OUP user menu

Is depression associated with health risk-related behaviour clusters in adults?

Pierre Verger , Caroline Lions , Bruno Ventelou
DOI: http://dx.doi.org/10.1093/eurpub/ckp057 618-624 First published online: 29 April 2009

Abstract

Background: Depressive disorders have been linked to health risk-related behaviours (HRBs) considered separately. Our objective was to study whether depression is associated with the co-occurrence of HRBs in adults. Methods: A sample of 17 355 subjects aged ≥18 years, derived from the 2002–03 cross-sectional Decennial Health Survey; probable depression was assessed with the CES-D scale. A cluster analysis of various HRBs (tobacco use, alcohol use, binge drinking, physical inactivity, certain eating habits) was used to study their co-occurrence. Multiple regressions adjusted on demographic and socio-economic characteristics, Body Mass Index and chronic illnesses were performed to study associations between probable depression and the HRBs clusters obtained. Results: Five clusters were observed evidencing a gradient of cumulative exposure to HRBs: ‘healthy lifestyles (Cluster 1), ‘non-daily-consumers-fruit-and-green-vegetables’ (Cluster 2), ‘regular alcohol users’ (Cluster 3), ‘daily smokers’ (Cluster 4) and ‘cumulate risk takers’ (Cluster 5). Compared with Cluster 1, positive associations were found between probable depression and Clusters 2, 4 and 5: OR 1.49 (95% CI 1.26–1.76) for Cluster 2; OR 1.81 (95% CI 1.54–2.12) for Cluster 4; OR 2.05 (95% CI 1.68–2.51) for Cluster 5. For Cluster 3, no association was found: OR 1.01 (95% CI 0.84–1.21). Conclusions: HRBs tend to co-occur in the general population, more frequently in case of probable depression. Further research is necessary to disentangle the direction of the links between depression and HRB clusters. Nonetheless, these results question the classic design of education campaigns considering HRBs separately. Moreover, screening for depression should be systematic during prevention consultations and various HRBs should be monitored when treating depressive patients.

  • cluster analysis
  • depressive disorder
  • health behaviours
  • general French population

Introduction

Depression, with a lifetime prevalence of 14% in Western Europe1 and 17% in the USA,2 has severe consequences in terms of social impairment3 and socio-economic cost.4 An increasing number of studies have pointed to an association between depressive disorders and various health risk-related behaviours (HRBs) in various population groups (general population, young adults, adults, elderly, etc.): tobacco use or dependency,5,6 alcoholism,7 insufficient physical or sporting activity,8–11 and unhealthy eating habits.8,12,13 But only a few studies, which mainly concern adolescents,14–16 have sought to determine whether or not depression increases the probability of co-occurrence of several HRBs. These studies suggest that risk-related behaviours tend to cluster or co-occur rather than occurring in isolation17 and this may have important implications in terms of health education and prevention.

In 2002–03 a national cross-sectional survey was conducted on health and health-related behaviours in the French general population (the Decennial Health Survey). The present article aims, in the adult population, to study whether depression is associated with the co-occurrence of various HRBs—unhealthy eating, physical inactivity, tobacco and alcohol use.

Methods

Study design and population

Households were selected randomly from those identified (addresses) in the national census. The survey concerned all individuals in a given household for whom the randomly drawn address was their main residence. People living in institutions (e.g. retirement homes, religious communities, prisons and hospitals) or mobile homes or who were homeless were not included. Out of the 25 086 addresses randomly selected, 21 656 (about 51 000 inhabitants) met the inclusion criteria for households (the randomly drawn address was the main residence), and 14 813 households (35 073/51 000 individuals (68.8%), of whom 26 341 were ≥18 years) took part in the three interviews; 20 315/35 073 (57.9%) handed in completed, useable questionnaires. In the present study, the following were excluded from the analysis: (i) Individuals who were under 18 (n = 2733); (ii) individuals who failed to complete information on their weight and height (n = 12); (iii) women who were at least 5 months pregnant (n = 102); (iv) individuals for whom the data required for the present study as recorded during the visits/interviews was incomplete (n = 115). In all, the study included 17 355 individuals.

Study variables

The Decennial Health Survey used a combination of three face-to-face interviews during three home visits 1 month apart and a self-administered questionnaire to collect data at individual level. The self-administered questionnaire was given to respondents at the time of the first visit, and returned on the second or third visit. It comprised in particular the Center of Epidemiological Studies Depression Scale (CES-D). This widely used instrument, which has a validated French version, enables assessment of current intensity of depressive symptoms, and screening for potential cases of depression.18,19 From the 20 items that it comprises, a depressiveness score has been constructed (range 0–60, the highest score relating to the most severe symptoms). Cases of ‘probable depression’ have been defined as scores over 17 for men and over 23 for women.18 The self-administered questionnaire also included questions on tobacco use: ‘Are you a smoker?’—Yes, every day; yes, not every day; no, but previously I smoked every day; no. A dichotomous indicator for tobacco use was constructed: current non-smoker (and previous or occasional smokers) vs. current smoker. Regarding alcohol use, the questionnaire included the following two questions from Audit-C:20 ‘How often to you drink beverages containing alcohol?’; ‘How often do you drink six glasses or more of alcoholic beverages on a single occasion? For these two questions, the response scale used was different from that of the Audit-C: never, once a month or less, twice or 3 times a month, once or twice a week, 3 or 4 times a week, 4 or 5 times a week, almost every day. A dichotomous indicator for alcohol use and binge drinking was constructed: consumption of alcohol at least 4 or 5 times a week (regular users) vs. less, consumption of at least six glasses on the same occasion at least twice a month vs. less.

At the time of the first visit, information was collected on gender, age, profession, income, health insurance, and current pregnancy (yes/no). The present profession (or that of last employment) was recoded into the following categories according to the recommendations of the French National Institute for Statistics and Economic Studies:21 white collar/managerial; blue collar; unskilled workers/labourers; self-employed/trade; intermediate professions; has never worked. The income per consumption unit (hereafter ‘income category’) was calculated on the basis of total income of all the members of the household over the last 12 months (net amount in Euros) and the number of members of the household, using the Organization for Economic Co-operation and Development (OECD) equivalence scale.22 The questionnaire asked the participants whether they had a complementary health insurance, (providing a better reimbursement rate than the ‘Sécurité sociale’ coverage alone) [no; yes via the state assisted ‘couverture médicale universelle’ (CMU) for individuals with low incomes; yes, via a non-assisted contract].

During the second visit, participants were asked to report presence of chronic illness, weight and height. Obesity was defined as a BMI (relationship of weight to squared height) of ≥30 kg/m2, overweight as between 25 and 29.9 kg/m2, underweight as under 18.5 kg/m2.

During the third visit, information was collected on any new chronic illness, any new pregnancy, and on eating habits and physical activity. Because of constraints on the length of interviews, a brief qualitative food frequency questionnaire used in the preceding Decennial Health Survey (1991–1992) was used do collect information on eating habits.23 Questions were asked on the consumption on various food groups, including fruit and green vegetables, with the following scale of consumption frequency: ‘Nearly every day; at least once a week; less than once a week; seldom or never’. We restricted our analyses to fruit and green vegetable consumption to avoid an over-representation of eating behaviours compared with other HRBs included in the analyses and because eating fruit and vegetables each day is recommended in public educational campaigns in France. A dichotomous indicator of fruits and green vegetables consumption was constructed: consumption daily or almost daily vs. the other responses grouped. For physical activity, the question was as follows: ‘Do you regularly engage in some sport or activity that you would view as sport in terms of intensity or duration?’ Yes/no.

Statistical analysis

The sample was weighted to take account of the characteristics of non-participants in face-to-face interviews and individuals who did not complete the self-administered questionnaire, and to obtain a representative sample of the French general population for gender, age, educational level of the reference subject in the household, type and size of household, region, size of locality, type of housing.21

A cluster analysis was conducted to identify clusters of individuals determined by the association of different HRBs. Cluster analysis enables the study of the statistical proximity of individuals on the basis of the factors under study (HRBs), without preconception as to any relationship among these factors. Clustering implies that the distribution of the factors studied is not independent, but instead reflects their common features. Thus a classification procedure enables the evidencing of risk factors that have been statistically defined according to criteria of similitude or distance.24 To focus on behaviours deviating from recommendations, we used the following dichotomous variables: daily tobacco use, low consumption of fruit and green vegetables, little sporting activity, regular alcohol use and frequent binge drinking. We used the Ward distance which links clusters on the basis of the degree of similarity between observations in the same cluster: this is done by minimizing the within-cluster sum of the squares of each cluster when clusters are linked.25 The corresponding decision tree (figure 1) was constructed using the SAS proc tree procedure. The number of clusters was chosen in accordance with the inter-cluster loss of inertia that could be visualized on the decision tree: the decision rule was to select the cluster number maximizing inter-cluster loss of inertia (figure 1).26 To check any impact of a potential household effect on the results a concordance analysis was conducted between the clusters initially observed and those obtained from a sub-sample comprising a single individual drawn randomly from each household.

Figure 1

Cluster analysis result: tree diagram

The association between belonging to one of the clusters (dependent variable) and the presence of probable depression (yes/no)—or the depressiveness score—(explicative variables) was tested with non-ordered polytomous logistic regressions (generalized Logit Model). Indeed, the structure of the dependent variable—more than two mutually exclusive clusters—requires the implementation of a polytomous logistic regression taking account of all the concurrent alternatives for this dependent variable.27 A ‘healthy lifestyle’ cluster was used as a reference. The clusters were ranked according to their distance from the reference cluster using the Ward method. The models were adjusted on the following: demographic and socio-economic variables; BMI categories (on account of the frequent association between depression and obesity);28 and the number of other chronic illnesses reported over the study period (CIM-10 coding) because of the frequent association between depression and chronic conditions.29 The Cochran–Mantel–Haenszel rank method was used to test for a monotonous relationship between the presence of probable depression and belonging to a cluster, on the hypothesis that these clusters were ordered according to an increasing gradient of accumulated HRBs. The statistical analyses were performed using SAS software version 9, using procedures to take account of the weightings (proc survey) and sampling (cluster option).

Results

Among the 17 355 subjects included in the study, 8356 were men and 8999 were women. Their average age was 47 years. The majority had normal weight, were blue-collar workers, had an above-average income, and complementary health insurance (table 1).

View this table:
Table 1

Socio-demographic characteristics of the population according to the clusters (weighted %).

TotalClustersP*
N = 17 3551a2b3c4d5e
Age (mean)47.1549.7743.3957.9237.0439.46<0.0001
Gender (%)
    Men47.9833.8446.2468.9648.8579.99<0.0001
    Women52.0266.1653.7531.0351.1520.01
BMI (%)
    Underweight2.492.082.861.124.193.21<0.0001
    Normal56.2254.9755.1951.1365.8056.28
    Overweight30.1730.5828.8537.7822.4330.64
    Obese11.1212.3713.099.977.579.88
Profession (%)
    Never worked8.558.7613.312.009.159.49<0.0001
    Unskilled worker24.1919.7524.5125.2530.5931.29
    Blue collar26.8829.9928.3617.5430.8418.86
    Self-employed/trade10.0310.388.8915.254.9210.16
    Intermediate professionals18.6718.9916.4521.5717.3118.73
    White collar/managerial11.6712.138.4818.387.1911.46
Income (%)
    ICU ≤ poverty threshold8.728.089.346.0511.7410.22<0.0001
    P threshold < ICU ≤ median ICU37.8436.7841.7533.5441.5135.93
    ICU > median ICU53.4455.1448.9160.4146.7453.84
Complementary health cover (%)
    CMU: entitled or dependant beneficiary23.8827.4524.3315.5225.5918.60<0.0001
    Mutual or private insurance company67.7465.4265.8577.6363.1572.52
    No complementary cover8.377.139.826.8511.268.88
Number of chronic illnesses (mean)0.890.990.871.130.560.62<0.0001
Probable depression (CES-D)f (%)
    No (CES-D < threshold)85.5787.9983.5886.7582.6880.76<0.0001
    Yes (CES-D ≥ threshold)14.4312.0116.4213.2517.3219.24
  • *Test for comparing clusters: Chi2 for qualitative variables, Anova (Fisher test) for quantitative variables

  • a: Healthy lifestyles’

  • b: Non-daily-consumers-fruit-and-green-vegetables’

  • c: ‘Regular alcohol users’

  • d: ‘Daily smokers’

  • e: ‘Cumulate risk-takers’

  • f: Reference: CES-D < 17 in men and CES-D < 23 in women

A five-cluster solution was identified (figure 1, table 2). Cluster 1 groups 41.0% of the sample and is made up of subjects with relatively healthy lifestyles. Cluster 2 comprises 18.2% of the sample and mainly differs from the preceding cluster by comprising the highest proportion (100%) of non-daily-consumers-fruit-and-green-vegetables. Cluster 3 concerns 17.0% of the sample and mainly differs from the two preceding ones by the highest proportion (100%) of regular alcohol consumers. Cluster 4 includes 16.3% of the sample and comprises 100% smokers, and cluster 5 includes 7.5% of the sample and is characterized by 100% frequent binge drinkers. The cumulate frequency of HRBs shows a graduation in the accumulation of HRBs from Cluster 1 (healthy lifestyles) to Cluster 5 (which cumulates all the HRBs) (table 2). The weighted Kappa value comparing the cluster solution observed on the initial sample with the solution obtained from the subsample comprising one individual per household was 0.65 showing good cluster agreement between the two samples.

View this table:
Table 2

Characteristics of the 5 clusters according to health risk-related behaviors (weighted %).

Health risk-related behavioursClusters
1a:2b:3c:4d:5e:
Daily smoker0022.5410046.29
Non regular sporting activity53.2360.7450.8563.7753.23
Alcohol ≥ 4 times per week00100043.14
Binge drinking ≥ 2 times per month0000100
Non-daily consumers of fruit010027.5244.9444.51
Non-daily consumers of green vegetables15.6250.5429.3948.9651.31
Cumulate frequency of risk-related behaviours16.4735.2138.3842.9456.41
Total, N (%)7105 (40.98)3174 (18.17)2873 (17.05)2896 (16.34)1307 (7.46)
  • a: ‘Healthy lifestyles’

  • b: ‘Non-daily-consumers-fruit-and-green-vegetables’

  • c: ‘Regular alcohol users’

  • d: ‘Daily smokers’

  • e: ‘Cumulate risk-takers’

Cluster 1 was mainly made up of women (table 1) but comparable with the overall sample for mean age, weight, socio-economic variables and chronic illness. Cluster 2 comprised mainly women and subjects younger than in Cluster 1, but was otherwise comparable for the other variables, with the exception of prevalence of probable depression, which was higher. Cluster 3 was distinct from the others in particular by mean age, the proportion of overweight subjects, and by subjects from the higher socio-professional categories. The prevalence of probable depression in this cluster was however close to that of cluster 1. Clusters 4 and 5 were made up of the youngest individuals, with a large proportion of non-qualified workers with low incomes. In contrast to Cluster 4, Cluster 5 comprised a majority of men.

The multiple regressions showed that belonging to Clusters 2 (non-daily-consumers-fruit-and-green-vegetables), 4 (daily smokers) and 5 (cumulate risk-takers) was significantly more probable in the presence of probable depression, but belonging to Cluster 3 (regular alcohol users, table 3) was not. Odds ratios for the presence of probable depression or the depressiveness score increased with the rank of the clusters (cumulate frequency of HRBs) (Cochran–Mantel–Haenszel test: P < 0.0001).

View this table:
Table 3

Associations between health risk-related behaviour clusters (dependent variable) and probable depression and depressiveness score (explicative variables) in the general population (n = 17 355) (two separate logistic non-ordered multiple polytomous multiple regressionsa).

Dependent variableProbable depressionb (Model 1)PWaldDepressiveness score (Model 2)PWald
OR (95% CI)cOR (95% CI)c
Non-daily-consumers-fruit-and-green-vegetables (Cluster 2)1.49 (1.26–1.76)<0.00011.02 (1.01–1.02)<0.0001
Regular alcohol users (Cluster 3)1.01 (0.84–1.21)1.00 (0.99–1.01)
Daily smokers (Cluster 4)1.81 (1.54–2.12)1.03 (1.02–1.03)
Cumulate risk-takers (Cluster 5)2.05 (1.68–2.51)1.04 (1.03–1.05)
  • a: Adjusted on gender, age, BMI, profession, income, complementary health cover, number of chronic illnesses

  • b: Reference: CES-D < 17 in men and CES-D < 23 in women

  • c: vs. the ‘healthy lifestyles’ cluster

Discussion

These results need to be viewed in the perspective of the methodological limitations of the study. First, although the participation rate was fair (68.8%), the percentage of useable questionnaires among the participants (57.9%) was rather low. The sample was therefore weighted to take account of the characteristics of non-participants as well as of individuals who did not provide a useable questionnaire. This limits, but does not remove, selection bias, in particular with respect to the health of individuals. Second, the cross-sectional design of the study does not make it possible to establish a temporal relationship between the supposed causal factor in the present analysis (depression) and the HRBs clusters. The notion that certain HRBs might favour depressive states has been brought up in the literature, for HRBs considered separately, notably concerning tobacco,5,6 restrictive eating behaviours,13 physical inactivity9,10,30 and chronic alcohol misuse.31 In particular, tobacco use could favour depression since the neurotransmitter system affected by certain components of tobacco smoke bears similarities with the neurotransmission mechanisms implicated in the biological mechanisms of depression.32 Data of this nature are not however, to our knowledge, available on the subject of the co-occurrence of these HRBs.

As the data used in the present study is self-reported, depressive state at the time of the interview might have affected the way in which respondents reported their behaviours (reporting bias). The questionnaire used in the Decennial Health Survey to document health behaviours is brief. It was used because it is well adapted to a population survey with major time constraints, since many aspects of health and health behaviours were evaluated in this survey. The stability of responses on this type of measure has been demonstrated over short periods in the European Health Behaviour Survey33 and also long-term stability (several months) in studies in the USA.34,35 These studies indicate a strong habit component for many HRBs, which lessens the risk of reporting bias.

The prevalence of cases of probable depression was identical among men and women, a finding that is not in line with data in the literature.1,2 This result is probably linked to the fact that the recommended thresholds in the French-validated version of the CES-D are different for men and women.18 However, the international threshold (CES-D score 16 or more for either gender)19 is lower than the French thresholds and not very specific. Using this threshold in our study, the prevalence of probable depression would have been 16.4% for men and 29.6% for women, which are high estimates in relation to those obtained with French diagnostic tools,36 but the results of the analyses of the links between probable depression and belonging to a cluster were similar to those obtained using the French thresholds. Alongside the above-mentioned methodological limitations, this analysis has the advantage of being adjusted on several confounding factors.6 Indeed certain authors indicate that the links observed between tobacco use and depression are affected by confounders, in particular social parameters.5 The present analysis is adjusted for different socio-economic factors (income, socio-professional category, complementary health coverage) which are associated both with HRBs and with depression risk.37 It is also adjusted on reported chronic morbidity and BMI categories, which can be at once linked to the HRBs under study and to depression.28,29

Probable depression was associated with the co-occurrence of HRBs in different health domains (eating habits, physical activity, psychoactive substance use): this is a new result for the adult general population. Our results suggest that in presence of probable depression, various HRBs tend to cluster statistically more frequently than would be expected in the absence of depression: the greater the number of HRBs, the stronger the association with probable depression (PWald associated with probable depression <0.0001, table 3). Moreover, the higher the severity of depressiveness, the greater the number of HRBs (dose–effect relationship). Our results also suggest that depression is associated with different patterns of HRBs (Clusters 2, 4 and 5) but not with all patterns (Cluster 3, although it corresponds to non-negligible HRBs). Clusters 4 and 5 were characterized by the highest cumulate frequencies of HRBs and the highest frequencies of daily smokers and binge drinkers respectively; their mean age was lower than that of other clusters. Cluster 2 presented a different pattern, as it did not include daily smokers or binge drinkers, but individuals with low fruit and vegetable consumption and non-regular sporting activity: it was composed of a majority of middle-aged women with low or modest socio-economic level. Finally Cluster 3, the oldest group, included the highest percentage of regular alcohol drinkers and was not associated with probable depression, probably because regular alcohol use, as defined in this analysis, is not a indicator of impaired mental health. The associations between the clusters and probable depression cannot be explained by their specific socio-demographic characteristics, as they were adjusted for in the multiple analyses.

Little research has approached the possible links between depression and HRBs with respect to their co-occurrence in the general adult population. Clustering of HRBs has been increasingly pinpointed in studies on young people.17,38–40 Some studies have established a relationship between health-risk behaviour clustering among young people and psychosocial factors such as poor mental health (stress, low self-esteem, anxiety, depression) poor educational achievement or personality traits.39 Regarding the links between depression specifically and clusters of HRBs, again published work concerns young people.14–16 These studies explore types of behaviour (such as driving after using alcohol, carrying weapons, binge drinking, marijuana or illicit substance use, risk-prone sexual intercourse) involving immediate risk or risk of chronic illness. They do not explore physical activity or eating habits. Most of these studies are not adjusted on socio-economic factors.

Depression could influence HRBs in several ways: anhedonia and withdrawal (central symptoms in depression) could account for diminished interest in health, and lesser receptiveness to health education messages (dietary recommendations for example), lesser sporting or physical activity,10 and difficulties quitting tobacco use.41 Depression has also been found associated with restrictive eating habits that could explain lesser consumption of fruit and vegetables (Cluster 2).12 It is also possible that depression alters risk perception, which is itself associated with HRBs.39 Finally, community surveys show substantial comorbidity between depression and alcoholism:7 alcoholism often develops secondarily to depression or anxiety and alcohol is commonly used to self-medicate symptoms of negative affect.7 In our study, only the cluster of binge drinkers was associated with depression, not the cluster of regular alcohol consumers (Cluster 3): binge drinking is an indicator of increased risk of alcoholism (alcohol abuse or dependence) while regular consumption of alcohol alone is not.20

Although our findings should be confirmed in other settings with prospective designs, their potential implications are important. First, as we show that certain population subgroups cumulate various HRBs, this questions the usual design of health education campaigns and interventions at two levels: (i) should they target specific groups? Indeed, receptiveness to such campaigns probably varies in population subgroups according to their more or less healthy lifestyles; (ii) should they address several risks at the same time, rather than solely separately, in compartmentalized manner? Second, our results could be useful for preventive action in the context of medical consultations. In prevention consultations, the psychological state of the subjects needs to be taken into account, and depression should be systematically screened for. Third, care delivery in depression should systematically entail particular vigilance as to the way in which various health recommendations are complied with in the course of treatment.

Key points

  • Certain subgroups of the adult general population cumulate various health risk-related behaviours in different domains (eating habits, physical activity, psychoactive substance use).

  • Our results suggest that probable depression is associated with a higher risk of co-occurrence of these health risk-related behaviours in the adult general population, independently from socio-economic factors and somatic comorbidity.

  • These results question usual designs of health education campaigns that do not target multiple risks takers.

  • In prevention consultations, the psychological state of the subjects needs to be taken into account, and depression should be systematically screened for.

  • Care delivery in depression should systematically entail particular vigilance as to the way in which various health recommendations are complied with in the course of treatment.

Funding

Conseil Régional Provence-Alpes-Côte d’Azur; Institut National de la Santé et de la Recherche Médicale (INSERM); Ministère de la Santé et des Solidarités)/Direction de la Recherche, des Etudes, de l’Evaluation et des Statistiques (DREES)/Mission Recherche (MiRe).

Conflicts of interest: None declared.

Acknowledgements

We would like to thank Angela Swaine Verdier for translating the manuscript.

References

View Abstract