the last 12 months prior to the interview.Alcohol drinking during the last 12 months prior to the interview was assessed by the Alcohol Use Disorders Identification Test, first three questions (Alcohol Use Disorders Identification Test Consumption, AUDIT-C 35,36 ): "1.How often did you have an alcoholic drink in the past 12 months?"(0: never, 1: once www.nature.com/scientificreports/ a month or less, 2: 2 to 4 times a month, 3: 2 to 3 times a week, 4: 4 times a week or more often).Those who had an alcohol drink in the last 12 months received questions 2 and 3: "2.If you have an alcoholic drink, how many glasses do you typically drink at 1 day?" (a small glass or a bottle of beer, a small glass of wine or sparkling wine, Spirits or liquor; 0: 1-2, 1: 3-4, 2: 5-6, 3: 7-9, 4: 10 or more), "3.How often did you drink 6 or more glasses in a row?" (0: never, 1: less than once a month, 2: once a month, 3: once a week, 4: daily or almost daily).The sum of the three item scores is the AUDIT-C sum score.Body overweight was estimated using the body mass index (BMI) as weight in kilograms divided by height in meters squared.Physical inactivity was assessed by a score with the value range 5 to 20 based on five questions about single areas of leisure time physical activity: ride a bicycle; practice gymnastics, aerobics or dancing; playing sports such as soccer, volleyball, handball, tennis; hiking or going on longer walks; working in house or garden.For each of these activities the respondent was asked to indicate how often s/he practiced the activity.We collapsed the answer categories into four: daily or almost daily, several times per week, once a week, less than once a week or never.We used a score of 1 to 4. It was the higher the less frequently the activity was practiced.
Mental disorders included mood disorders, anxiety disorders, and somatoform disorders 34 .These had been assessed for the lifetime before baseline according to the criteria of the Diagnostic and Statistical Manual of Mental Disorders, fourth edition, American Psychiatric Association 32 .The Composite International Diagnostic Interview (M-CIDI) 33,37 had been used.

Mortality follow-up
Vital statistics data were retrieved from the residents` registration files at the place of the last residence for allcause mortality.For cardiovascular and for cancer mortality, we used the death certificate information.Based on the information of the residents' registration office about the date of death, we received the death certificates from the local health authorities at the place of residence of the individual.The death certificates included health disorders which inferred death, disorders which were a main cause of, and disorders which may have contributed to death.A maximum of 15 disorders and 11 disorders as found by autopsy could be documented by the certificating physician.We used cardiovascular disorders and cancer among the disorders that inferred or were a main cause of death based on the International Classification of the Diseases, version 10 38 .

Data analysis
The data analysis was performed in three steps: First, we determined the ranks of the four single BRF variables.We started with four ranks per variable.We collapsed the four values each of body overweight and physical inactivity into three based on the number of death cases per group.A score 0 to 3 for each of tobacco smoking and alcohol drinking and a score 0 to 2 for body mass index and for physical inactivity turned out.The four ranks of tobacco smoking were: 0 never or ever less than daily smoker, 1 former daily smoker, 2 current daily smoker less than 20 cigarettes per day, 3 current daily smoker 20 or more cigarettes per day.Alcohol drinking included: 0 AUDIT-C = 1-4, 1 AUDIT-C = 5-7, 2 AUDIT-C alcohol abstinent, 3 AUDIT-C = 8-12.The ranks of body overweight were: 0 BMI < 27, 1 BMI 27-< 32, 2 BMI 32 or higher.Physical inactivity had the ranks: 0 physical inactivity score 16 or lower, 1 physical inactivity score 17-19, 2 physical inactivity score 20.
Second, the BRF sum score was constructed as the sum of the values of the four single variables (0 to 10).The values 7 to 10 included 117 persons only and were collapsed to one group.Thus, the final BRF sum score had eight ranks (0-7).We built the 16 combinations of the four BRFs after dichotomizing tobacco smoking into never daily smokers vs. ever daily smokers, alcohol drinking into low risk (AUDIT-C: 1-4) vs. high risk drinking (AUDIT-C: 5-12) or alcohol abstinence in the last 12 months prior to the interview.We used alcohol abstinence and AUDIT-C 5-12 as one group because evidence had shown that alcohol abstinence included former problem drinkers and had been related to death in a similar way as high alcohol consumption 39 .Body overweight was used as normal weight (BMI lower than 27) vs. overweight (BMI 27 or higher), and physical inactivity as physically active (physical inactivity score: lower than 17) vs. physically inactive (physical inactivity score: 17 or higher).
Third, we analyzed the prediction of all-cause, cardiovascular and cancer mortality by the single BRFs, the BRF sum score, and the BRF combinations.For Cox Proportional Hazards models the proportional hazards assumption was tested using Schoenfeld residuals 40,41 .The Cox Proportional Hazards assumption was not fulfilled in one or more variables in each of the models that have been calculated.Therefore, we used logistic regression analysis.Odds ratios (ORs) with 95% confidence intervals (CIs) are presented.With respect to the single BRFs (Table 3), we analyzed three models.Model 1 included each single BRF adjusted for age and sex.Model 2 was adjusted for age and sex and the three further BRFs.Model 3 in addition was adjusted for mental disorders.The BRF sum score and combinations (Table 4) were analyzed by two models.Model 1 was adjusted for age and sex.Model 2 was adjusted for age, sex, and mental disorders.We used three groups of mental disorders in the data analysis: mood, anxiety, and somatoform disorders.Mood disorders included major depression, dysthymia, bipolar disorders, and mood disorders based on a medical condition.Anxiety disorders were panic disorders, phobias, anxiety disorders not further specified, compulsive disorders, posttraumatic stress disorder, and anxiety disorder on grounds of a medical condition.Somatoform disorders included somatization disorder, conversion disorder, pain disorder, and hypochondriasis.
Missing values were found in the number of cigarettes per day among the current daily smokers (14; 0.35% of 4028) and in the alcohol drinking (35; 0.87% of 4028), body overweight (30; 0.74% of 4028), and physical inactivity variables (40; 0.99% of 4028).The missing values were replaced by the means of 5-year age groups among females and males each.For sex, age and mental disorders no values were missing due to the rules of the computer-assisted interview.They included that the study participant was recontacted if any information in the interview was missing and could not be provided by answers to other questions in the interview 37 .The datasets used and analyzed during the current study are available from the corresponding author on reasonable Vol:.(1234567890

Results
At baseline, 49.90% of the final sample had a BRF sum score 3 or greater (2010 study participants; Table 1).The likelihood of a mental disorder was higher for all BRF sum score ranks compared to no BRF.The data revealed an OR 1.86 (1.32-2.63)for a mental disorder among the persons with two BRFs who were daily smokers and alcohol high risk drinkers.Among the study participants with three BRFs, those who were daily smokers, alcohol high risk drinkers and physically inactive had an OR 2.38 (1.74-3.25)for a mental disorder.The BRFs were associated with each other except tobacco smoking with body overweight (Table 2).All-cause mortality was predicted by each of the single BRFs after adjustment for age, sex, and the other three BRFs (Table 3).The OR per rank was 1.47 (1.35-1.60)for tobacco smoking and 1.45 (1.30-1.62)for alcohol drinking in model 2. For cardiovascular mortality, the data revealed increased ORs for all four BRFs.The highest rank of alcohol drinking had an OR 4.78 (2.43-9.41) in model 2.
The higher the BRF sum score was, the higher were the proportions of deceased persons (Table 4).Among the study participants who had no BRF, 7.91%, among the study participants who had seven to ten BRFs, 34.19% had been deceased.The data revealed 38% higher odds of all-cause mortality for each of the ranks, the study participants who had none of the BRFs being the reference.For persons with a sum score 3 or higher, the ORs were significantly increased compared to the persons who had no BRF.After adjustment for mental disorders in addition to age and sex, the OR per sum score rank was 1.38 (1.31-1.46)and 8.09 (4.66-14.02)for the sum score rank 7 to 10. Mood, anxiety and somatoform disorders were not related with an increased mortality risk.
For the combinations of the single BRFs, the study participants who had no BRFs were the reference group.Among the person groups with two or more BRFs, all ORs for all-cause mortality were significantly increased.Among the participants with two BRFs, ever daily smokers and alcohol high risk drinkers had the highest OR (2.91; 1.80-4.70)after adjustment for age, sex, and mental disorders.The lowest OR was found in the group that had neither a tobacco-nor an alcohol-related BRF.Among the persons with three BRFs, those with ever daily smoking, alcohol high risk drinking and physical inactivity had the highest OR (4.79; 3.08-7.43).All ORs that were statistically significant after adjustment for age and sex remained to be significant after additional adjustment for mental disorders (model 2).
According to cardiovascular mortality, in both models increased ORs of the BRF sum score were found compared to persons without BRFs.The data revealed 49% higher odds of mortality for each of the ranks with the study participants who had none of the BRFs as the reference (model 1).The data of the combinations revealed that among persons who had three or four BRFs all combinations showed significantly increased ORs.
According to cancer mortality, in model 1 increased ORs were found for the BRF sum score (1.35; 1.25-1.46).Persons with any mood, anxiety or somatoform disorder did not have an increased risk of cancer death.For the combinations of the BRFs, the data revealed eight increased ORs.Among these, seven included ever daily

Discussion
This study has four main findings.First, the BRF sum score turned out to be related with mortality risk in a dosedependent manner.Second, among the specific combinations of the single BRFs alcohol drinking and tobacco smoking seem to strongly contribute to the prediction of mortality.Third, cardiovascular mortality was predicted by the BRFs in a particularly strong manner.Fourth, mental disorders did not change the findings considerably.
The BRF sum score predicted mortality.The score turned out to have three advantages.First, it gave evidence for the dose-dependent relation with mortality.Each of the seven ranks indicated a 38% higher likelihood to die within the 20 years.This finding gives evidence on a dose-relation between BRFs and mortality risk.Many of the studies before had been limited to BRFs as being present or absent 8,9,43 .Second, the sum score may provide more information about risk of death than single BRFs because of covering four behavior-related health risk factors and eight ranks.Third, the data revealed rank 3 of the sum score as the lowest one that indicates an increased risk of death.This finding seems to be particularly important for public health.Among all 4028 study participants, 49.90% had a BRF sum score 3 or higher.
According to specific combinations, the findings suggest that alcohol consumption might be of particularly strong influence.Persons with high risk alcohol drinking had seven ORs greater than 2 for all-cause mortality.This was the largest number of ORs greater than 2 among all BRFs.High risk alcohol drinking was also involved in all three combinations of the BRFs with an OR larger than 3.
According to cardiovascular mortality, the BRF sum score turned out to be related with the risk of death in a particularly high dose manner.Each rank of the sum score was 48% higher compared to the respective lower rank after adjustment for age, sex and mental disorders.Nine combinations of BRFs had an OR larger than 2.Among them, the most frequent BRF was alcohol high risk drinking.Persons with high risk drinking were involved in seven of the nine ORs larger than 2. According to cancer mortality, the BRF sum score was also in a linear relation to the mortality risk with a 35% higher risk per rank after adjustment for age and sex.Daily tobacco smoking was included in seven out of eight BRF combinations that predicted cancer mortality.Tobacco smoking was the only BRF that predicted mortality among persons with one BRF.This lends support to the assumption that in cancer mortality tobacco smoking might have a particularly strong influence among the four BRFs.
Mood, anxiety or somatoform disorders did not add to the prediction of mortality by BRFs.Our findings at first view seem to contradict findings of a higher mortality among subjects with a mental disorder than among healthy comparison samples 16 .One reason might be that our sample had been drawn from the general population.The study participants with a mental disorder in their majority included those who had not utilized psychiatric treatment.Among them, there might be those with a low severity of the disorder.Persons with a mental disorder but no treatment survived longer than those with a treated mental disorder 34 .
Strengths of our study include that a sum score with the potential of ten ranks has been used.This provides the opportunity to detect more detailed dose-relations between BRFs and mortality.Second, 15 combinations of the BRFs in addition to no BRF turned out to add information to the relations between BRFs and mortality.Subgroups of persons at particularly high risk among the general population and combinations of single BRFs that might be more important than others in the prediction of mortality have been detected.Third, we assessed mental disorders using an internationally standardized interview and the diagnostic criteria of the American Psychiatric Association.Our study added findings about BRFs, mental disorders, all-cause, cardiovascular, and cancer mortality in the time frame of 20 years.Limitations of our study include that we had self-report data only.Reporting bias is likely.This particularly might be the case for alcohol consumption.But for the other BRFs also, underreporting may have taken place.It seems plausible that residents with high ranks of BRFs might be particularly prone to underreport BRFs.Feelings of guilt and shame might be responsible for that.Our study was limited to four BRFs.More may be relevant for death.BRFs may have evolved or been discontinued during the 20 years.We reported a considerable number of statistical test results.Problems of multiple testing should be kept in mind.

Table 1 .
Characteristics of the sample at baseline: Study participants with mental disorder among all study participants with characteristic.Logistic regression analysis for study participants with any of mood, anxiety or somatoform disorder in life before baseline adjusted for age and sex.N number of study participants at baseline, n number of study participants with mental disorder.OR odds ratio, CI 95% confidence interval, ref reference group, AUDIT-C Alcohol use disorders identification test consumption.*includes the original values 7-10, **ever daily smoking, ***alcohol high risk drinking or alcohol abstinence last 12 months.

Table 2 .
Associations among the behavior-related health risk factors at baseline.Spearman correlation coefficients.p probability of error, ns not significant. .Ever daily smokers turned out to be the only subgroup with an increased OR among the study participants with one BRF.The highest OR (4.14; 2.04-8.40)was found for those who had four BRFs.

Table 3 .
Single behavior-related health risk factors and mortality.N number of study participants at baseline.n number of study participants who had been deceased in the time after baseline until mortality follow-up.Model 1: adjusted for age and sex.Model 2: adjusted for age and sex, and the other three behavior-related health risk factors at baseline.Model 3: adjusted for age, sex, the other three behavior-related health risk factors, and three groups of mental disorder (mood disorder, anxiety disorder, somatoform disorder) in life before baseline.Logistic regression due to Cox Proportional Hazards Assumption not having been fulfilled in one or more single variables.CI 95% confidence interval, OR odds ratio, ref reference group, AUDIT-C Alcohol use disorders identification test consumption.

Table 4 .
Behavior-related health risk factor sum score, health risk factor combinations and mortality.N number of study participants at baseline.n number of study participants who had been deceased in the time after baseline until mortality follow-up.Model 1: adjusted for age and sex.Model 2: adjusted for age and sex and three groups of mental disorder (mood disorder, anxiety disorder, somatoform disorder) in life before baseline.Logistic regression due to Cox Proportional Hazards Assumption not having been fulfilled in one or more single variables.Behavior-related health risk factor sum score: sum of ranks of tobacco smoking, alcohol drinking, body overweight, and physical inactivity.OR odds ratio, CI 95%-confidence interval, ref reference group.-not applicable.