Challenging the “jolly fat” hypothesis among older adults: High body mass index predicts increases in depressive symptoms over a 5-year period

Abstract

Several investigators have observed lowered risk of depression among obese older adults, coining the “jolly fat” hypothesis. We examined this hypothesis using baseline and a 5-year follow-up body mass index, depressive symptoms, and covariates from 638 community-based older adults. High objectively measured body mass index and functional limitations predicted increased future depressive symptoms. However, symptoms did not predict future body mass index. Self-reported body mass index showed similar associations despite underestimating obesity prevalence. Results did not differ on the basis of gender. Results for this study, the first longitudinal reciprocal risk analysis between objectively measured body mass index and depressive symptoms among older adults, do not support the “jolly fat” hypothesis.

Keywords

depression epidemiology geriatrics obesity physical function

Recent meta-analytic evidence suggests the existence of a relationship between obesity and depression (De Wit et al., 2010). However, the directional nature of this relationship remains less clear among differently aged and gendered samples. Among longitudinal studies, there is some evidence that depression status predicts later obesity. In a meta-analysis of 17 samples, Blaine (2008) estimated a large effect size of baseline depression status on subsequent obesity when controlling for baseline obesity and other covariates. However, the only study composed entirely of older adults (age > 50 years; DiPietro et al., 1992) found that baseline depression was associated with marginally lower risk of future obesity, results consistent with the “jolly fat” hypothesis which posits that there are protective effects of obesity on risk of depressive symptoms (Crisp et al., 1980; Crisp and McGuiness, 1976; Palinkas et al., 1996).

Results supportive of the “jolly fat” hypothesis have primarily been observed for older cohorts, have sometimes been stronger for males than for females, and have been obtained mostly from cross-sectional studies. Indeed, a major conclusion from a systematic review of epidemiological research is that there is a dearth of high-quality longitudinal studies which measure the effect of obesity on future depression status (Atlantis and Baker, 2008). Of the four studies which met their quality inclusion standards, three were composed of older adults, all from the same sample in Alameda County, CA (Roberts et al., 2000, 2002). In contrary to the “jolly fat” hypothesis, both 1- and 5-year longitudinal analyses revealed that self-reported obesity predicted significantly higher risk of depressive symptoms even with control for an array of covariates. These results have been supported by more recent longitudinal analysis of older adults, though findings vary by obesity measurement used. While studies using self-reported body mass index (BMI_SR) have found obesity is increasing risk of depressive symptoms among women (Pan et al., 2012), but not in men (Xiang and An, 2015), a study using objectively measured body mass index (BMI_O) found this association only among men (Vogelzangs et al., 2010).

To our knowledge, only two longitudinal studies have systematically assessed the reciprocal effects of depression and obesity among older adults (Pan et al., 2012; Roberts et al., 2003). Both studies found that obese individuals were at a higher risk of reporting clinical levels of depressive symptoms, anti-depressant use, or receiving a diagnosis of depression 5–10 years later. This was true even with control for baseline depression status and a range of covariates. Conversely, although Pan et al. (2012) found evidence of a bidirectional relationship, Roberts et al. (2003) did not find that depressed individuals were at a higher risk of obesity 5 years later. Compared with prior longitudinal studies on depression and obesity, the covariate models employed by Roberts et al. and Pan et al. were considerably more extensive.

Although these studies have methodological strengths, both were also limited to using self-reported obesity. Prior studies have found systematic differences between BMI_O and BMI_SR. Specifically, individuals tend to overestimate height and underestimate weight, which results in a likely underestimation of obesity prevalence, particularly among older adults (Keith et al., 2011; Kuczmarski et al., 2001). Additionally, Keith et al. (2011) noted that BMI_SR failed to consistently predict mortality outcomes which were associated with BMI_O. As such, it is possible that BMI_SR and BMI_O may associate differently with other psychobiological outcomes.

This study uses objective measurements of BMI from the Maine-Syracuse Longitudinal Study (MSLS) to investigate the reciprocal effects of depressive symptoms and BMI and to further test the jolly fat hypothesis among older adults. We hypothesize that BMI will be positively associated with concurrent and future depressive symptoms; that is, will be seen with cross-sectional and longitudinal analyses.

Methods

Participant selection

The MSLS includes seven waves of demographic, cardiovascular risk factor, and cognitive performance data gathered over a period of 35 years employing a sample of community-dwelling adults residing in Central New York. Since the original MSLS baseline (1976), individuals have been excluded based on diagnosed alcoholism, diagnosed psychiatric disorder, or inability to speak English. In this and other subsequent MSLS studies, additional exclusionary criteria have been included on a study-by-study basis. The University of Maine Institutional Review Board approved this study (reference number: 2005-07-04) and informed consent was obtained from all participants.

Attrition

At baseline, older adults were selected for this analysis (n = 893, ⩾50 years). Participants no longer residing in New York State were lost to the study due to changes in research funding limiting travel of the investigative team out of state. Of the baseline participants, 71.4 percent responded again and were eligible at wave 7 follow-up (n = 638). Additionally, participants were excluded from regression analysis for missing depressive symptom data (n = 27), for missing objectively measured (BMI_O; n = 42) or self-reported BMI (BMI_SR; n = 117), or baseline covariate data (n = 25).

Compared with individuals who were lost to follow-up or excluded, individuals retained at follow-up were 2.52 years younger (confidence interval (CI): 1.04–4.00, p = .002), 2.5 points lower on the Zung Depression Inventory (ZDI; CI: 1.06–3.95, p = .001), and had 0.48 fewer physical deficits (CI: 0.30–0.66, p < .001) at baseline. However, neither group differed in baseline educational level, marital status, number of recent life changes, social activity/isolation level, number of chronic illnesses, or BMI_SR/O (p > .05). Table 1 presents the baseline descriptive statistics for participants retained at follow-up.

Table 1.

Wave 6 baseline variables.

Variable	Value	Female (%)	Male (%)	p < .05
Age (years)	50–59	32.0	29.6
	60–69	25.7	35.6
	70–79	32.8	26.4
	80 or older	9.5	8.4
Years of education ⩾ 12		94.6	95.6
Married/cohabitating		54.5	79.6	*
Recent life events		76.8	66.4	*
Social activity	Mean (SD)	21.7 (5.3)	20.3 (4.8)
Moderate/high isolation		30.4	18.3	*
Zung Depression Inventory	<40	43.8	47.8
	40–49	31.8	34.9
	50–59	13.7	18.3
	⩾60	6.1	3.6
Obesity rate (objective)		38.3	39.6
Obesity status (self-report)		29.0	29.5
Chronic conditions	0	34.8	32.8
	1	19.8	57.8
	2	16.0	20.0
	3 or more	29.4	22.4
Physical deficits	None	53.6	60.4
	Low	15.7	16.8
	Medium	14.7	6.8	*
	High	16.0	16.0

SD: standard deviation.

p < .05 refers to differences in category prevalence by gender.

Study design

This study uses BMI, depressive symptoms, and covariate information from wave 6 of the MSLS (2001–2005) as baseline data. Approximately 5 years later (2006–2010) at wave 7 follow-up, BMI and depressive symptoms were obtained. Objective measurement of height, weight, and physical performance obtained at waves 6 and 7 allowed for more comprehensive prospective models than were available in previous waves.

Measures of obesity and depressive symptoms

Objectively measured height and weight at baseline and follow-up was assessed in the lab without shoes or heavy clothing to calculate BMI (kg/m²). In addition, self-reported height and weight were used to calculate an alternative index of BMI. BMI was analyzed as a continuous outcome and baseline adjustment variable, and obesity was defined based on the World Health Organization (WHO, 2010) obesity standard (BMI ⩾ 30).

Baseline and outcome depressive symptoms were measured with the ZDI (Zung, 1965). The ZDI is composed of 20 self-reported, Likert scale items ranging in frequency from 1 “a little of the time” to 4 “most of the time” and summed to create a total raw score. Raw scores were then divided by 0.8, resulting in an index score range from 25 to 100 (Hunter and Murphy, 2011). At baseline, ZDI items showed good internal consistency (α = .80). The ZDI has been reported to lack sufficient sensitivity to act as a categorical diagnostic measure (Thurber et al., 2002). As such, index scores were used to represent the continuum of depressive symptom severity rather than major depressive disorder incidence. Individuals missing more than two items were excluded from analysis, while scores for two or fewer missing items were imputed based on the mean of available ZDI data.

Covariates

Covariates were selected based primarily on the extensive models used by Roberts et al. (2003), their theoretical relevance as posited by Atlantis and Baker (2008), and their associations with BMI and ZDI within our own sample. This included adjustments for age, gender, education in years, marital status (married/living w/partner vs other), social isolation (no close friendships and/or co-residents vs other), social activity (see Robbins et al., 1994), recent life events (yes/no), physical functioning, and chronic medical conditions.

For the chronic illness composite, participants completed questionnaire items asking whether their doctor had told them whether they had a range of medical issues during the past 6 years. This included summing occurrence data on heart trouble, high blood pressure, asthma, bronchitis/emphysema, rheumatoid arthritis, diabetes, stroke, cancer, cataracts, and circulatory issues.

The physical functioning deficit composite was measured with three timed performance tasks. Participants were asked to (1) stand up three times from a hard-backed chair with their arms folded, (2) walk forward 10 feet, turn 180°, and walk back to their starting position, (3) from a standing position, complete a full 360°turn. Participants unable to complete a task or whose task completion time was among the slowest quartile in the baseline sample were considered to have a deficit in that area of performance. Total deficit scores ranged from “0” (no deficit) to “3” (high deficit). Prior research has shown that a performance deficit in any of these tasks is predictive of future functional dependence in activities of daily living (Gill et al., 1995).

Self-reported family income, smoking status, and ethnicity were all collected and considered for analysis but ultimately dropped from our final models. Family income was considered because it has been found to be related to depressive symptoms (Mendes De Leon et al., 1994). However, it was dropped from the analysis because (1) a large number (n = 106) of participants indicated that they “would rather not answer” this item and (2) when tested as a model covariate, family income did not change the results reported below. Smoking status was considered based on the quality assessment criteria by Atlantis and Baker (2008) but was dropped due to a lack of observed association with BMI_SR/O or ZDI scores. Although ethnicity may moderate the relationship between obesity and depressive symptoms (Xiang and An, 2015), detailed group analysis was not conducted due to a lack of sufficient sample size for non-White participant groups (Black/Hispanic/Asian/American Indian combined; n = 42).

Statistical analysis

Hierarchical multivariate regression models were used to explore the cross-sectional and longitudinal association between BMI and ZDI with adjustment for covariates. Baseline (wave 6) values for ZDI and BMI, respectively, were used as covariates for models in which wave 7 ZDI or wave 7 BMI was the outcome of interest. Model 1 of the cross-sectional and longitudinal analyses included baseline age, sex, education, and marital status as covariates. Model 2 contained all covariates from Model 1 in addition to life events, social isolation, and social activity. Model 3 included all covariates from Model 2 in addition to chronic health conditions and physical performance deficits.

Model variables met assumptions for parametric multivariate regression (Durbin-Watson; 1.94 ⩾ d ⩽ 2.04) and minimal collinearity was observed (maximum variance inflation factor ⩽ 1.97). Tests for non-linear associations between BMI and ZDI were conducted with the addition of quadratic predictor terms with adjustment for linear terms. Neither inclusion of quadratic terms for wave 6 BMI_SR/O (p > .25) nor wave 6 ZDI (p ⩾ .09) improved fit across all models. Statistical analyses were conducted with SPSS software (version 23.0).

Results

Objective versus self-reported BMI

Bivariate and paired-samples t-test analysis were used to examine the convergent validity between objective and self-reported height and weight at baseline for participants who completed both waves 6 and 7. Participants both significantly underestimated their weight (t(601) = −10.71, p < .001) by an average of 2.10 kg (95% CI = 1.72, 2.48 kg) and overestimated their height (t(574) = 6.02, p < .001) by an average of 0.94 cm (95% CI = 0.63, 1.25 cm). However, self-reported and objective measurement of height (r = .93, p < .001) and weight (r = .97, p < .001) were also highly associated. BMI_SR obesity prevalence (29.2%) was significantly lower than what would be expected ( χ ² (1, 572) = 348.57, p < .001) based on the BMI_O obesity prevalence in our sample (38.8%).

Model analyses

Results of analyses for cross-sectional and longitudinal multivariate regression models are presented in Table 2. With regard to the cross-sectional association between baseline BMI and ZDI, across all models the adjusted R² values ranged from .07 to .20. BMI_O predicted higher ZDI in Model 1 and Model 2 (ps < .01), but not in Model 3 (p = .16). BMI_SR predicted higher ZDI in Model 1 (p < .01), but not in Model 2 or Model 3 (ps > .10).

Table 2.

Regression coefficients showing relations between BMI and depressive symptoms.

Cross-sectional
	Objective BMI^a			Self-report BMI^b
	b	SE	p-value	B	SE	p-value
Model 1	0.16	0.06	.008	0.17	0.06	.006
Model 2	0.16	0.06	.005	0.15	0.06	.11
Model 3	0.08	0.06	.16	0.06	0.06	>0.25
Longitudinal: BMI predicting depressive symptoms
Objective BMI^c				Self-report BMI^d
	b	SE	p-value	B	SE	p-value
Model 1	0.23	0.05	<.001	0.24	0.06	<.001
Model 2	0.23	0.05	<.001	0.24	0.06	<.001
Model 3	0.20	0.05	<.001	0.20	0.06	<.001
Longitudinal: depressive symptoms predicting BMI
Objective BMI^e				Self-report BMI^f
	b	SE	p-value	B	SE	p-value
Model 1	0.01	0.01	>.25	−0.01	0.02	>.25
Model 2	0.02	0.01	>.25	<0.01	0.02	>.25
Model 3	0.02	0.01	.20	<0.01	0.02	>.25

BMI: body mass index; SE: standard error.

Model 1 = age, sex, education, and marital status. Outcome variable adjusted at baseline if longitudinal model; Model 2 = Model 1 + recent life events, social isolation, and social activity; Model 3 = Model 2 + chronic illness and physical performance deficits.

n = 810.

n = 761.

n = 582.

n = 544.

n = 540.

n = 495.

Regression models examining the longitudinal association between baseline BMI and 5-year follow-up ZDI included adjustment for baseline ZDI as well as the other covariates. Across all models, adjusted R² values ranged from .44 to .46. Baseline ZDI was strongly correlated with follow-up ZDI (r = .65, p < .001), accounting for 26.6–35.6 percent of the total variance across all models. Both higher BMI_O and BMI_SR at baseline predicted higher ZDI at follow-up across all three regression models (ps < .001).

In addition to BMI, the covariates age and physical functioning were significant predictors of follow-up ZDI. For instance, in the BMI_O models (1) older age was associated with higher follow-up ZDI in Model 1 (b = 0.07, standard error (SE) = 0.03, p = .029) and in Model 2 (b = 0.07, SE = 0.03, p = .023) but not when adjusting for physical ailments and physical functioning in Model 3 (p > .25) and (2) baseline physical functioning deficits predicted higher follow-up ZDI (b = 0.95, SE = 0.29, p = .001) in the full regression model (Model 3).

Regression models examining the longitudinal association between baseline ZDI and follow-up BMI included adjustment for baseline BMI and the other covariates. Across all models, adjusted R² values ranged from .68 to .78. Baseline BMI was highly correlated with follow-up BMI (r_objective = .89 and r_SR = .83), accounting for 56.0–67.7 percent of the total variance across all models. For both BMI variables across all three models, baseline ZDI failed to predict follow-up BMI (b ⩽ 0.02, p ⩾ .20).

Supplementary analyses were conducted to examine gender differences, given that evidence supporting the “jolly fat” hypothesis has been found predominantly for males, that is, others have found that obese men and women have differential levels of risk of depressive symptoms (Atlantis and Baker, 2008; Xiang and An, 2015). Adjusted for all other cross-sectional (wave 6) Model 3 covariates, females reported higher ZDI scores than males (in BMI_O model, b = 2.18, SE = 0.66; in BMI_SR model, b = 2.16, SE = 0.68, ps < .01). However, results for BMI*gender interaction terms included in the BMI_SR/O cross-sectional and longitudinal models were non-significant (ps > .25) when predicting either baseline ZDI or longitudinal change in ZDI.

Discussion

In this longitudinal study, we tested the “jolly fat” hypothesis in the context of assessing the relationship between depressive symptoms and BMI among a community sample of older adult men and women. We found that older adults with higher baseline BMI reported increases in depressive symptoms over a 5-year period. This was true even when adjusting for baseline depressive symptoms, age, sex, education, marital status, social isolation, social activity, chronic illness, and physical functioning. These results were unchanged whether self-reported or objectively measured BMI was used, thus validating the self-report findings of Roberts et al. (2003).

Strengths

This study adds methodological strength to the currently limited longitudinal research on obesity and depression among older adults, much of which is lacking objective measurement of BMI (Atlantis and Baker, 2008). Although we found in our sample that objective measurement and self-reported height and weight were highly correlated and similar in predicting baseline level and change in depressive symptoms, there was also a small but consistent tendency for participants to underestimate weight while overestimating height, resulting in a considerable underestimation of the true obesity prevalence within our sample. While using self-reported BMI as a continuous variable is likely justifiable for associative analysis, logit models, in which dependent variables are categorical (i.e. self-reported obesity; Cox, 1958), may experience a considerable loss of predictive power due to misclassification. Additionally, researchers studying obesity prevalence are encouraged to either use objective BMI or to consider statistical adjustments for reporting bias among older adults.

Regression models also incorporated a composite measure of physical functioning which consisted of objectively timed performance tasks predictive of difficulty in activities of daily living (Gill et al., 1995), which up until this point has been missing from associative research examining depression and obesity. Additionally, although Roberts et al. (2003) adjusted for self-reported functional limitations (e.g. bathing and dressing), they did not include objective measurement of physical functioning. Prior studies have reported moderate overlap but distinct differences between self-reported and objectively measured physical functioning. Until now, studies have been unable to account for the potentially confounding associations between self-reported physical functioning and well-known dimensions of depression such as self-efficacy and mastery rather than actual performance (Becofsky et al., 2013; Kempen et al., 1996).

Finally, in this study, we observed changes in BMI and depressive symptoms over time by including both baseline and follow-up levels in our models. This allowed us to simultaneously test the bidirectional associations between BMI and depressive symptoms among older adults, adding to the scant research with such a design (Pan et al., 2012; Roberts et al., 2003). Although there is evidence for symptoms of depression being both an outcome and predictor of incident obesity (Atlantis and Baker, 2008; Blaine, 2008), to our knowledge no meta-analysis has statistically accounted for potential publication bias. As such, the strength of pooled evidence for bidirectionality as assessed by non-reciprocal models may be overestimated. In the context of a reciprocal analysis and a sufficiently powered sample, we observed significant directional associations with BMI as predictor, but not as an outcome of baseline depressive symptoms.

Limitations

It should be noted that our sample was both highly educated and predominately White. This limited our ability to conduct a more thorough analysis of socioeconomic and ethnic group differences which may moderate the association between obesity and depressive symptoms (Xiang and An, 2015). However, with regard to our main predictor/outcome variables, obesity and depressive symptoms, our sample did not vary considerably from national estimates among older adults (Center for Behavioral Health Statistics and Quality, 2015; Ogden et al., 2014). For additional health demographic information of the MSLS sample, see Crichton et al. (2014).

Regarding attrition, participants who were available at follow-up tended to be younger, less likely to report depressive symptoms, and less functionally impaired compared with those who were unavailable. Although functional impairment and to a lesser extent advanced age were associated with increased depressive symptoms in our models, it is possible that the magnitude of their effects were partially attenuated. Similarly, the effect of depressive symptoms on BMI may have been partially attenuated; however, there was little evidence for a bidirectional trend.

Our results indicate that higher BMI may contribute to increases in depressive symptoms over time. Despite our findings, it is possible that chronic illness and low physical functioning, which tend to increase risk of depressive symptoms, may contribute to weight loss among aging adults as well as weight gain due to lowered physical activity and loss of muscle tone (St-Arnaud-McKenzie et al., 2010). However, only one individual in our wave 6 sample was measured to be underweight by objective measurement (BMI < 18.5). As such, our conclusions are limited to normal and overweight/obese older adults and our results may mask associations between low-weight and depressive symptoms mediated by nutrient deficits as a result of appetite loss (i.e. B12, folate; Kim et al., 2008). However, our results are consistent with the homeostasis theory of obesity which posits that some individuals may over-consume energy-dense, low-nutrient foods (e.g. sugar-sweetened beverages) as a form of emotion regulation (Marks, 2016). These homeostatic disturbances extended over the course of several years may contribute to continued weight gain and declines in physical health which in turn increases risk of experiencing depressive symptoms.

Conclusion and future directions

The results of this study clearly do not support the “jolly fat” hypothesis among older adults. This is an important finding because the notion that obesity is protective may lead to behaviors that are not consistent with good health practices. Higher BMI was related to greater risk, not less risk, of depressive symptoms. There was no indication that the BMI-associated risk of depressive symptoms interacted with gender. Studies that obtained results supportive of the “jolly fat” hypothesis among older adults, and predominantly for males, were conducted two or more decades ago (Crisp et al., 1980; Crisp and McGuiness, 1976; DiPietro et al., 1992; Palinkas et al., 1996). It is possible that increased awareness of the diversity and severity of health risks associated with obesity has made positive associations with depressive symptoms more likely not only through resultant individual perceptions of threat to health but also through an increased prevalence of perceived weight-based discrimination overall among US adults (Andreyeva et al., 2008). Although women likely experience greater weight-related discrimination overall compared with men, more recent analysis suggests that these differences may not extend into older adulthood (Eichner, 2002; Puhl et al., 2008; Roehling et al., 2007). Finally, recent findings do suggest that older adults and especially the oldest-old (⩾80 years) experience considerably lower rates of weight-based discrimination compared with younger adults (Jackson et al., 2015; Puhl et al., 2008), which may explain previously inconsistent findings between younger and older samples.

Conversely, we did not find any association between baseline depressive symptoms and BMI change 5 years later, in contrary to the 10-year bidirectional results reported by Pan et al. (2012). It is possible that behaviors associated with depressive symptoms (e.g. appetite disturbance and lowered activity) require a longer incubation period in order to cause significant weight gain among older adults. In our own 5-year models, BMI was measured to be highly stable, which may have masked the potential for more subtle weight gain over a longer duration.

Regardless of gender, higher baseline BMI was associated with a net increase in depressive symptoms. Although this latter finding is similar to the longitudinal results of Roberts et al. (2003), it differs from other cross-sectional and longitudinal results that have shown obese women to be at a higher risk of depressive symptoms in comparison to men (Atlantis and Baker, 2008; Xiang and An, 2015). Our results indicate that although older adult women may be at a higher risk of depressive symptoms, their risk of depressive symptoms as a function of BMI does not differ significantly from that of older adult men. It is possible that gender differences in weight-based discrimination diminish in older adulthood or that other forms, such as age-based discrimination, become more salient (Puhl et al., 2008). Future research should investigate the potential mediating role of weight-based discrimination among obese older adults in the development of depressive symptoms. In a review recently published by Marks (2015), only one state (i.e. Michigan) was identified as having laws prohibiting weight discrimination. However, this also opens up the possibility of tracking the efficacy of such legislation on reducing the prevalence of weight-based discrimination and incident depression among overweight/obese individuals.

Objective physical impairment also independently predicted changes in depressive symptoms and its combined association with chronic illness attenuated much of the association between depression and aging. Future analysis should explore the possible mediating effect of physical functioning interventions on the relationship between obesity and depressive symptoms. It is possible that older obese individuals, who are more likely to experience depressive symptoms, may benefit from interventions targeting limitations in physical functioning. Increased physical activity, which has been shown to reduce the risk of depressive symptoms among obese individuals (King et al., 2013), may play a key role by slowing the rate of functional impairment independent of weight loss (Chalé-Rush et al., 2010).

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Supplementary Information

Additional information relating to this article may be obtained by request by emailing the primary author.

References

Andreyeva

Puhl

Brownell

(2008) Changes in perceived weight discrimination among Americans, 1995–1996 through 2004–2006. Obesity 16(5): 1129–1134.

Atlantis

Baker

(2008) Obesity effects on depression: Systematic review of epidemiological studies. International Journal of Obesity 32(6): 881–891.

Becofsky

Baruth

Wilcox

(2013) Physical functioning, perceived disability, and depressive symptoms in adults with arthritis. Arthritis 2013: 525761. Available at: http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=3777208&;tool=pmcentrez&rendertype=abstract

Blaine

(2008) Does depression cause obesity? A meta-analysis of longitudinal studies of depression and weight control. Journal of Health Psychology 13(8): 1190–1197.

Center for Behavioral Health Statistics and Quality (2015) Behavioral health trends in the United States: Results from the 2014 national survey on drug use and health. Available at: http://www.samhsa.gov/data/sites/default/files/NSDUH-FRR1-2014/NSDUH-FRR1-2014.pdf

Chalé-Rush

Guralnik

Walkup

et al . (2010) Relationship between physical functioning and physical activity in the lifestyle interventions and independence for elders pilot (LIFE-P). Journal of the American Geriatrics Society 58(10): 1918–1924. Available at: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2952066/

Cox

(1958) The regression analysis of binary sequences (with discussion). Journal of Royal Statistical Society: Series B 20: 215–242.

Crichton

Elias

Davey

et al . (2014) Cardiovascular health and cognitive function: The Maine-Syracuse longitudinal study. PLoS ONE 9(3): e89317. Available at: http://www.ncbi.nlm.nih.gov/pubmed/24595096 (accessed 26 July 2016).

Crisp

McGuiness

(1976) Jolly fat: Relation between obesity and psychoneurosis in general population. British Medical Journal 1(6000): 7–9. Available at: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1638245/

10.

Crisp

Queenan

Sittampaln

et al . (1980) “Jolly fat” revisited. Journal of Psychosomatic Research 24(5): 233–241. Available at: https://www-sciencedirect-com.web.bisu.edu.cn/science/article/pii/0022399980900136

11.

De Wit

Luppino

Van Straten

et al . (2010) Depression and obesity: A meta-analysis of community-based studies. Psychiatry Research 178(2): 230–235. Available at: https://www-sciencedirect-com.web.bisu.edu.cn/science/article/pii/S016517810900170X (accessed 27 May 2015).

12.

DiPietro

Anda

Williamson

et al . (1992) Depressive symptoms and weight change in a national cohort of adults. International Journal of Obesity 16(10): 745–753.

13.

Eichner

(2002) Apolipoprotein E polymorphism and cardiovascular disease: A HuGE review. American Journal of Epidemiology 155(6): 487–495. Available at: http://aje.oxfordjournals.org/content/155/6/487.full (accessed 23 March 2016).

14.

Gill

Williams

Tinetti

(1995) Assessing risk for the onset of functional dependence among older adults: The role of physical performance. Journal of the American Geriatrics Society 43(6): 603–609.

15.

Hunter

Murphy

(2011) Zung Self-Rating Depression Scale. In: Kreutzer

DeLuca

Caplan

(eds) Encyclopedia of Clinical Neuropsychology. New York: Springer, pp. 2746–2747. Available at: https://dx-doi-org.web.bisu.edu.cn/10.1007/978-0-387-79948-3_2017

16.

Jackson

Steptoe

Beeken

et al . (2015) Perceived weight discrimination in England: A population-based study of adults aged ⩾ 50 years. International Journal of Obesity 39(5): 858–864. Available at: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4309989/

17.

Keith

Fontaine

Pajewski

et al . (2011) Use of self-reported height and weight biases the body mass index-mortality association. International Journal of Obesity 35: (3): 401–408. Available at: https://dx-doi-org.web.bisu.edu.cn/10.1038/ijo.2010.148

18.

Kempen

Steverink

Ormel

et al . (1996) The assessment of ADL among frail elderly in an interview survey: Self-report versus performance-based tests and determinants of discrepancies. Journals of Gerontology Series B: Psychological Sciences and Social Sciences 51(5): P254–P260.

19.

Kim

J-M

Stewart

Kim

et al . (2008) Predictive value of folate, vitamin B12 and homocysteine levels in late-life depression. The British Journal of Psychiatry: The Journal of Mental Science 192(4): 268–274. Available at: http://bjp.rcpsych.org/content/192/4/268 (accessed 13 April 2016).

20.

King

Kalarchian

Steffen

et al . (2013) Associations between physical activity and mental health among bariatric surgical candidates. Journal of Psychosomatic Research 74(2): 161–169. Available at: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3556899/

21.

Kuczmarski

Najjar

(2001) Effects of age on validity of self-reported height, weight, and body mass index: Findings from the third national health and nutrition examination survey, 1988–1994. Journal of the American Dietetic Association 101(1): 28–34. Available at: https://www-sciencedirect-com.web.bisu.edu.cn/science/article/pii/S0002822301000086

22.

Marks

(2015) Homeostatic theory of obesity. Health Psychology Open 2(1): 2055102915590692. Available at: http://hpo.sagepub.com/lookup/doi/10.1177/2055102915590692 (accessed 27 September 2016).

23.

Marks

(2016) Dyshomeostasis, obesity, addiction and chronic stress. Health Psychology Open 3(1): 2055102916636907. Available at: http://hpo.sagepub.com/lookup/doi/10.1177/2055102916636907 (accessed 27 September 2016).

24.

Mendes De Leon

Rapp

Kasl

(1994) Financial strain and symptoms of depression in a community sample of elderly men and women: A longitudinal study. Journal of Aging and Health 6(4): 448–468.

25.

Ogden

Carroll

Kit

et al . (2014) Prevalence of childhood and adult obesity in the United States, 2011–2012. Journal of the American Medical Association 311(8): 806–814. Available at: http://www.ncbi.nlm.nih.gov/pubmed/24570244

26.

Palinkas

Wingard

Barrett-Connor

(1996) Depressive symptoms in overweight and obese older adults: A test of the “jolly fat” hypothesis. Journal of Psychosomatic Research 40(1): 59–66.

27.

Pan

Sun

Czernichow

et al . (2012) Bidirectional association between depression and obesity in middle-aged and older women. International Journal of Obesity 36(4): 595–602. Available at: http://www.nature.com.prxy4.ursus.maine.edu/ijo/journal/v36/n4/full/ijo2011111a.html (accessed 4 May 2016).

28.

Puhl

Andreyeva

Brownell

(2008) Perceptions of weight discrimination: Prevalence and comparison to race and gender discrimination in America. International Journal of Obesity 32(6): 992–1000. Available at: https://dx-doi-org.web.bisu.edu.cn/10.1038/ijo.2008.22

29.

Robbins

Elias

Croog

et al . (1994) Unmedicated blood pressure levels and quality of life in elderly hypertensive women. Psychosomatic Medicine 56(3): 251–259.

30.

Roberts

Deleger

Strawbridge

et al . (2003) Prospective association between obesity and depression: Evidence from the Alameda County Study. International Journal of Obesity and Related Metabolic Disorders: Journal of the International Association for the Study of Obesity 27(4): 514–521.

31.

Roberts

Kaplan

Shema

et al . (2000) Are the obese at greater risk for depression? American Journal of Epidemiology 152(2): 163–170.

32.

Roberts

Strawbridge

Deleger

et al . (2002) Are the fat more jolly? Annals of Behavioral Medicine 24(3): 169–180. Available at: https://dx-doi-org.web.bisu.edu.cn/10.1207/S15324796ABM2403_02

33.

Roehling

Pichler

(2007) The relationship between body weight and perceived weight-related employment discrimination: The role of sex and race. Journal of Vocational Behavior 71(2): 300–318. Available at: https://www-sciencedirect-com.web.bisu.edu.cn/science/article/pii/S0001879107000437

34.

St-Arnaud-McKenzie

Payette

Gray-Donald

(2010) Low physical function predicts either 2-year weight loss or weight gain in healthy community-dwelling older adults. The NuAge longitudinal study. Journals of Gerontology Series A: Biological Sciences and Medical Sciences 65 (12): 1362–1368.

35.

Thurber

Snow

Honts

(2002) The Zung Self-Rating Depression Scale: Convergent validity and diagnostic discrimination. Assessment 9(4): 401–405.

36.

Vogelzangs

Kritchevsky

Beekman

et al . (2010) Obesity and onset of significant depressive symptoms: Results from a prospective community-based cohort study of older men and women. The Journal of Clinical Psychiatry 71(4): 391–399. Available at: http://www.psychiatrist.com/JCP/article/Pages/2010/v71n04/v71n0403.aspx (accessed 4 May 2016).

37.

World Health Organization (2010) Obesity. Available at: http://www.who.int/topics/obesity/en/

38.

Xiang

(2015) Obesity and onset of depression among U.S. middle-aged and older adults. Journal of Psychosomatic Research 78(3): 242–248. Available at: https://www-sciencedirect-com.web.bisu.edu.cn/science/article/pii/S0022399914004449 (accessed 9 March 2016).

39.

Zung

(1965) A self-rating depression scale. Archives of General Psychiatry 12(1): 63–70. Available at: https://dx-doi-org.web.bisu.edu.cn/10.1001/archpsyc.1965.01720310065008