Association Between Depression and Physical Conditions Requiring Hospitalization

This cohort study examines data from the UK Biobank and 2 Finnish cohorts to gauge the association between depression and an array of physical conditions requiring hospital treatment.

M ental health problems are a major contributor to disease burden in high-income countries, with depression ranking among the top 10 causes of years lost to disability. 1 While major depressive disorder (MDD) affects approximately 5% of the general population, 2 its prevalence has been found to be markedly higher in patients with chronic medical conditions, such as myocardial infarction (29%), 3,4 type 2 diabetes (28%), 5 Parkinson disease (23%), 6 stroke (18%), 7,8 cancer (16%), 9,10 and Alzheimer disease (13%). 11 There is growing evidence that depression may act as an etiological risk factor for the development of specific physical illnesses [12][13][14][15][16][17][18] and that comorbid depression may exacerbate disease progression. [19][20][21] It has also been hypothesized that some depression-disease associations may be bidirectional, including those of depression with coronary heart disease, 18,22 stroke, 16,23 and diabetes. 15,24 To date, the associations of depression with a full range of serious diseases are not well understood. Existing studies on multiple disease outcomes are often characterized by smaller sample sizes, 25-27 thereby limiting the array of end points under study. Furthermore, these investigations did not assess the role of depression severity in disease risk, evaluate the extent to which the associations between depression and diseases are bidirectional, or examine the most common causes of hospitalizations in people with depression. 14, [25][26][27][28][29] Using data from the UK Biobank, a prospective cohort study, we conducted a large-scale, outcomewide investigation of the association of self-reported and physiciandiagnosed depression measures and subtypes with the incidence of 77 common health conditions requiring hospital treatment. To test reproducibility of our findings, we re-peated analyses in an independent study population drawn from a country with a different health care system.

Key Points
Question What are the most common conditions requiring hospital treatment in people with depression?
Findings In this cohort study that included 240 433 individuals, depression was associated with an increased risk of 29 hospital-treated conditions. For 12 of these conditions, there was evidence for a bidirectional relationship with depression; the highest cumulative incidence was observed for diseases of the endocrine, musculoskeletal, and circulatory systems.
Meaning In this study, the main causes of excess hospitalizations in people with depression were endocrine, musculoskeletal, and vascular diseases. Analyses were repeated in a pooled data set of 2 Finnish cohort studies: the population-based Health and Social Support Study (HeSSup) 31

Assessment of Depression and Covariates at Baseline
In UK Biobank, depression was ascertained from the 9-item version of the Patient Health Questionnaire (PHQ-9), which assesses how often over the past 2 weeks an individual experienced each of the 9 depressive symptoms embodied in the DSM-IV (0 for not at all, 1 for several days, 2 for more than half the days, and 3 for nearly every day). 33 Total scores (range, 0-27) were computed by summing individual item responses, and thereafter, 3 severity categories were derived: no depression (0-4, reference group), mild to moderate depression (5-14), and severe or moderately severe depression (≥15). In addition, we computed a binary depression variable (yes, no) with scores of 10 or higher indicating depression. 34 Further information on depression was obtained from 3 previously established, UK Biobank-based depression indicators, 35 denoting probable recurrent severe major depression (yes, no), probable recurrent moderate major depression (yes, no), and probable single major depression episode (yes, no). (Details of the methods used to derive and operationalize these variables have been described by the UK Biobank. 36 ) Data on these measures were collected from participants' responses to a touchscreen questionnaire at recruitment to UK Biobank. Questions were based on the Structured Clinical Interview for DSM-IV Axis I Disorders. While the focus of this study was on unipolar depression measures, we also report data on probable bipolar depression (yes, no); these results are shown in eTables 5, 8, and 15 in Supplement 1.
In HeSSup and FPS, available measures of depression included self-reported physician-diagnosed depression (yes, no) and a history of recorded hospitalizations with a depression diagnosis (yes, no, based on codes F32-F33 from the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision [ICD-10]) drawn from national hospital admission registers.
We used several covariates captured at baseline. Sociodemographic variables were age and sex. Educational qualification (low, medium, high) was used as an indicator of socioeconomic position. Health behaviors included self-reported smoking status (never, previous, current), alcohol consumption (none/low, medium, high), and physical activity (physically active, not active). In UK Biobank, self-reported ethnic origin (White or non-White) was added as an additional covariate. Participants chose among the categories Asian/Asian British, Black/Black British, Chinese, White, mixed (multiracial), and other ethnic group. With most UK Biobank participants having White ethnic origin, there were too few health events at follow-up to facilitate analyses across individual minority groups. We therefore categorized these data as White or non-White.

Ascertainment of Incident Hospital-Treated Conditions at Follow-up
In UK Biobank, newly developed physical and mental health conditions requiring hospital treatment were ascertained from linkage data to the UK National Health Service (NHS) Hospital Episode Statistics database for hospital admissions and the NHS Central Registry for mortality, from March 1995 until May 2021. In HeSSup and FPS, participants were linked to national hospital discharge and mortality registries. These electronic health records provided information on the cause and date of hospital discharge or mortality, or both, from January 1996 to December 2018. The mean (SD) follow-up was 4.6 (0.14) years in UK Biobank, 12.5 (3.93) years in FPS, and 13.8 (1.26) years in HeSSup.
In all 3 cohort studies, individual diseases and disease categories were coded according to the ICD-10. We investigated 77 predefined ICD-10 disease chapters and diagnostic groups designed for outcomewide analyses. 37,38 For each health outcome, participants with the disease at or before baseline, as ascertained from the linked health registry data, were excluded from the analysis of incident hospital-treated conditions.

Statistical Analysis
In primary analyses of UK Biobank participants, and after confirmation that the proportional hazards assumption had not been violated (eTables 1, 2, 3, and 4 in Supplement 1), we performed separate Cox proportional hazards regression models 39 to examine the associations of each depression measure with 77 incident disease outcomes. Hazard ratios (HRs) and accompanying 95% CIs were adjusted for age and sex and, additionally, ethnic origin, education, smoking, alcohol consumption, and physical activity. To identify the most robust associations, 40 subsequent analyses were restricted to associations that remained statistically significant at a Bonferronicorrected α level of P < 6.49 × 10 −4 (ie, adjustment for 77 tests) and had an HR greater than or equal to 1.50. In a sensitivity analysis, we accounted for missing data on covariates by producing a multiple imputation model based on chained equations (n = 20). 41 To examine total disease burden over time in individuals with severe/moderately severe depression, we calculated the cumulative incidence per 1000 persons for disease categories that were most consistently associated with depression.
To investigate whether depression was associated with disease progression, we examined the risk of being hospitalized because of circulatory conditions in a subgroup of UK Biobank participants with self-reported heart problems but no record of hospitalization due to circulatory conditions at baseline (n = 23 509) and the risk of being hospitalized because of diabetes in a subgroup of UK Biobank participants with diabetes who had no record of hospitalizations due to this condition at baseline (n = 4161).
To examine the reproducibility of the associations between depression and incident disease, we repeated analyses in pooled data from HeSSup and FPS by computing HRs and 95% CIs with the same statistical adjustments and for the same 77 disease end points. In these analyses, cohort was used as an additional covariate. To examine the bidirectionality of robust depression-disease associations in a subpopulation of 57 166 FPS participants with repeat data on both depression and diseases, we used multivariable-adjusted logistic regression analyses and explored whether participants with the hospital-treated disease of interest but no depression at baseline were at increased risk of developing depression at follow-up relative to individuals without depression and the disease at baseline.
All UK Biobank analyses were conducted using Stata version 17.0 and those of FPS and HeSSup using SAS version 9.4. Statistical code is provided in the eMethods in Supplement 1.

Results
The analytical sample of UK Biobank participants consisted of 130 652 individuals (71 565 Table 1). Differences in characteristics between included and excluded participants are shown in eTable 6 in Supplement 1.
In Figure 2, we summarize the multivariable-adjusted associations of each depression measure with main ICD-10 disease groups according to strength and statistical significance in UK Biobank (results for all 77 disease outcomes are reported in eTables 7-11 and 16 in Supplement 1). After adjustment for age, sex, ethnic origin, education, smoking status, alcohol consumption, and physical activity, severe/moderately severe depression according to PHQ-9 score was most strongly associated with incident disease. The effect estimates for mild to moderate depression (PHQ-9) and a probable single major depression episode (UK Biobank definition) were smaller. In analyses using an alternative PHQ-9 cutoff of 10 or higher, the effect sizes were between those of mild/moderate depression and severe/moderately severe depression.
In Table 2, we show the associations of severe/moderately severe depression with 29 nonoverlapping disease outcomes that were robust to multivariable adjustment, had an HR of greater than or equal to 1.50, and were statistically significant at a Bonferroni corrected α level, P < 6.49 × 10 −4 in UK Biobank. Individuals with severe/moderately severe depression experienced an increased risk of hospitalizations due to poisoning, falls, and injuries (range of HRs, 1.86-23.03); diseases of the endocrine, genitourinary, and digestive systems (HRs, 1.67-6.97); and mental and behavioral disorders and diseases of the nervous system (HRs, 3.67-16.47). They also experienced a more than 1.5-times increased risk for diseases of the musculoskeletal system (HRs, 1.80-6.80); diseases of the respiratory system (HRs, 2.65-4.11); diseases of the circulatory system, blood, and related symptoms (HRs, 1.76-4.38); diseases of the ear or eye (HRs, 1.52-2.76); and infections and diseases of the skin (HRs, 2.01-2.52). Use of multiple imputa- tion analysis to supplement missing values on covariates did not materially change the results (eResults 1 and eTable 12 in Supplement 1). In analyses exploring whether depression was associated with disease progression in UK Biobank participants with prevalent heart conditions, individuals with mild/moderate depression had a 1.26-times increased risk (95% CI, 1.14-1.40) and those with severe/moderately severe depression a 1.92-times increased risk (95% CI, 1.52-2.43) of hospitalizations due to circulatory conditions at follow-up relative to participants without depression at baseline. Similar results were found in a subgroup of participants with self-reported physician-diagnosed diabetes at baseline who had not been hospitalized because of the condition. Relative to participants without depression at baseline, individuals with mild/moderate depression had a 2.35-times increased risk (95% CI, 1.38-3.98), and those with severe/moderately severe depression a 3.60-times higher risk (95% CI, 1.46-8.90) of hospitalization due to diabetes at follow-up (eResults 2 in Supplement 1).
To describe the burden of depression-related comorbidities over time in absolute terms in UK Biobank, we computed the cumulative incidence during the follow-up for 8 disease categories in people with severe/moderately severe depression (Figure 3). These categories combine conditions that were most consistently associated with this depression measure after adjustment for covariates and multiple testing ( Table 2). The highest 4-year cumulative incidence and the greatest difference in cumulative incidence between participants with and without severe/moderately severe depression were observed for endocrine and related internal organ diseases, with 245 per 1000 persons with depression vs 147 per 1000 persons without depression at baseline requiring hospital treatment because of these diseases (absolute excess risk compared with people without depression, 9.8%). Musculoskeletal diseases ranked second   diagnosed depression, and 492 (0.5%) had been hospitalized because of a depression diagnosis at baseline (Table 1). Physiciandiagnosed depression in the replication cohorts was almost as strongly associated with incident disease as severe/ moderately severe depression in UK Biobank. Accordingly, robust associations of physician-diagnosed depression were confirmed for 25 of the 29 nonoverlapping diseases that were associated with severe/moderately severe depression in UK Biobank participants (Figure 2 and Table 2). The patterns of associations were similar for people who were hospitalized because of depression, albeit with smaller effect estimates ( Figure 2). In an analysis of repeat data from 57 166 FPS participants (Table 2, right column, and eTable 13 in Supplement 1), we found support for a bidirectional association of depression with poisonings, falls, digestive and abdominal symptoms, diseases of the digestive system, mood disorders, sleep disorders, sciatica, soft tissue disorders, circulatory and respiratory symptoms, and bacterial infections, with odds ratios ranging from 1.2 (diseases of the digestive system) to 5.4 (mood disorders).

Discussion
The findings of this multicohort study suggest that, compared with individuals without depression, those with selfreported severe or moderately severe depression have at least a 1.5-times higher risk of 29 nonoverlapping conditions across multiple organ systems. The greatest absolute risk was observed for endocrine and related internal organ diseases, followed by musculoskeletal diseases and diseases of the circulatory system and blood. Among people with depression, the cumulative incidence for hospital-treated mental disorders was lower, possibly due to conditions such as mood and anxiety disorders being predominantly treated in primary care. There was also evidence for bidirectional depression-disease relationships, such that poisonings; falls; diseases or symptoms of the circulatory, respiratory, digestive, and musculoskeletal systems; and severe infections were associated with subsequent onset of depression.
By applying an outcomewide approach to a single analytical setting, this study provides novel insights into depressionrelated hospitalizations and also confirms the results of several previous single-outcome studies. Overall, our findings emphasize the important role of depression as a risk factor for physical illnesses requiring hospital treatment. The observed elevated risk for poisonings; headaches; neurotic, stress-related, and somatoform disorders; sleep disorders; obesity; back pain; and chronic obstructive bronchitis in individuals with depression confirm previous meta-analytic investigations and large-scale studies. [27][28][29]42,43 Our results also support earlier evidence on the prospective links between depression and diabetes, 15 Parkinson disease, 12 and heart disease. 27,44,45 In addition, they are in agreement with a recent large-scale study of around 6 million people that captured data on mood disorders and subsequent medical diseases from linkage to national registries. 28 In that study, in accordance with our findings, strong associations were observed with diseases of the blood and endocrine system and weak or nonexisting associations with cancers. It has been hypothesized that depression may also affect disease progression, for example, by reducing adherence to prescribed treatments or poorer self-care. 20,21, 46 We found that depression was associated with worse disease prognosis in people with selfreported heart problems and diabetes at baseline.
There are several plausible mechanisms for the observed associations. Behavioral pathways may involve smoking, par-  ticularly in relation to diseases such as chronic obstructive bronchitis 47 and heart disease 46 ; lack of physical activity for conditions such as obesity, diabetes, cardiovascular disease, and musculoskeletal disorders; and high alcohol consumption in relation to diseases of the endocrine, digestive, genitourinary, and circulatory systems. 46,48 There is also some evidence pointing toward shared genetic variation between depression and an array of neurological, circulatory, musculoskeletal, and respiratory disorders. 29 Further shared biological mechanisms that may underlie multiple comorbid conditions in depression include alterations in levels of monoamines, which have been linked to sleep problems 49 and pain 50 ; as well as a dysregulated hypothalamic-pituitary-adrenal axis for cardiometabolic diseases. 46,51 Another plausible mechanism is inflammation. A recent case-control analysis of UK Biobank participants suggests that levels of systemic inflammation were significantly higher in people with depression compared with the reference group without depression. 52 Systemic inflammation has also been found to contribute to a wide range of conditions, including obesity 53 and cardiovascular disease. 46 Lastly, the observed bidirectional associations in our study demonstrate that severe physical illnesses may be associated with poor mental health. In recent years, there has been emerging interest in precision medicine approaches to psychiatry. In our study, the identified depression-disease associations were largely consistent across depression measures, although there was evidence for a dose-response effect by severity and chronicity. We also observed widespread but slightly weaker depressiondisease associations for inflammation-related 5 4 and obesity-related 55 symptom profiles, as well as for probable bipolar depression, suggesting that most depression-disease associations persist across varying symptom expressions (eTable 8 in Supplement 1). Thus, our findings support only partially the notion that different symptoms may have distinct underlying etiological pathways that contribute to variability in associations with disease pathologies. 51,54 These hypotheses warrant further testing in future large-scale studies.

Limitations
Interpretation of our findings requires consideration of various limitations. First, causal inference is not possible given that all the studies included in the present analysis are observational. Conclusions about the directions of associations may also be hampered owing to missing data on undiagnosed diseases or disease diagnoses made in primary care. In addition, the utilization of data-driven HR thresholds (1.5) and a predefined set of common diseases with public health relevance may have resulted in an underestimation of the actual hospitalization burden associated with depression. Second, while we were able to test disease associations for various presentations of depression, no data were available on other subtypes, such as dysthymia or seasonal affective disorder. This highlights the need for future investigations to explore whether these subtypes are differentially related to disease risk.
Third, despite being directionally consistent, effect estimates for some depression-disease associations were higher in UK Biobank than in the Finnish cohorts. There are at least 2 explanations for these differences. The participants of FPS are employees, and there is evidence suggesting that occupational cohorts have more favorable risk factor profiles and a lower disease incidence compared with the general population. 56 Another plausible explanation is that the primary exposure in UK Biobank was severe/moderately severe depression, whereas severity levels of depression were not considered in the Finnish cohorts. It may be that the Finnish depression measure also captured milder cases of depression, which may explain some of the lower effect estimates observed in the analyses of the Finnish cohorts. Fourth, with the large majority of participants in the present studies having a White ethnic background, the generalizability of our results to minority groups is unknown.

Conclusions
Our findings show that the associations of depression with physical disease are widespread across multiple organ systems. The most common causes of hospitalization in our cohort among people with depression were endocrine, musculoskeletal, and vascular diseases. This suggests that depression should be considered more widely as a target for somatic disease prevention and treatment.