In this study, data from the Survey of Health, Ageing, and Retirement in Europe was used to investigate our research question aiming at analyzing a possible bi-directional association between the BMI of male and female individuals from wave 4 and mental health, including QoL and depression status, in wave 8 and vice versa, in the context of a longitudinal analysis using a two-wave cross-lagged panel approach over a defined time lag of 10 years.

The Survey of Health, Ageing and Retirement in Europe was conducted among 50+ individuals and started in 2004, including 11 countries within Europe. Baseline study in wave 1 involved Scandinavia (Denmark, Sweden), central Europe (Austria, France, Germany, Switzerland, Belgium, Netherlands) and the Mediterranean including Greece, Italy and Spain. At the end of 2004, Israel was additionally integrated into the survey. Estonia, Portugal, Slovenia and Hungary joined the third regular panel in wave 4 carried out in 2010, including 58,000 participants constituting the gross data basis for this study. Fieldwork for the eighth wave in October 2019 was interrupted due to the COVID-19 pandemic in March 2020, resulting in a considerably shortened questionnaire. Therefore, the following countries were included: Austria, Belgium, Bulgaria, Cyprus, Croatia, Czech Republic, Denmark, Estonia, France, Germany, Greece, Israel, Italy, Luxembourg, Poland, Spain, Sweden, Lithuania, Latvia, Malta, Finland, Switzerland, Slovenia, Hungary, Netherlands and Slovakia (54) resulting in a gross sample size of 46,733 of which 19,915 participants completed main interviews in the fourth wave.

The target population of the SHARE survey consisted of people who were 50 years of age or older at the time of the surveys and who had their regular residence in the relevant country of the survey. Persons who were either hospitalized, incarcerated, or out of the country for the entire duration of the interview were excluded from the interviews. Other exclusion criteria included insufficient language skills with regard to the respective national language or moving to an unknown address. All persons of a household born in 1960 or earlier (wave 4) and 1969 or earlier (wave 8) were admitted to interviews (55). The participation in the interviews was voluntary and confidential. All collected data from the survey interviews was linked to a wave-specific identifying variable (ID) (54).

All data was collected using computer-assisted personalized interviews (CAPI) and face-to-face interviews. This process was further enhanced by self-completed paper & pencil questionnaires (54). Each interviewer was using a laptop when conducting face-to-face interviews on which the CAPI tool was installed. Due to the need of executing various physical tests, this personalized approach was necessary for the survey. SHARE uses an ex-ante harmonization concept, that is, one general questionnaire is translated into the respective national languages. This is made possible through the use of an internet-based translation tool, which carries out an automatic translation into the CAPI instrument (55). Based on a significant dropout of several respondents over the time period of 10 years who either died, moved from their regular residence, could no longer be contacted, or refused to participate, we finally obtained a net sample of 16,184 from 19,915 individuals who participated in both wave 4 and wave 8 providing complete data without defined missing values for further analysis. Another reason for the withdrawal of participants in the eighth wave from the survey might have been the beginning of the COVID-19 pandemic in 2019/2020, which is why the study was discontinued.

The study was conducted according to the guidelines of the Declaration of Helsinki and obtained approval from the Ethics Committee of the Medical University of Graz, Austria (Document Number 32-305 ex 19/20).

Internal consistency reliability of mental health measures was calculated using Cronbach’s alpha. The EURO-D scale which was applied in wave 4 and wave 8 resulted in values ranging from 0.62 to 0.78 and proved to be rather moderately internally consistent in all included European countries (62). Moreover, the abridged CASP-12 scale indicates sufficient Cronbach’s alpha scores between 0.56 and 0.76 across different countries (63).

The first step was to match relevant SHARE data sets including central characteristics from waves 4 and 8 as well as sociodemographic variables. The matching procedure was accomplished using a person identifier variable as a key variable. Given that some cases yielded data on wave 4 but not on wave 8, all following calculations, including respective hypotheses tests, refer entirely to the cases in which all data was available. Remaining, non-paired cases were excluded using a filter variable. The age of respondents was transformed into categories representing three age groups: “50–59”, “60–74”, “75+”. Collected information on the “living situation” of respondents was transformed into a binary coded variable for “living alone,” and “living with a partner.” Furthermore, we created dummy variables of the categorical predictors, including sex, BMI, and educational level to allow the analysis of their impact on the respective outcome variables. Responses such as, “refuse,” “do not know,” “implausible or suspected wrong,” were defined as missing values and were excluded from the analysis.

Statistical analysis was entirely performed using IBM SPSS statistical software (Version 26). Addressing the long-term effects, according to our hypothesis, was accomplished using multiple linear regression analysis to define associations between BMI categories and mental health variables, depression and QoL. The level of significance (value of p) was defined at p < 0.05 (α = 5%). In order to evaluate the association of predicting variables from wave 4 on outcome variables in wave 8, a two-wave-cross-lagged panel approach (64–66) with a time lag of about 10 years was conducted in the context of a longitudinal analysis. Central variables, BMI, depression status and QoL were measured at two different time points (wave 4 and 8) leading to six possible relations (65), which were analyzed using multiple regression analysis to determine stability (autoregressive) and cross-lagged effects, whereby synchronicity of constructs at both time points was assumed. Synchronicity, stability and cross-lagged associations of central variables in waves 4 and 8 result in two synchronous (cross-sectional), two stability (autoregressive), and two cross-lagged relations (65). In order to investigate the stability of the relation between BMI values at two time points, BMI was used as a metric predictor. Accordingly, autoregressive and cross-lagged effects between different measures in waves 4 and 8 were calculated simultaneously to determine the size of stability of each construct and reciprocal associations over time (Figure 1). Accordingly, the analysis of stability and cross-lagged effects resulted in 3 models (Table 1).

Based on these results, the impact of sociodemographic variables, including age, sex, living situation and educational level (ISCED-97) on respective outcome variables was analyzed. Additionally, further multiple linear regression analysis adjusting for sociodemographic characteristics was conducted.

In the longitudinal analysis, 16,184 people (aged 50 and above) were included, consisting of 9,428 (58.3%) female and 6,756 (41.7%) male respondents. The mean age of the participants was 63.6 years (SD = 8.1). The youngest participants were 50 and the oldest was 92 years of age. Most of the participants were from Estonia (13.1%) and the Czech Republic (10.1%). The majority of respondents reported living together with their partner (74.9%). In addition, 37.6% of the sample population indicate an upper secondary education (level 3) (Table 2).

The mean BMI score of the study population was 27.0 kg/m² (SD = 4.7). The majority of respondents were overweight (42.1%) followed by normal weight (35.2%) and people living with obesity (21.8%) – underweight people only account for a minority of 0.9%. The mean depression score according to the EURO-D scale was 2.2 (SD = 2.1) reflecting an overall non-depressed state of the population. After measuring QoL using the CASP-12 scale ranging from 0 to 12, mean values and standard deviations were calculated, indicating a mean QoL score of 38.4 (SD = 5.9) indicating a high level of QoL within the study population (Tables 3, 4).

All three models demonstrate good to high fit to the data. The first model (model 1) was significant in terms of the stability of QoL (CASP) scores and BMI categories predicting QoL over time in the eighth wave (F (4,16,179) = 1575.441, p < 0.001, R² = 28.0%). As a result, QoL scores in wave 4 seem to significantly predict QoL scores (autoregressive path) at a second time point in wave 8 (β = 0.515, p < 0.001). This mental health characteristic indicates high stability over a lag time of about 10 years. Cross-lagged analysis of BMI categories predicting QoL indicates that people living either with overweight, obesity or underweight show significantly decreased levels of QoL in comparison to respondents of normal weight, where people living with obesity show the strongest effect (β = −0.084, p < 0.001).

Our second model (model 2) yields a significant result as well (F (4,16,179) = 820.656, p < 0.001, R² = 16.8%). The Depression score proved significantly stable in the long term (β = 0.403, p < 0.001). The analysis of cross-lagged associations yields that BMI categories, underweight, overweight and obesity significantly predict depression status over time, resulting in increased scores of depression compared to people of normal weight. Again, with the strongest effect among individuals who live with obesity (β = 0.063, p < 0.001).

The third model (model 3) was also significant (F (3,16,180) = 820.656, p < 0.001, R² = 61.2%) demonstrating a significantly stable association regarding the autoregressive path. The BMI score in wave 4 predicts the BMI score in wave 8, indicating high stability (β = 0.782, p < 0.001), whereas depression state and QoL do not significantly predict BMI over time (Table 5).

In order to investigate the impact of sociodemographic covariates on respective outcome variables, we performed a second run of analysis adjusting for baseline characteristics, including sex, age, living situation, and educational level. In addition, cross-lagged associations between central constructs and stability correlations were retested according to models 1 to 3 described above, including confounding predictors.

The first model examining the effect of BMI categories and confounding factors on QoL was significant (F (12,16,171) = 647.113, p < 0.001, R² = 32.4%). The analysis reveals that respondents of older age indicate a significantly lower QoL, which comparably demonstrates the strongest effect among sociodemographic characteristics within this model in terms of predicting QoL (β = −0.166, p < 0.001). Moreover, our first adjusted model provides no evidence that sex of respondents significantly affects QoL over time. The education of respondents seems to significantly affect QoL. In fact, the higher the level of education the higher the QoL in the long term, while educational level 5, referring to tertiary education, represents the strongest educational predictor of QoL (β = 0.118, p < 0.001). Individuals who live with their partners show significantly higher QoL than those who live alone (β = 0.032, p < 0.001).

The second model examined the impact of confounding factors on depression scores, yielding a significant result (F (12,16,171) = 345.542, p < 0.001, R² = 20.3%). Similar to model 1, age of respondents seems to play a significant role in predicting depression scores over time. People of older age show higher depression scores over a time-lag of about 10 years (β = 0.138, <0.001). Furthermore, females significantly indicate increased depression scores compared to males (β = 0.091, <0.001). The level of education seems to affect depression scores. More precisely, higher educated individuals indicate significantly lower depression levels over a 10 year period in contrast to individuals with low education. Again, tertiary educated individuals, who belong to educational level 5, show the strongest effect in terms of decreased depression scores (β = −0.094, p < 0.001), while living with a partner has no significant effect on depression status compared to living without a partner.

The third model proved to be significant as well (F (11,16,172) = 2421.248, p < 0.001, R² = 62.2%). Interestingly, older respondents seem to have significantly lower BMI scores (β = −0.100, p < 0.001), whereas sex has no significant effect on BMI scores. With regards to education, in contrast to QoL and depression status, education levels 5 and 6 (tertiary education) make a significant contribution to the prediction of BMI (β = −0.028, p < 0.001; β = −0.013, p = 0.008). Thus, tertiary education seems to be negatively associated with BMI, indicating that tertiary educated individuals show decreased BMI scores. Overall, living situations, that is “living with a partner,” does not significantly predict BMI (Table 6).

In summary, the stability of central constructs proved significant over a 10 year time period in all three models. Significant cross-lagged effects can be observed in one direction between BMI predicting QoL and depression state, but not vice versa. Hence, people living with underweight, overweight or obesity show lower QoL and higher depression scores than people of normal weight, whereas obesity indicates the strongest effect regarding the prediction of both QoL and depression state over a lag-time of 10 years. Accordingly, results do not confirm any reciprocal association. The depression state and QoL do not significantly predict BMI in older adults. Moreover, attention has to be paid to the rather small strength of the respective cross-lagged effects.

Sociodemographic predictors such as educational level and age of respondents seem to significantly affect QoL, depression state, and BMI over time. Higher education seems to be associated with lower BMI and depression scores. More highly educated people show greater QoL levels than those with lower levels of education. Female gender obviously predicts increased depression scores, whereas no significant impact on QoL and BMI can be observed. The older age of respondents also proved to be a significant predictor for increased depression scores and decreased QoL, but is negatively associated with BMI scores, implying that older individuals display lower BMI scores. However, the effects are rather small.

In contrast to the performance of randomized control trials (RCTs) for investigating causal effects among study populations, longitudinal studies of observational data such as the Survey of Health, Ageing and Retirement in Europe (SHARE) offers the opportunity to study long-term, and hypothesized bi-directional associations over a defined time-lag. Although results can never prove complete causality. Nevertheless, this study provides a longitudinal design that is highly desirable in order to investigate cause-effect associations between mental health and overweight over time. Despite a large sample size, including respondents from all over Europe, this study does not account for further possible confounding factors such as physical activity, eating behavior, physical comorbidities or the use of psychotropic medication, which may substantially contribute to causing obesity and metabolic side effects (40).

In fact, information on the intake of psychotropic drugs of respondents may have an impact on results considering that prior research (34–39) suggests that certain antidepressants are suspected to potentially increase the risk of weight gain, in addition to antipsychotic drugs. Following the suggestions from an early review conducted by Fava, older antidepressants such as mirtazapine, within the subgroup of tricyclic antidepressants (TCAs), and monoamine oxidase inhibitors (MAOIs), may more frequently increase body weight compared to the newer generation of selective serotonin reuptake inhibitors (SSRIs) (35), which is also in accordance with suggestions by Serretti and Mandelli who claim in their meta-analysis that mirtazapine, among other antidepressants, is considered a cause of weight gain (36, 37, 39). Gaining weight from these medications may in some cases result from complex hormonal mechanisms, including the antihistaminergic effect of both antipsychotics and certain antidepressants on H₁ histamine receptors, and imbalances of the orexigenic and anorexigenic hormones ghrelin and leptin (77–79). Yet, it should be considered that the mechanisms through which antidepressants may potentially be associated with weight gain are unclear and poorly understood (80) assuming that these findings may not warrant causal inference, while the effect of residual covariates possibly contribute to overrate this association (81).

According to Rogosa, investigating long-term effects using a two-wave cross-lagged procedure is not useful for determining causal inference (82), whereas Selig and Little suggest that this panel design may generate more insight, better understanding of longitudinal associations, and stimulate further research (83). Another limitation of this study is that the anthropometric characteristic body weight was self-reported, which carries the risk of response bias. Moreover, we were not able to account for weight bias and stigmatization of people living with obesity, which obviously may affect people’s mental health status. Individuals who live with obesity often experience weight bias and being targeted for jokes because of their body weight, especially via the media, which further promotes the stereotype that larger body size does not comply with the norm, and is socially not acceptable (84).

In fact, one strength of this study is that the analysis is based on observational data derived from a large-scale panel approach which may significantly contribute to generating an adequate sample population encompassing the older adults in their private household setting. The analysis of longitudinal data provides a reliable tool for identifying cause-effect- and bi-directional associations in the long term. Since data collection was performed on people of the general population from 15 European countries, the sample population is assumed to be representative and provides an appropriate sample size yielding generalizable results.

Given that both mental illness and overweight are public health issues of great importance, gaining more insight into the complexity of their relationship and exploring the underlying nature of these two entities seem to be crucial in respect of disease prevention and developing innovative treatment strategies considering that mental health issues have often been overlooked in the past (85). The findings of this study should initiate further research focusing on this subject, accounting for links between mental health issues and their impact on body weight, and vice versa, particularly, in the older adults among whom overweight and obesity is highly prevalent. Thus, integrating mental health status into treatment strategies following a biopsychosocial approach and implementing new evidence in clinical settings could improve future treatment and prevention strategies.

In this regard, this study should encourage researchers to set up innovative investigations addressing associations between mental health and obesity, and further analyze interactions between common mental illnesses such as depression and anxiety, with respect to overweight and obesity, particularly with regard to bi-directional long-term effects and corresponding casualties in terms of a cause-effect principle. In fact, evidence on associations between mental health and obesity is still inconclusive (45) and most of the studies that have been carried out in the past are based on a cross-sectional approach, while prospective studies using repeated measurements of both mental illness and overweight are scarce. In particular, further large-scale investigations are required in order to provide profound studies by which potential cumulative effects can be analyzed and discussed in a broader sense. Moreover, generalizability of previous findings needs to be verified (43). Nowadays, both mental illness and obesity are among the greatest global public health challenges. Exploring and understanding the underlying nature of these complex relationships is crucial and highly desirable, considering that the associations between obesity and mental health highlight the importance of addressing both physical and mental health when treating obesity. By addressing mental health concerns, individuals with obesity may be better equipped to manage their weight and improve their overall health and well-being.

This work is intended to create stronger awareness and to stimulate further discussion on this issue. In particular, this research provides multiple measuring points within a panel approach, which is highly necessary to assess reciprocal interactions. Most of the prior research focusing on the relationship between obesity and mental health predominantly determines synchronous effects of a given population leading to results that are generally rather inconsistent (43). Our investigation provides a longitudinal design that allows us to estimate reciprocal interactions, which is quite rare in view of the existing evidence. Since it is well known that both obesity and mental health issues do often occur among older individuals, indicating a highly prevalent issue, our findings provide valuable statements about the associations over a long period of time, and hint to possible causal influences that need to be further analyzed by future research activities. Moreover, most of the studies carried out previously yielded results that do not allow an overall conclusion to be drawn for all age groups, which may have an impact on respective results (48). This study contributes to obtaining a more detailed picture of long-term effects and interactions among older people, which is essential in terms of determining the direction of the relationship. On this foundation, treatment strategies and prevention programs could be further improved by identifying potential risk factors, considering that obesity may be a risk factor for the development of depression, interventions that target overweight and obesity may improve the prevention of depression and low QoL. Furthermore, the results of our investigation should serve as an important foundation for clinical practice suggesting that clinicians and professional care providers should account for mood monitoring in patients living with overweight and obesity, especially when treating older patients with long-term obesity, which may further facilitate the establishment of treatment guidelines.

In addition, we hope that the findings of this study strengthens public awareness about the importance of addressing both obesity and mental health in order to reduce stigmatization and encourage affected individuals to seek treatment. This work should further contribute to identifying early signs of pathological developments and improving interventions for people at a higher risk. We really hope that this research stimulates further scientific discussions and investigation to verify the nature of the relationship between mental health and obesity. Based on our findings, we do believe that further epidemiological research is warranted to determine these underlying mechanisms, notably with regard to possible moderators and confounding factors within the European population. Finally, we hope that our findings inform policy decisions related to public health.