Source of data and selection criteria

PMSI-MCO is the main source of data regarding hospitalization in France (MCO for medical, surgical, gynecological-obstetric), and the RIM-P (P for Psychiatry) is the main source for French psychiatric hospitalization data. These databases collect all of the public and private hospital stays for each patient (inpatients and outpatients) and allow patients to be followed over time thanks to a unique and anonymous linkage number. In the present study, we assessed depressed or obese people registered in the public medico-administrative data of the PMSI-MCO and RIM-P in the year 2016.

Patients were identified with specific ICD-10 codes for obesity or depression. The zip code of residence is recorded in the PMSI and the RIM-P for each patient, making it possible to map the hospital prevalence of obesity and depression.

French zip codes designate administrative areas ranging from hundreds (villages) to hundreds of thousands of inhabitants (major cities). In this study, zip codes were used in relation to the geographic codes found in PMSI or RIM-P data in order to localize our selection of patients.

Duplication of in or outpatient stays was carefully avoided and only single individuals with obesity and/or depression were recorded in the database, irrespective of each hospital stay during the year 2016.

This study analyzed inpatients and outpatients, males and females aged 18 years and more hospitalized in 2016 for depression or/and obesity. The patients were identified as follows:

1. Patients hospitalized for obesity
   • from the PMSI-MCO database:
     • Patients with main or associated diagnosis of obesity (ICD-10 code E66).
2. Patients hospitalized for depression
   • from the PMSI-MCO database:
     • Patients hospitalized for depressive disorder as main diagnosis (ICD-10 codes F32, F33);
     • Patients hospitalized for a suicide attempt (ICD-10 codes X6*, X7*, X8* as associated diagnosis) and a depressive disorder (ICD-10 codes F32, F33 as associated diagnosis) during the same stay.
   • from the RIM-P database:
     • Patients with a main or associated diagnosis of depressive disorder (ICD-10 codes F32-F33).

We used the same algorithms as other French studies which were conducted using national PMSI and RIM-P data [23,26].

Spatial analysis

Prevalence of in-hospital obesity and depression were calculated from PMSI geographic codes, indicating each patient’s zip code of residence (5,627 PMSI geographic codes in France). It was standardized for age and sex according to the direct method using national census data from the French national census institute, the INSEE (Institut National de la Statistique et des Etudes Economiques) as a reference. We used INSEE data from the 2013 census for the PMSI geographic codes; 2013 data were the most recent for decennial age categories. Prevalence rates are expressed for 100,000 inhabitants.

In order to investigate the spatial overlap between obesity and depression, the mapping presents standardized prevalence rates for each condition with 2016 PMSI geographic codes. The prevalence rates were presented in two classes for each disease:

1. lower than the national average prevalence;
2. higher than the national average prevalence.

Four classes of geographic codes were thus defined, following a High-High/High-Low/Low-High/Low-Low discretization. For example, High-High means that prevalence rates of both obesity and depression are higher than the national average prevalence.

All spatial analyses were performed using GIS (Geographic Information System) MapInfo 11.0 and statistical analyses were performed with SAS 9.4.

Multivariable analysis

A logistic regression analysis was performed to assess the relationship between the standardized prevalence rates of obesity and depression (by tertile) and the influence of environmental and socio-economic determinants. The independent factors entered into the multivariate model were the deprivation index, the level of urbanization, and depression (when obesity was the outcome) and obesity (when depression was the outcome).

Spatial analysis

The study highlighted some significant differences between France regions (Fig 2), with some clusters of co-occurrent over-representation (Brittany, Massif Central, North and North East part of France), and some clusters of under-representation (Pays de La Loire, Mediterranean belt, East of Auvergne-Rhône-Alpes area). Depression and obesity rates were simultaneously higher than the national average in 15.7% of the French territory, and depression and obesity were simultaneously lower than the national average in 44.0% of the territory, meaning that in 60% of the French territory the rates for the two conditions were conjointly above or below the national average.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Hospital prevalence of obesity and depression.

https://doi.org/10.1371/journal.pone.0210507.g002

Multivariable analysis

We then analyzed the independent relationship between obesity and depression using the geographic codes.

The multivariable analysis revealed a highly significant mutual interaction between the geographic prevalence of obesity and depression (Table 2), with an increase in the risk of obesity in highly depressed regions (Odds Ratio (OR) of 1.295; 95% CI [1.128–1.486], p = 0.0002) (above average prevalence), but a protective effect in regions with moderate levels of depression (OR 0.704 [0.611–0.811], p<0.0001) (below average prevalence).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Multivariable analysis of geographical interactions between obesity and depression.

https://doi.org/10.1371/journal.pone.0210507.t002

An analysis of the risk of depression as a function of the geographical prevalence of obesity showed similar results (Table 3). Geographically, the risk of depression was 1.323 [1.155–1.517] (p<0.0001) in regions with high obesity, and, conversely, the risk of depression was 0.724 [0.627–0.835] (p<0.0001) in regions with lower rates of obesity.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Multivariable analysis of geographical interactions between depression and obesity.

https://doi.org/10.1371/journal.pone.0210507.t003

In addition, the analysis demonstrated that areas of major material or social deprivation were associated with a high obesity rate: OR 1.484 [1.217–1.809] (p<0.0001) and 2.243 [1.835–2.741] (p<0.0001) respectively, and 3.583 [2.837–4.526] (p<0.0001) for people living in the most deprived socio-residential context. Concerning MDD, areas with material and global deprivation seemed to be more at risk of depression (OR 1.233 [1.011–1. 502], p = 0.0383 and OR 1.648 [1.308–2.078], p<0.0001), while individuals living in advantaged areas were less likely to develop depression (OR 0.724 [0.627–0.835], p = 0.0024). However, a high rate of depression was not associated with contexts of social deprivation.

Depression was also influenced by residence in urban or rural areas. Living in smaller rural cities (small and mid-size centers) was associated with depression (OR 2.218 [1.811–2.716], p<0.0001), and people living in the suburbs of major cities were less likely to be depressed (OR 0.765 [0.662–0. 885], p = 0.0003). Rurality was also associated to high rates of depression (OR 1.281 [1.093–1.501], p = 0.0022). All types of areas were associated with high rates of obesity with taking as reference the major urban centers.

Social deprivation and urbanicity

Mainly deprived areas, and areas with material deprivation to a lesser extent, were associated with depression, while the most advantaged areas were negatively associated with depression. However, social deprivation was not associated with depression. As mentioned above, we can speculate that people in the most advantaged regions adopt healthy lifestyles whereas people in most deprived areas have a poor diet and lack of physical activity, in addition to other psycho-social factors. A number of studies on depression have pointed out the complexity of socio-environmental factors, and it is worth noting that material deprivation has not been directly associated; several reports on well-being have underlined that the correlation between income and happiness is far from straightforward [37,38]. In our study, the most deprived geographic codes (i.e. both social and material deprivation) were strongly associated with increased depression, enhancing the need for progress in public policies in terms of socio-economic poverty reduction and development of health education programs. Regarding obesity, all deprived areas were linked to obesity regardless of material or social deprivation, which parallels existing literature [39,40]. Indeed, the relationship between obesity and poor socio-economic status or modest living conditions is well established. In the present study, we may assume that cold spots for obesity and depression (i.e. Pays de La Loire, East of Auvergne-Rhône-Alpes region, West and South-East France) are regions known for a quality of life, high socio-economic level and comparably low unemployment while hot spots for both comorbid obesity and depression are regions known for their socio-economic difficulties (North and East part of France), where global deprivation is significant. Previous reports have underlined the influence of socio-economic status on obesity [41,42], and the relationship is clearly present in France. The impact of regional socio-economic specificities likely limited the obesity/MDD mapping overlap. Indeed, if obesity seems highly related to low economic status as discussed previously, the prevalence of MDD depends on much more personal and environmental factors not directly related to socio-economic status, conveying different regional patterns for MDD and obesity. However, most deprived areas were associated with both obesity and depression, while individuals in most advantaged areas were less likely to develop both conditions. These crucial data require adjusted public policies for the most deprived zones in order to reduce geographical health disparities. On the one hand, very few French health campaigns have attempted to enhance public awareness of depression or to encourage screening and psychiatric care [43]. When developed, those rare programs focused mainly on mental health issues without addressing metabolic disturbances or other somatic outcomes possibly associated with depression [44]. In addition, there have been no specific programs for deprived regions or community groups. On the other hand, some public health media campaigns have focused on nutrition and healthy food, with a focus on the physical health consequences of a poor diet rather than a psychosocial approach. Yet recent reports [4] have stressed the need to develop health promotion in the most deprived French communities where obesity has increased dramatically over the last 10 years. Integrative biopsychosocial health programs appear appropriate for this context considering that motivational strategies are multifaceted and depend on mental determinants. Food intake and emotion share some fundamental pathways in terms of neuro-cognitive process, especially in people with high BMI. This has been demonstrated by several studies where healthy food behaviors were heightened by a focus on pleasure and tastiness rather than on a strict health rationale [45]. Thus, an integrative approach promoting health and positive emotions in communication and education may lead to healthier food habits in people with obesity, particularly in the most deprived areas. In this aim, our geographical and epidemiological analysis could inform the development of such integrative health programs in the most deprived areas, where both obesity and depression are highly prevalent.

Regarding urbanicity, though early reports underlined an association between population density and depression [46–48], the latest studies have dismissed this link, enhancing instead a specific risk in semi-rural regions and mid-sized urban centers in rural territories as illustrated here. These results appear to be in line with Reeves et al. and Breslau et al. who showed the deleterious aspects of rurality and small urban centers vs. large metropolitan centers [49,50], which might be due to the lack of psychiatric resources in small towns and rural areas. An older study conducted in England highlighted the link between suicides and the remoteness from large urban centers, concluding that individual isolation in certain rural areas might lead to major depression and suicide [51]. Concerning obesity, our study highlighted that the suburbs of major cities had an increased risk of high prevalence rates. Similarly, American studies have pointed out the negative effect of urban sprawl on obesity rates. People living in large suburbs may develop worse life habits, such as a sedentary lifestyle or more motorized transports [41,52,53]. This has been referred to as an “obesogenic environment” [54] even if the interactions between urban environment and obesity remain complex [55,56]. Recent data revealed that inhabitants of rural areas, where specific socio-residential risks like precarity may be associated to a lower access to health facilities, would be more at risk of developing overweight [57], which is in line with our study. In the United States, some data showed a similar geographical association between obesity and diabetes in geographic mapping, in spite of cultural specificities and differences [17,58]. So, the additional socio-environmental data concur with existing literature, and emphasize our main result, the partial but robust geographical overlap between obesity and depression.

Age and gender differences

The demographic characteristics of our study population are similar to those generally reported in epidemiological surveys for both mood disorders and obesity. The depression peak observed in the present report corresponds to the over-representation of depression between 40 to 60 years, mainly in women, often described in mental health epidemiology [10]. As far as obesity is concerned, many reports have shown that the prevalence of obesity increases with age. In France, recent studies have demonstrated that obesity was twice as prevalent in the 55–74 age group than in the 18–39 age group [4]. Thus, our data are in line with existing epidemiological data.

The present study highlights some gender differences in terms of health care access. If depression rate in males and females is consistent with the standard epidemiology of MDD (females 66% vs. males 33%) [59–61], it is not the case for hospital prevalence of obesity. Indeed, our study demonstrates that women suffering from obesity are hospitalized much more frequently than men until the age of 60. This observation is similar to a previous study on health perceptions and underassessment of body weight in males [62]. Thus, while epidemiological reports describe a balanced sex ratio in obesity prevalence and more overweight men than women, it is worth mentioning that men and women have different beliefs and behaviors relating to weight issues. Overweight males are less likely to see themselves as overweight than females in the same situation. Furthermore, other reports demonstrated than men are underrepresented in weight-loss programs [63]. In addition to a health perception bias, men see most weight loss programs as unattractive or designed for women’s needs [64]. Similarly, there has been a marked gender difference in patients undergoing bariatric surgery (males 20% vs. females 80%) for several decades [65]. All these data illustrate the major psycho-behavioral obstacles that discourage men from taking charge of their weight problems. The very low rate of hospitalized males with obesity in the present study confirms a detrimental gender bias. An understanding of this bias should encourage the development of specific programs for men, from primary care to specialized nutritional services.

Limitations

This study has several limitations. First, as we used aggregated data at the scale of French zip codes to evaluate the overlap between two pathologies and the potential association with several socio-residential indicators, our results may be subject to an ecological bias [66]. However, the object of this study was not to investigate individual propensity to develop both conditions but to explore spatial tendencies for a whole country. Several spatial studies using the same type of data and scale of analysis have demonstrated the robustness of this type of analysis [21,67,68]. Second, we could not explore obesity and depression in primary care, but our study probably identified the most severe cases of depression and obesity that required in-hospital health services. In addition, the obesity and depression rates demonstrated in this research are logically minor in comparison with French annual prevalence of both diseases [4,69] because of the in-hospital approach. However, we may think that in-hospital spatial distribution of prevalence rates might reflect the global spatial prevalence of the two diseases, irrespective of severity. Third, in this study we included overweight patients (25<BMI<30) but we are not sure that we could identify them exhaustively. This limitation might have led to an under-evaluation of metabolic dysregulations due to the ICD-10 code threshold. We may speculate that more comorbid depressed-overweight people would have been revealed by including all non-obese overweight patients, which would have probably strengthened the main statistical trends of the present study, i.e. the robust statistical association between obesity and depression. Fourth, our analysis was standardized for age and sex but did not take into account lifestyle or antidepressant medication, though those confounding factors were not strongly associated with obesity in depressed people in several reports. Indeed, epidemiological evidence supports the association of depression and obesity in both Western and non-Western countries, with a robust bidirectional relationship, independently of antidepressant medications [5,70]. In fact, weight gain related to psychotropic medications appears more to be linked to polygenic risk more than comorbid depression/obesity mechanisms [71]. Despite these limitations, other reports have demonstrated similar spatial correlations in linked pathologies (e.g. obesity, diabetes and stroke) [16,18], and we believe our study provides useful additional epidemiological data including group-level associations between obesity and depression which underline shared clinical and biological mechanisms.