The Longitudinal Association From Obesity to Depression: Results From the 12-year National Population Health Survey

Authors

  • Genevieve Gariepy,

    1. Douglas Mental Health University Institute, McGill University, Montreal, Quebec, Canada
    Search for more papers by this author
  • JianLi Wang,

    1. Department of Psychiatry, University of Calgary, Calgary, Alberta, Canada
    2. Department of Community Health Sciences, University of Calgary, Calgary, Alberta, Canada
    Search for more papers by this author
  • Alain D. Lesage,

    1. Fernand Seguin Research Centre, Louis-H. Lafontaine hospital, University of Montreal, Montreal, Quebec, Canada
    2. Department of Psychiatry, University of Montreal, Montreal, Quebec, Canada
    Search for more papers by this author
  • Norbert Schmitz

    Corresponding author
    1. Douglas Mental Health University Institute, McGill University, Montreal, Quebec, Canada
    2. Department of Epidemiology, Biostatistics and Occupational Health, McGill University, Montreal, Quebec, Canada
    3. Department of Psychiatry, McGill University, Montreal, Quebec, Canada
    Search for more papers by this author

(norbert.schmitz@douglas.mcgill.ca)

Abstract

Prior observational studies have investigated the association between obesity and depression but evidence remains weak and mixed. There has been a call for high-quality longitudinal studies to elucidate the etiologic relationship from obesity to depression. The main objective of this study was therefore to investigate whether obesity was a risk factor for depression in a nationally representative sample followed for 12 years. Seven waves of data collection (1994–1995 to 2006–2007) were obtained from the National Population Health Survey (NPHS). Our analyses included 10,545 adults without depression at baseline. Past-year major depression episode (MDE) was assessed from the Composite International Diagnostic Interview-Short Form for Major Depression (CIDI-SFMD). Obesity was estimated using baseline BMI from self-reported weight and height (obesity: BMI ≥30 kg/m2). Kaplan–Meier survival curves were generated and Cox proportional hazard regression modeling was used to estimate the risk of MDE by obesity status, controlling for sociodemographic and health and lifestyle variables. We found that obesity at baseline did not significantly predict subsequent MDE in women (adjusted hazard ratio (AHR): 1.03, 95% confidence interval (CI) 0.84–1.26) and negatively predicted MDE in men (HR: 0.71, CI 0.51–0.98), after adjusting for important confounders. In summary, our findings suggest that obesity is a significant (negative) predictor of depression in adult men but not in women. These results moderate prior evidence supporting a positive link from obesity to depression.

Introduction

The prevalence of obesity has been on the rise in Western countries (1). In Canada, nearly one-quarter (23.1%) of adults are estimated to be obese (2), increasing their risk for a number of chronic conditions, such as diabetes mellitus, cardiovascular diseases, and certain types of cancers (3,4), as well as disability and impaired quality of life (5,6). Depression is similarly prevalent. The 2001 Canadian Community Health Survey estimated the 12-month prevalence of major depression at 4.8% (ref. 7). Depression is linked to various somatic conditions such as cardiovascular and lung diseases (8,9), and is also known to affect quality of life (10).

There has recently been a growing interest in the psychological correlates of obesity and the role of comorbid depression and obesity in treatment and health outcomes (11). The cross-sectional association between depression and obesity has been researched extensively (12) but only a few population-based studies have assessed the longitudinal link from obesity to depression (13,14,15,16,17,18,19). The evidence overwhelmingly points to obesity as a positive predictor of depression. For example, Roberts et al. studied a sample of 2,298 participants aged ≥50 years interviewed in 1994. Obesity was found to predict depression 1 year (13) and 5 years later (14,15).

However, prior longitudinal studies presented some shortcomings. Most analyses were based on only two time points. Evaluating depression incidence on more than one occasion could capture potential changes of effects. Additionally, some studies reported prevalence rather than incidence ratios. The distinction is important. Prevalence is a measure of disease status in the population whereas incidence quantifies new occurrences of disease. The drawback of prevalence measures is that past depression is known to predict future depression (20). As a result, it becomes difficult to assert whether it was past depression or past weight status predicting future depression.

In a recent systematic review, Atlantis and Baker found only a weak level of evidence supporting the link between obesity and depression (12). The authors called for additional high-quality prospective studies to clarify the etiological relationship from obesity to depression. This call has been echoed by other reviewers (11,21,22). The longitudinal Canadian National Population Health Survey (NPHS) offers a unique opportunity to examine this relationship in a representative community sample. The objectives for this study were to (i) estimate the prevalence and temporal change in prevalence of obesity–depression comorbidity and (ii) investigate the longitudinal association from obesity to depression. Based on previous research, we hypothesized (i) an increasing prevalence of obesity–depression comorbidity over time and (ii) higher incidence rates of depression among obese participants compared to their normal-weight counterparts.

Methods and Procedures

Data source

The NPHS collects longitudinal information related to health status, health-care utilization, and health determinants from a representative sample of Canadians. Full details of the collection procedures and response profile are extensively described elsewhere (23). Briefly, the NPHS is a national health survey using multiple-stage, stratified random sampling procedures (23). The 1994/1995 NPHS participants (n = 17,276) formed a longitudinal cohort that was reinterviewed every 2 years by trained interviewers. This study focused on seven cycles of interviews (1994/1995; 1996/1997; 1998/1999; 2000/2001; 2002/2003; 2004/2005; 2006/2007). Cycle response rates were 83.6, 92.8, 88.3, 84.9, 80.8, 77.6, and 77.0% for cycles 1–7, respectively. For this study, only respondents aged between 18 and 80 years were selected (n = 13,618).

Dependent variable

Past-year major depression was evaluated using the Composite International Diagnostic Interview-Short Form for Major Depression (CIDI-SFMD) (24). The CIDI-SFMD is a brief version of the major depression section from the CIDI, a fully structured diagnostic interview designed to be used by trained interviewers who are not clinicians. A major depression episode (MDE), as defined in the NPHS, represents a purported 90% predictive cut-point for the CIDI-SF. It has been noted that unlike the full version of the CIDI, the CIDI-SFMD does not exclude depressive moods due to physical illness or bereavement and may therefore be vulnerable to false positives (25). Nevertheless, the overall classification accuracy of the CIDI-SFMD in identifying an MDE is 93% based on the Diagnostic and Statistical Manual of Mental Disorders-IV criteria (24). The sensitivity and specificity of the CIDI-SFMD using this cut-point when compared against the full version of the CIDI are 90 and 94%, respectively (24).

Independent variable

Weight status was based on self-reported height and weight (BMI, in kg/m2), and categorized according to guidelines from the National Heart, Lung, and Blood Institute (<18.5 kg/m2 for underweight; 18.5–24.9 kg/m2 for normal weight; 25.0–29.9 kg/m2 for overweight; ≥30.0 kg/m2 for obesity) (ref. 26). Obesity status was then dichotomized into obese and normal/overweight categories. Prior research has found minor or inconclusive associations of overweight with health outcomes (5,6), weight stigma (27), and psychological health, including depression outcomes (18,28). Underweight individuals were excluded from analyses, as our focus was on obesity and depression.

Covariates

The NPHS collected data on gender, age, ethnicity (white/nonwhite), marital status (married/common law; single; divorced/separated/widowed), and highest educational attainment (less than secondary school/secondary school graduation/postsecondary graduation). Physical activity level and smoking status were also assessed. Physical activity levels were estimated from total daily energy (kcal/kg/day) expanded during leisure time activities. An index was created to classify participants as inactive (<1.5 kcal/kg/day), moderately active (1.5–2.9 kcal/kg/day), or active (≥3.0 kcal/kg/day) (ref. 29). Alcohol consumption was not included in the analysis because of its relatively high number of missing data (25.6% of adults).

Self-perceived health status was assessed with one item: “In general, would you say your health is excellent, very good, good, fair, or poor?” Responses were dichotomized (poor/fair vs. good/very good/excellent). The validity of this measure is supported by several studies showing that this single-item is a strong and independent predictor of morbidity and mortality (30). Chronic somatic conditions referred to 10 illnesses diagnosed by health professionals and expected to last ≥6 months: asthma, arthritis, back problems, high blood pressure, migraine, bronchitis, diabetes, cardiovascular disease, cancer, and stomach ulcers.

The social support index assessed perceived social support by adding the number of affirmative responses to four questions: having someone to confide in, to count on in a crisis, to count on for advice, and who makes them feel loved and cared for. Higher score indicates better perceived social support. The scale was researched and designed for the NPHS to measure the instrumental, informational, appraisal, and emotional dimensions of social support (Cronbach's α = 0.95) (ref. 31). Baseline recent negative life events included physical abuse, unwanted pregnancy, abortion or miscarriage, major financial difficulties, and serious problems at work or in school. The presence of any adverse events (vs. no event) was reported. NPHS participants were also asked whether, as a child or teenager, they experienced parental divorce, a lengthy hospital study, prolonged parental unemployment, frequent parental alcohol or drug use, physical abuse, and being sent away from home. An affirmative answer to any question was determined as having been exposed to childhood traumatic events.

Statistical analysis

Respondents were included in the analyses if they had complete information on depression and weight status at baseline. Data screening and examination of descriptive statistics and bivariate associations were performed. Cross-sectional prevalence point of MDE was tabulated for each survey cycle by weight status.

Two approaches were used to analyze the incidence risk of major depression. We first generated Kaplan–Meier survival curves for depression onset by gender and obesity status. We then used the Cox proportional hazards regression model to calculate hazard ratios (HRs) of major depression incidence by obesity status. The Cox model is the preferred model for analyzing NPHS data because it takes into account different lengths of follow-up and does not require assumptions about the distribution of survival time. Participants with MDE at baseline were excluded from incidence analyses in order to identify new cases of depression. Subjects with incomplete information on depression for one survey cycle were right-censored at that cycle. As interviews were conducted every 2 years, we chose to censor these individuals early because a hiatus of ≥4 years might contain an MDE which would be missed. We tested three models: (i) unadjusted model, (ii) age adjustment only, (iii) adjustment for all baseline sociodemographic and health variables. Regression analyses were stratified by gender because sex differences in depression outcomes are known to exist. Stratification by age group was not possible because of sparse data in some cells. Preliminary analysis found the obesity–age interaction term nonsignificant in the model.

Data were weighted to adjust for differential response rates and variation in probabilities of selection into the sample. The bootstrap resampling technique for precision of estimates was used to account for complex sampling design. Analyses were conducted in SAS, version 9.1 (SAS Institute, Cary, NC), and bootstrapping techniques were done in STATA, version 10 (StataCorp, College Station, TX).

Results

Our initial sample consisted of 11,220 participants with complete information on baseline weight status and depression score. From our baseline sample of adults (n = 13,618), 752 participants were excluded because of missing BMI information, 310 were excluded because of underweight status, and 1,336 were excluded because of missing MDE information. Figure 1 shows the cross-sectional proportion of MDE by obesity status for each survey cycle. Depression prevalence was slightly higher for obese individuals than their normal/overweight counterparts except at baseline (cycle 1) (sample size for cycle 1: 11,220; cycle 2: 10,500; cycle 3: 9,775; cycle 4: 8,862; cycle 5: 7,932; cycle 6: 7,437; cycle 7: 6,417). Prevalence remained relatively stable over time.

Figure 1.

Major depression episode (MDE) prevalence of adults in the National Population Health Survey (NPHS) by obesity status for each NPHS cycle.

At baseline, 780 subjects reported MDE and were therefore excluded from incidence analysis; data from the remaining 10,545 participants (4,928 males, 5,617 females) were examined. Preliminary analyses revealed a higher risk of mortality for obese individuals compared to normal/overweight individuals, and for males more than females, after controlling for age (mortality HR for obese compared to normal/overweight was 1.4 (95% confidence interval (CI) 1.2–1.8) in men and 1.0 (95% CI 0.8–1.3) in women).

Women participants were less likely to be single compared to men, and they reported more childhood traumatic events but slightly better social support (Table 1). Regarding health characteristics, women were on average less active than men and less likely to be smokers. Women were also more likely to rate their health as fair or poor and to suffer from a somatic comorbidity compared to men. Approximately 13% of our sample was classified as obese (13.8% of women and 12.7% of men).

Table 1.  Descriptive characteristics of participants included in analyses
inline image

Figure 2 illustrates Kaplan–Meier survival curves by obesity status and gender. After 12 years, men were half as likely as women to have suffered from at least one MDE (8.5% normal/overweight and 7.2% obese men vs. 14.7% normal/overweight and 15.0% obese women).

Figure 2.

Kaplan–Meier survival curve of major depression episode by gender and obesity status for adults in the National Population Health Survey.

Results from the Cox proportional hazards models are shown in Table 2. Unadjusted HRs showed a negative effect of obesity on MDE in men (HR: 0.70, CI: 0.49–0.99) and a nonsignificant association in women (1.03, CI 0.78–1.36). Adjusting for sociodemographic and health variables, HRs remained relative unchanged (for men: 0.71, CI 0.51–0.98; for women: 1.03, CI 0.84–1.26). In a subgroup analysis by obesity severity (results not shown), obesity had a nonsignificant effect in severely obese men (fully adjusted HR (AHR) 0.89, CI 0.46–1.69) but a significant negative effect in obese men (AHR 0.67, CI 0.46–0.96); obesity had a nonsignificant effect in obese and severely obese women (AHR 0.99, CI 0.78–1.25 and AHR 1.12, CI 0.80–1.58, respectively).

Table 2.  Hazard ratio and 95% confidence interval of major depression episode by gender and obesity status
inline image

Discussion

The main goal of this study was to investigate whether obesity was a risk factor for depression, in a representative community sample followed for 12 years. According to our results, obesity was a significant (negative) predictor of major depression in men but not in women, even after controlling for baseline demographic variables and health factors. Although inconsistent with most prospective findings (13,14,15,16,17,18,19), our results are nevertheless supported by some cross-sectional studies (32,33).

Several methodological differences could explain our divergent results. We provided measures of incidence whereas others reported prevalence, and we calculated HRs instead of odds ratios. Moreover, we controlled for confounding by physical and psychosocial factors, which have been previously excluded by some. Recent analyses have highlighted the role of these covariates in linking obesity to depression. For example, Carr et al. (34) found that obesity was associated with more frequent negative affect and less frequent positive affect; but this association was reversed after controlling for physical health and interpersonal relationships. The authors concluded that obesity was not inherently distressing but that the physical and interpersonal consequences of excess weight were instead driving the association.

Our regression point estimates suggested a negative effect of obesity on depression in men and a weak positive effect in women. Gender differences in the association between obesity and depression have been described in previous cross-sectional studies (35,36). Carpenter et al. (35) reported obesity to be negatively associated with depression in men (odds ratio 0.63, CI: 0.60–0.67) but positively in women (odds ratio 1.37, CI: 1.09–1.73). Among the three prospective studies that stratified by gender: one found a positive association for women but nonsignificant for men (17), the other reported a positive association for men but nonsignificant for women (16), and the last presented positive results for both genders (19). Psychosocial factors might explain gender-specific effects. Obese males are less likely than their female counterparts to experience obesity-related stigma and discrimination (37). Obese men also suffer from fewer body image disturbances than obese women (37), particularly in societies that values heavier frames as part of the muscular male ideal.

For each survey wave, we found that obese individuals were more depressed than their nonobese counterparts, and that prevalence of MDE remained relatively stable for both groups. This might indicate that little change in depression status occurred for obese participants who stayed obese. It might also suggest that participants who switched from obese to nonobese (or vice versa) also switched depression status proportionately with the depression prevalence of their new weight status group. If obesity is not a predictor of depression, then two explanations are plausible. First, depression might be a risk factor for obesity. Prior longitudinal research supports this hypothesis (15,38) but a recent paper based on the first six cycles of NPHS concluded that depression was not a significant risk factor for obesity (39). The second possibility is that there is an association between obesity and depression (unidirectional or reciprocal) but this takes place early in the weight gain process during a time frame that was not captured by our study. This can be understood not only in terms of age-specific effects of obesity on depression but can also be thought of in terms of study design. Incidence analysis requires that persons with baseline depression be excluded from analyses. However, by excluding obese (and nonobese) respondents with baseline depression, this may have eliminated a subgroup in which obesity was most linked to depression.

Some methodological limitations should be kept in mind. Differential loss to follow-up could have impacted our results. Preliminary analyses demonstrated that obese individuals, especially those with severe obesity and obese men, had a higher risk of mortality than average-weight subjects when controlling for age. It is also possible that depressed subjects were more likely to be lost to follow-up than nondepressed subjects. Further, BMI was based on self-reported data which, evidence has shown, may underestimate the prevalence of obesity (40). Approximately 13% of our sample was classified as obese whereas a previous report based on measured anthropometry estimated the Canadian average to be 23% (ref. 2). This suggests important biases in self-reporting height and weight in our sample.

In addition, we did not assess depression by clinical diagnosis. Although the CIDI-SFMD is a well-validated instrument, the use of this scale may result in an overestimation of major depression because of its vulnerability to false positives (25) and because obesity may cause somatic symptoms that were attributed to depression (e.g., feeling tired). The NPHS does not collect information on lifetime history of depression and obesity. Consequently, though every effort was made to include only incident cases in our risk analyses, it is possible that participants with a history of depression were included in our sample. Further, some baseline variables such as weight status and marital status could have changed over time. As a result, time-varying covariates could have distorted some measures of association. Lastly, residual confounding might have biased our results. For example, psychiatric conditions are often comorbid and some psychiatric illnesses (such as binge-eating disorder) might partly account for the link from obesity to depression (22).

Our study had several methodological strengths. It provided longitudinal data from a large population-based sample. Depression symptoms were measured with a validated depression scale. The study also controlled for known confounders, such as chronic conditions and physical activity levels, which have been lacking from some previous analyses. Further, we attempted to capture the true incidence rate by excluding individuals with depression at baseline. Reducing the incidence of disease is the objective of primary prevention and the role of public health but incidence estimation has not always been feasible in prior longitudinal studies.

Conclusion

Across the 12-year span of this study, obese individuals had an overall higher prevalence of depression compared to their nonobese counterparts. Incidence analysis, however, suggests that obesity negatively predicted subsequent depression in men and did not predict depression in women. These results moderate findings from previous longitudinal studies that found obesity to positively predict depression later in life. Gender and obesity severity appeared to play a role in the determination of depression in obese individuals. As previously recommended (21), future studies should bear in mind these potential moderators in their analyses. Additionally, researchers might consider accounting for obesity onset and weight change as well as other time-varying covariates when examining the role of obesity in depression.

Acknowledgments

We thank Danit Nitka for help with manuscript preparation. Funding for this project was provided by the Canadian Institutes for Health Research (CIHR grant MOP-79464).

Disclosure

The authors declared no conflict of interest.