Evidence of depression affecting multimorbidity in survey data from the Brazilian population

Introduction

Depression is one of the leading health concerns worldwide today^1,2. Over 350 million people are affected by this disorder and prevalence in the general population is estimated between 3 to 10%¹. It is the single largest cause of years lost to disability², as it is a chronic disorder with symptoms that last for long periods and are recurring. It is also related to worse quality of life, medical morbidity and higher mortality³.

There is evidence that depression is associated with multimorbidity, the occurrence of two or more chronic diseases. It has been shown that the direction of this association seems to run from multimorbidity to depression^4,5. As individuals age, they tend to become alone, losing friends and family members, which renders them susceptible to depressive symptoms⁶. Furthermore, with age they are subject to brain vascular injury due to hypertension, diabetes mellitus, obesity, and high blood cholesterol⁷. Examining the relation between multimorbidity and depression is important as a global health preventive measure, as it can help match resources to the individuals who are hit the hardest in crises like the recent Covid19 pandemic. For example, during the COVID19 pandemic, multimorbidity was associated with increased anxiety and depressive symptoms both during the social isolation phase ⁸ and post-infection⁹. A systematic review published by Read et al. (2017) reported that people with multimorbidity have twice the risk of depressive disorders⁷. Socioeconomic factors seem to play an important role in the multimorbidity of mental disorders and chronic diseases, as this combination of disorders are more common in more deprived areas, where they particularly affect younger individuals^10,11.

Multimorbidity is known to increase with age, reaching higher prevalence in very old individuals. There is an ongoing effort of trying to identify patterns of clustering for these conditions, which varies significantly across studies¹². A three-cluster pattern seems to be emerging, with musculoskeletal, cardiometabolic, and neuropsychiatric clusters^4,12. Furthermore, being overweight is associated with multimorbidity in a dose-response pattern. A recent Finnish study reported a hazard ratio of 2.83 for obesity and 2.67 for those overweight¹³ in developing multimorbidity. It is therefore essential to take ageing and body mass index (BMI) into consideration in every multimorbidity study.

By contrast, the effect of depression on multimorbidity is seldom investigated. Subjects with depression are vulnerable to multimorbidity due to poor lifestyle choices (such as tobacco smoking or taking less exercise), limited adherence to clinical treatment and poor access to health¹⁴. There is also evidence that depression and multimorbidity have common biological and psychosocial drivers (inflamation, stress, economic burden) that might interact and result in one or both conditions¹⁸. Additionally, some medications for depression are associated with metabolic syndromes, obesity, and cardiovascular events^15–17. So, though it is clear that multimorbidity has an important effect on depression, there is a potential causal path in the opposite direction. This paper aims to investigate this, the effect of depression on multimorbidity, using a publicly available Brazilian survey including 28,382 respondents from 2014, and incorporating age and BMI as controls, in a structural equation model that accounts for error measurement.

Methods

Data analysis

All analyses were performed using R version 4.1.3 (https://cran.r-project.org/). All analyses used an alpha of 0.05. The overall missingness of the entire dataset was below 10% (Supplemental material). Missing data were filled by pooling from 5 multiply-imputed data frames using the R mice package with the classification and regression trees algorithm²⁰. The analytic strategy used structural equation modelling, allowing both investigation of latent constructs and accounting for measurement error. The structural equation model specification can be seen in Figure 1. The estimation of the coefficients was performed using the lavaan survey package version 1.1.3.1 and incorporated the rake weights. This package does not allow categorical variables. This is the reason all variables were included in the model as continuous. For complex survey data, i.e. with weights, there currently is no alternative to this limitation. Identification was automatically tested by loading the model specification into the umx package²¹ and using the mxCheckIdentification tool²². The model is locally identified. The path diagram of figure 1 was built using Onyx²³. The estimation was maximum likelihood based, using the “MLM” option to obtain robust fit statistics.

• Download figure
• Open in new tab

Figure 1

SEM path specification. Dashed lines are negative coefficients. Standard SEM representation is used, circles are latent variables, boxes are measured variables. Standardized estimates. Model Fit: χ²(23) = 671.442, p < 0.001; Robust CFI = 0.971;Robust TLI = 0.955; Robust RMSEA = 0.051 (90%CI0.048,0.055); Robust SRMR = 0.026. *, statistically significant. Age_dep, age in which the respondent was diagnosed with depression; Psy_med_use, whether the respondent used psychiatric medications; Severity, variable coding whether the respondent had depression and if yew, the burden it produced in their lives (no, mild, severe, very severe); n_med_chr, number of medications used to treat chronic conditions; n_chr_condition, number of chronic conditions; multimorbidity, whether the respondent endorsed multimorbidity; time_medicated, for how long the respondent has been using medications

Results

After pre-processing, the set contained 28,382 respondents. Demographic and clinical statistics are shown in Table 1. Continuous variables were analysed with the average of the descriptive measures, median, minimum, maximum and standard deviation (SD) and categorical variables were described with the total number of observations and their corresponding percentages stratified by sex. The data set consisted of 8,780 (30.9%) males and 19,602 (69.1%) females. The individuals had a mean age of 61.04 (SD 14.04) years, and most patients were aged between 60 and 99 years (Males: 60.4% and Females: 52.7%). Both males and females had on average 2 chronic diseases (Males: 2.05 (SD 1.30) and Females: 2.30 (SD 1.38)). This was also observed in the number of medications for chronic disease treatment (Males: 2.10 (SD 1.28) and Females: 2.30 (SD 1.32)). Most individuals resided in the South (Males: 25.3% and Females: 25.0%) and Southeast regions (Males: 23.8% and Females: 23.6%), while individuals from the North of the country were the minority in this data set (Men: 14.9% and Women : 12.9%). Of those responding, 64.8% were considered multimorbid, 59.6% of males and 67.2% of females. Only 0.8% of respondents used psychiatric medication, 0.4% of males and 1.1% of females. Females engaged in intense physical exercise on average 7.06 (SD 22.25) days per week, while males exercised on average 4.62 (SD 10.26) days per week. Males smoked on average 17.30 (SD 18.44) cigarettes per day and females 15.11 (SD 17.36) cigarettes per day. On average, individuals with depression were diagnosed at 48.48 (SD 21.02) years of age; males at age 52.12 (SD 19.74), and females at age 47.76 (SD 21.19). In response to the depression severity variable, the majority of individuals did not endorse any life limitations from depression (Males: 4.3% and Females: 8.4%) others acknowledged few limitations (Males: 3.0% and Females: 7.2%), moderate limitations (Males: 1.7% and Females: 4.1%), intense limitations (Males: 1.1% and Females: 2.9%) and very intense limitations (Males: 0.7% and Females: 1.6%). Note that this variable also encoded subjects who did not have a diagnosed depression, which was the majority of both sexes (Table 1).

The path of interest for the SEM was the regression between the latent variable representing depression to the latent variable representing multimorbidity. The model fit was good: χ²(23) = 671.442, p < 0.001; Robust CFI = 0.971; Robust TLI = 0.955; Robust RMSEA = 0.051 (90%CI 0.048,0.055); Robust SRMR = 0.026. All observed variables were modelled with error measurement variances (Figure 1). The standardized coefficient found for this path was 0.355 and was statistically significant (p<0.05), which means that for each SD increase in the depression syndrome latent variable, an increase of 0.35 SD in multimorbidity is expected. This effect is medium size. Age contributed a positivemoderate effect on multimorbidity, and BMI contributed a small positive effect on multimorbidity, though neither were statistically significant). The upper part of the model (BMI and age) was specified so that these variables could act as controls in the relationship between depression and multimorbidity.

Age and BMI both had significant small negative effects on the depressive syndrome latent variable. The former was surprising, given that depression is positively associated with ageing²⁴.

Discussion

This is a cross-sectional, maximum likelihood based analysis of the association between depression and multimorbidity using complex survey data from Brazilian individuals. It found a significant association in the direction of depression affecting multimorbidity, a direction seldom studied in the multimorbidity literature. The size of the effect found (0.355) was medium, and the model presented a good fit in several fit indices.

Most of the present literature in multimorbidity has focused on the effect of multimorbidity in depression. We have shown that the effect of depression on multimorbidity is of importance. In our analysis, age had a moderate effect on multimorbidity (0.332) and BMI had a small effect on (0.06, not significant) multimorbidity. The former confirms the pattern found in the literature ¹², the latter was less than what was expected based on what had been reported of the effect of BMI on multimorbidity ¹³. In this analysis, BMI and age had little effect on depressive syndrome, which was surprising give that data comprises older subjects and the known relationship between age and depression ²⁵.

Obesity (measured by BMI or waist circumference) is a highly prevalent health condition associated with early death and high health costs²⁶. The rising incidence of obesity in the low-middle income countries is associated with multimorbidity. Romano et al. (2021) reported based on a sample of elderly (over 60 years) individuals that obesity (BMI≥ 30 kg/m2) was associated with 1.43 times higher odds for multimorbidity. The relation was significantly positive between obesity and hypertension, arthritis, asthma, diabetes and chronic back pain²⁷. In another study from Finland using participants from the UK, researchers found that individuals with obesity started presenting complex multimorbidity by age 55, whereas individuals with healthy weight would only show multimorbidity by age 75. Also, they found that the degree of obesity was associated with complex multimorbidity in a dose-response relationship¹³.

The co-occurrence of depression and multimorbidity has been described by numerous previous studies^18,28. Zhang et al. (2022) found that multimorbidity was associated with increased odds of having depressive symptoms (adjusted odds ratio of 1.51), and developing depressive symptoms (adjusted odds ratio of 1.65). Depression worsens the outcomes of people living with multimorbidity increasing disability, accelerating cognitive decay and also increasing costs in older individuals^5,7,29. Moreover, depression was associated with a higher incidence of somatic disease, cardiometabolic disease, musculoskeletal diseases and digestive diseases in a 6-year follow-up cohort in the Netherlands³⁰. Momen et al. (2020) conducted a cohort study using 16 years of health data in Denmark and found that individuals with a mood disorder had a cumulative incidence within 15 years after the diagnosis of a mental disorder for circulatory conditions of 40.9% (95% CI, 40.5 to 41.3). Another study using the Center for Epidemiological Studies Depression Scale (CES-D) found a positive association between depression score and chronic disease incidence. Every one point symptom increase in the CES-D scale resulted in a 5% increase in incidence of chronic diseases³¹. Given the prevalence of depression, the public health implications of this are substantial.

The study findings should be interpreted in the context of some limitations. We used the term ‘effects’ throughout the paper, as is the standard in SEM analyses, but it was a cross-sectional study, and therefore should not be considered a causal investigation. In the survey used in the study, the diagnosis of depression was self-reported, and no diagnostic instrument was used. The negative coefficients from age and BMI to depression may be related to the continuous treatment of some variables that are categorical or ordinal, which might have interfered in the variances for the latent variable representing depression. Despite these limitations, SEM allowed us to model measurement error, which is expected in survey data. This is often overlooked in psychiatry research and is an important shortcoming in survey data of psychiatric outcomes. Also, the fit indices suggest that the chosen model was parsimonious.

Conclusion

We have presented here evidence of the effect of depression on multimorbidity in a large, nationwide representative data set. Several mechanisms can explain the effect in this direction. Subjects with depression might adhere less to clinical treatments, engage less in healthy life style, exercise less often, or anti-depressant use may increase multimorbidity by its metabolic effects. However, this can only be confirmed in studies with longitudinal designs. A future direction in this field could be to investigate whether the effect of depression on multimorbidity is bidirectional, which can be assessed in quasi-experimental methods such as cross-lagged panel data models.

Individual contributions

All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by Elisângela da Silva Rodrigues and Rodrigo Lins Rodrigues. The first draft of the manuscript was written by Jacyra Paiva de Araujo and Luis Castro-de-Araujo. Richard AA Kanaan contributed to later versions of the manuscript. All authors commented on versions of the manuscript, read and approved the final manuscript.

Conflicts of interest

Authors report no conflicts of interest

Supplementary Material

Distribution of age by sex

• Download figure
• Open in new tab

Missing and categories and distribution in one picture

Missings, categories and distributions

Survey weights in structural equation modeling (SEM) with imputed data

• Download figure
• Open in new tab