Introduction
Depressive disorder affects approximately 280 million people worldwide (GBD Mental Disorders Collaborators, 2022) and shortens their lifespan by about 13 years relative to people without the disorder (Chan et al., Reference Chan, Correll, Wong, Chu, Fung, Wong, Lei and Chang2023). Evidence has suggested that physical illnesses could explain about two-thirds of premature mortality in patients with mental disorders (Walker et al., Reference Walker, McGee and Druss2015), and recent evidence has further shown that a substantial proportion of mortality in people with depression was due to cardiovascular diseases (CVDs) (Correll et al., Reference Correll, Solmi, Veronese, Bortolato, Rosson, Santonastaso, Thapa-chhetri, Fornaro, Gallicchio, Collantoni, Pigato, Favaro, Monaco, Kohler, Vancampfort, Ward, Gaughran, Carvalho and Stubbs2017; Lo et al., Reference Lo, Chan, Wong, Chung, Correll, Solmi, Baum, Thach, Sham and Chang2025). Type 2 diabetes mellitus is an important CVD risk factor (Vancampfort et al., Reference Vancampfort, Mitchell, De Hert, Sienaert, Probst, Buys and Stubbs2015a) and, in itself, a leading global disability (GBD 2019 Diseases and Injuries Collaborators, 2020), affecting approximately 529 million people (GBD 2021 Diabetes Collaborators, 2023). Diabetes and its complications affect multiple systems, particularly the cardiovascular system, including coronary heart disease and stroke, which are respectively the second and third leading causes of global disability (GBD 2019 Diseases and Injuries Collaborators, 2020).
Existing studies have in fact indicated bidirectional relationships between depression and type 2 diabetes (Fanelli et al., Reference Fanelli, Raschi, Hafez, Matura, Schiweck, Poluzzi and Lunghi2025; Pouwer et al., Reference Pouwer, Ehrmann, Strandberg, Ismail, Aalders, Rutters, de Wit, Speight, Iversen and Holt2026). Accumulating evidence has shown that patients with depression are at increased risk of hyperglycemia and type 2 diabetes compared to those without depression (Vancampfort et al., Reference Vancampfort, Correll, Wampers, Sienaert, Mitchell, De Herdt, Probst, Scheewe and De Hert2014, Reference Vancampfort, Stubbs, Mitchell, De Hert, Wampers, Ward, Rosenbaum and Correll2015b; Lindekilde et al., Reference Lindekilde, Scheuer, Rutters, Knudsen, Lasgaard, Rubin, Henriksen, Kivimaki, Andersen and Pouwer2022). Some studies have also demonstrated elevated cytokine and chemokine alterations in depression compared to the comparison (Köhler et al., Reference Köhler, Freitas, Maes, Andrade, Liu, Fernandes, Stubbs, Solmi, Veronese, Herrmann, Raison, Miller, Lanctôt and Carvalho2017), implying shared immunological pathways across both depression and diabetes (Moulton et al., Reference Moulton, Pickup and Ismail2015). Conversely, diabetes may increase the risk of depression due to factors such as physiological stress and emotional responses to long-term complications (Gonzalez et al., Reference Gonzalez, Fisher and Polonsky2011). In addition, the co-occurrence of depression and diabetes is presumably driven by numerous shared risk factors, including unhealthy lifestyle (Pearce et al., Reference Pearce, Garcia, Abbas, Strain, Schuch, Golubic, Kelly, Khan, Utukuri, Laird, Mok, Smith, Tainio, Brage and Woodcock2022) and polypharmacotherapy (Jeffery et al., Reference Jeffery, Bhanu, Walters, Wong, Osborn and Hayes2023). As diabetes is a major risk factor for CVD-related death in depression, it is clinically important to confirm the impact of depression on the outcomes of diabetes (Dragioti et al., Reference Dragioti, Radua, Solmi, Gosling, Oliver, Lascialfari, Ahmed, Cortese, Estradé, Arrondo, Gouva, Fornaro, Batiridou, Dimou, Tsartsalis, Carvalho, Shin, Berk, Stringhini, Correll and Fusar-Poli2023). A handful of studies have systematically evaluated the associations between depression and diabetes-related outcomes, including mortality and complication risks (van Dooren et al., Reference van Dooren, Nefs, Schram, Verhey, Denollet and Pouwer2013; Farooqi et al., Reference Farooqi, Khunti, Abner, Gillies, Morriss and Seidu2019; Nouwen et al., Reference Nouwen, Adriaanse, van Dam, Iversen, Viechtbauer, Peyrot, Caramlau, Kokoszka, Kanc, de Groot, Nefs and Pouwer2019), but the findings were hampered by several important limitations. First, the use of self-reported questionnaires to define depression in the vast majority of the included studies may identify depressive symptoms at a subclinical level rather than a diagnosis of clinical depression (Roy et al., Reference Roy, Lloyd, Pouwer, Holt and Sartorius2012). Second, these depressive symptoms might indicate diabetes distress (Fisher et al., Reference Fisher, Skaff, Mullan, Arean, Mohr, Masharani, Glasgow and Laurencin2007), which is clinically distinct from depressive disorder, and thus might potentially blur the effect of depressive disorder on outcome estimates. Third, using these tools with variable cut-off values to define depression might lead to an inconsistent assessment of depression across studies, leading to problems with the comparability of depression exposure. Importantly, studies examining the association between depression and diabetes complications were inconsistent, with some showing higher risks (Yoo et al., Reference Yoo, Han, Rhee and Lee2023; Scheuer et al., Reference Scheuer, Andersen, Carstensen, Diaz, Kosjerina, Lindekilde, Wild, Jackson, Pouwer, Benros and Jørgensen2024), and others lower or comparable risks (Wu et al., Reference Wu, Hsu and Wang2020; Ho et al., Reference Ho, Chan, Lo, Fang, Wong, Lee, Lai, Ng, Chen, Wong and Chang2025a). As the associations of depression with a wide spectrum of specific diabetes complications have not yet been systematically evaluated, quantitative synthesis is indicated to aggregate the results to inform healthcare policy-making, particularly the formulation of tailored preventive approaches of diabetes in patients with depression to improve their clinical outcomes.
To this end, we conducted a systematic review and meta-analysis to examine the risk of mortality and complications among patients with depression and co-occurring diabetes (i.e., depression-diabetes group) relative to patients with diabetes only (i.e., diabetes-only group), with an aim to compare between the two patient groups on their all-cause mortality rates, and if applicable, other cause-specific mortality rates, as well as the occurrence of specific diabetes complications. Additionally, stratified analyses by pre-specified study-level characteristics, including age, gender, study period, geographic region, follow-up duration and nature of diabetes sample, were performed.
Methods
Search strategy and selection criteria
We systematically searched across the electronic databases Embase, MEDLINE, PsycInfo and Web of Science for articles published from inception to 20 December 2024. No restrictions were placed on language or geographic regions. The search strategy was developed by combining keywords, including depressive disorder, diabetes mellitus, mortality and complications. The full search strategy for each database is provided in Supplementary Table S2. The references of eligible articles were manually reviewed to identify additional relevant studies. Studies were included if they fulfilled the following criteria: (1) included patients diagnosed with diabetes mellitus, (2) reported on the risk of mortality or complications in patients with depression (defined as major depressive disorder ± dysthymia) and co-occurring diabetes compared to those with diabetes alone, and (3) were observational in design. Studies with ascertainment of mental disorders based on self-reported measures or unclear diagnostic criteria, which could not be mapped to DSM/ICD criteria, or classified based on symptoms but not disorders, were excluded. Publications addressing solely on specific population subgroups, or with a sample size of fewer than 50, were excluded. Conference abstracts, editorials, protocols and reviews that lack original data were also excluded. This meta-analysis was conducted and reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) (Supplementary Table S1). The study protocol was registered prospectively with PROSPERO (CRD42024595145). Since this research was based on previously published literature, no ethical approval or patient consent was required.
Study selection and data extraction
Duplicates among different databases were identified and removed, after which titles and abstracts were screened for relevant studies. Full texts of potentially eligible publications identified at screening were evaluated by two reviewers (M.T.H.H. and W.C.Y.C.) for inclusion. Disagreement was resolved through discussions or consultation with the supervisor (L.L.W.T.). Reasons for exclusion at this stage were documented. Data extraction included the following information: first author’s surname, year of publication, geographic region, study design, study period, types of diabetes, nature of diabetes sample (incident or prevalent), sample size of patients with depression and diabetes, sample size of patients with diabetes only, percentage of male in both groups, age at baseline with its standard deviation in both groups, outcome measure, cumulative incidence of outcome, risk estimates with 95% confidence intervals (CI), methods to ascertain depression, depression diagnostic system, inclusion of pre-existing depression, methods to ascertain diabetes and mortality/complication outcome, follow-up duration and matching and/or adjustment variables.
Risk estimates, when not reported numerically in the report, were extracted from the figures using WebPlotDigitizer (https://automeris.io/WebPlotDigitizer) (De Prisco et al., Reference De Prisco, Tapoi, Oliva, Possidente, Strumila, Takami Lageborn, Bracco, Girone, Macellaro, Vieta and Fico2024). For studies reporting outcome measures at multiple time points, the outcomes at the latest follow-up time point were selected. When several adjusted risk estimates were reported for the same outcome, the most comprehensively adjusted estimate was selected. Diabetes complications were categorized in broad terms as microvascular complications and macrovascular complications (Zheng et al., Reference Zheng, Ley and Hu2018). To categorize specific types of diabetes complications, the Diabetes Complications Severity Index (DCSI) was adopted (Young et al., Reference Young, Lin, Von Korff, Simon, Ciechanowski, Ludman, Everson-stewart, Kinder, Oliver, Boyko and Katon2008), which encompasses seven categories of complications, including cardiovascular, cerebrovascular, peripheral vascular, nephropathy, neuropathy, retinopathy and metabolic complications. Metabolic complications in DCSI encompass ketoacidosis, hyperosmolar state and other coma. Risk estimates were reported with 95% CI, with greater than one denoting a statistically significant increase in risk of mortality/complications in patients with depression. Risk-of-bias was independently assessed by two reviewers (M.T.H.H. and W.C.Y.C.) using the Newcastle-Ottawa Scale (NOS) (Wells et al., Reference Wells, Shea and O’connell2009).
Statistical analysis
Data synthesis was performed quantitatively. Odds ratio (OR), hazard ratio (HR) and incidence rate ratio (IRR) were treated as equivalent measures of risk to maximize data inclusion, in line with prior meta-analyses on mortality in mental disorders (Chen et al., Reference Chen, Ejlskov, Laustsen, Weye, Sørensen, Momen, Dreier, Zheng, Damgaard, McGrath, Sørensen and Plana-Ripoll2024; Ronaldson et al., Reference Ronaldson, Santana, Carlisle, Atmore, Chilman, Heslin, Markham, Dregan, Das-munshi, Lampejo, Hotopf and Bakolis2024). The term ‘relative risk’ (RR) is used thereafter. The analyses were also restricted to risk estimates reported by HR. The extracted risk estimates were log-transformed to achieve a normalized distribution, and their CIs were converted to standard errors (SE). The converted risk ratios and SE were then used to generate pooled estimates. As the included studies were conducted over different time periods and regions, substantial heterogeneity was anticipated across studies, and thus, random-effects meta-analysis was selected. Heterogeneity was evaluated using the I 2 statistic (Higgins et al., Reference Higgins, Thompson, Deeks and Altman2003), and Cochran’s Q-test was performed to assess the statistical significance of the heterogeneity across studies. Results were displayed graphically by Forest plot, presenting pooled estimates with 95% CIs and weight. When multiple risk estimates by subgroup were provided, the risk estimates were combined to derive a single estimate per study (with reference to van Dooren et al., Reference van Dooren, Nefs, Schram, Verhey, Denollet and Pouwer2013), before they were pooled into a summary estimate across studies. Publication bias was examined visually through funnel plot and statistically by Egger’s regression asymmetry test, with P < 0.1 considered significant. Synthesized estimates subject to publication bias were further corrected for using Duval and Tweedie’s trim-and-fill procedure. Two sets of sensitivity analyses were conducted, including (1) restricting the analyses to studies classified as fair or good quality, and (2) leave-one-out analysis. Subgroup analyses were performed, stratified by pre-specified study-level characteristics, including the nature of the diabetes sample (incident or prevalent), the study period (categorized into ‘after year 2010’ or ‘before or in year 2010’ based on the middle year of the cohort data collection) and geographic regions. Univariate meta-regression was performed for age, gender (i.e., percentage male) and follow-up duration. Statistical analyses were performed in R (version 4.0.2) with metafor package. For all analyses, except Egger’s test, P < 0.05 was considered statistically significant.
Results
Study selection
The PRISMA flow diagram showing the screening and identification of studies is shown in Figure 1. A total of 53,181 records were identified in the literature search, and after removal of duplicates, 42,132 remained. After screening titles and abstracts, 79 full-text articles underwent full-text review. Of these, 56 were excluded due to: the exposure was not depressive disorder (n = 5), there were no DSM/ICD criteria for depressive disorder (n = 39), the comparison group was not patients with diabetes alone (n = 4); the study did not report relevant outcomes (n = 6), overlapping data (n = 1) and inadequate sample size (n = 1). An additional two records were identified from citation searching. Reasons for study exclusion are shown in Supplementary Table S3. Twenty-six studies met the inclusion criteria and were included in the meta-analysis.
PRISMA flow diagram for study selection.
Figure 1 Long description
The flowchart illustrates the PRISMA process for study selection. It begins with the identification of studies via databases, listing 53,181 records identified from sources like Embase, MEDLINE, PsycINFO and Web of Science. After removing 11,049 duplicates, 42,132 records remain. Additionally, 2 records are identified from citation searching. During screening, 42,053 records are excluded, leaving 79 full-text reports assessed for eligibility. Of these, 56 reports are excluded for reasons such as not meeting DSM/ICD criteria for depressive disorder, irrelevant outcome estimates, incorrect comparison group, overlapping data and inadequate sample size. Two reports are assessed for eligibility from other methods. Finally, 25 reports with 26 studies meet the eligibility criteria, including 10 mortality outcomes and 16 complication outcomes.
Study characteristics
Of these 26 studies, 19 studies were included in the meta-analysis for mortality outcomes and 16 studies for complication outcomes. As shown in Table 1, studies were conducted in the South Korea (n = 6), United States (n = 6), United Kingdom (n = 4), Taiwan (n = 3), Denmark (n = 2), Hong Kong (n = 2), Australia (n = 1), Canada (n = 1) and Spain (n = 1), which belonged to four continents including Asia (n = 11), Europe (n = 7), North America (n = 7) and Australia (n = 1).
Characteristics of the included studies
Table 1 Long description
Study characteristics for 23 longitudinal studies compare people with depression plus diabetes versus diabetes only, reporting location, design, years, diabetes type, incident or prevalent sampling, baseline age, sample sizes, sex distribution, ethnicity, and outcomes. Most studies are cohort designs spanning the 1990s through 2025, with many conducted in Asia and Europe and several in North America. Type 2 diabetes predominates, though multiple studies include both type 1 and type 2, and a few focus on type 1 only. Many samples are very large, including national cohorts with hundreds of thousands to millions of participants, while a few early studies are small. Baseline ages are typically in the late fifties to early seventies when reported, but age and sex data are frequently missing for some studies. Outcomes most often include all-cause mortality and cardiovascular outcomes, with additional reporting of cerebrovascular, peripheral vascular, nephropathy, neuropathy, retinopathy, and metabolic complications. Several studies explicitly assess whether depression predates diabetes, and both incident and prevalent diabetes cohorts are represented, which affects how temporality and risk estimates can be interpreted. Ethnicity is reported in only a subset of studies, commonly noting predominantly White samples in United Kingdom and United States cohorts, limiting cross-study comparability on demographic composition.
CVD = cardiovascular disease, MDD = major depressive disorder; NA = not available; N/A = not applicable; SD = standard deviation.
a The last name of the first author of the publication, in reverse chronological order of the date of publication.
b Refers to whether the study estimating the risk of mortality or complications based on an incident diabetes sample (i.e., patients with first diagnosed diabetes at the start of follow-up period) or a prevalent diabetes sample (i.e., patients with onset of diabetes mellitus prior to the start of follow-up period).
c Refers to whether the study examined the association by using pre-existing depressive disorder (i.e., depressive disorder that precedes diabetes) to delineate the temporal relationship of the association.
d DEP refers to people with depression and co-occurring diabetes mellitus (i.e., depression-diabetes group), and control refers to people with diabetes only (i.e., diabetes-only group).
Study periods spanned from 1982 to 2022 and were categorized into ‘after 2010’ (n = 10) or ‘before or in 2010’ (n = 16). Almost all studies used cohort design, with one adopting a nested case-control design. Seven studies focused on patients with incident diabetes, and 19 studies on patients with prevalent diabetes. Additional characteristics of individual studies are summarized in Supplementary Table S4. Twenty-four of the 26 studies ascertained the depression diagnosis by identifying diagnostic codes (ICD or DSM) in a health record database, while the remaining two studies used diagnostic interview (Clouse et al., 2003; Winkley et al., 2012) (Supplementary Table S4). The covariates of the individual studies are summarized in Supplementary Table S5, and the outcome estimates are summarized in Supplementary Tables S6 and S7. Of these 26 studies, the quality assessment showed that 25 were of fair to good quality, and 1 was of poor quality. It indicated that the majority of all studies were at low risk of bias. The corresponding NOS scores for each of the included studies on mortality and complication outcomes are presented in Supplementary Tables S8 and S9, respectively.
Meta-analysis and stratified analyses of mortality outcomes
The risk estimates for all-cause and CVD-specific mortality are presented in Figure 2. The pooled estimate showed that the depression-diabetes group was associated with increased risk of all-cause mortality (RR 1.30 [95% CI: 1.21–1.39]), with significant heterogeneity (Q 18 = 983.2, P < 0.001, I 2 = 98.3%, k = 19). Visual inspection of the funnel plot (Supplementary Figure S1) and Egger’s test (z = 1.49, P = 0.137) suggested no publication bias. The synthesized estimate remained consistent when studies identified as of poor quality were excluded (1.29 [1.20–1.38]; Q 17 = 769.0, P < 0.001, I 2 = 98.0%, k = 18) and in leave-one-out analysis (1.27–1.32). Among the 19 studies assessing the all-cause mortality risk, 16 used HRs, 2 used ORs and 1 used IRRs. The synthesized estimate was (1.30 [1.20–1.40]); Q 15 = 961.7, P < 0.001, I 2 = 98.6%, k = 16) when the analysis was restricted to HR. Similarly, the pooled estimates showed that the depression-diabetes group was associated with increased risk of CVD-specific mortality (1.15 [1.02–1.29]; Q 4 = 20.1, P < 0.001, I 2 = 81.4%, k = 5). No evidence of publication bias was found in Egger’s test (z = 0.56, P = 0.578). Among the five studies assessing the CVD-specific mortality risk, four used HRs and one used ORs. The pooled estimate was (1.14 [0.99–1.31]); Q 3 = 19.9, P < 0.001, I 2 = 87.1%, k = 4) when the analysis was restricted to HR, and remained consistent in leave-one-out analysis (1.11–1.20).
Forest plot for risk of all-cause mortality and CVD-specific mortality in diabetes patients with vs without depression. Studies were arranged in reverse chronological order (first author and their colleagues, year of publication).
Figure 2 Long description
Two stacked forest plots labeled “All-cause mortality” and “CVD-specific mortality,” each listing studies in a left column and showing square markers with horizontal confidence-interval lines aligned to a central vertical reference line at 1. All-cause mortality section: The right column header reads “RR (95 percent CI)”. The x-axis label reads “Estimate (95 percent CI)”, with tick labels 0.5, 1, 1.5, 2, 2.5, 3 and 3.5. The study rows and values shown are: Ho et al., 2025: 1.06 (1.02 - 1.10). Li et al., 2025: 1.26 (1.09 - 1.44). Kim et al., 2024: 1.31 (1.20 - 1.44). Fleetwood et al., 2023: 1.28 (1.22 - 1.35). Kim et al., 2022: 1.24 (1.22 - 1.27). Ung et al., 2021: 1.66 (1.60 - 1.72). Lughati et al., 2021: 1.32 (1.21 - 1.45). Fernandes de Abreu et al., 2020: 1.14 (1.07 - 1.22). Wu et al., 2020: 1.08 (1.04 - 1.12). Chiang et al., 2020 (Taiwan cohort): 1.23 (1.06 - 1.42). Chiang et al., 2020 (UK cohort): 1.41 (1.20 - 1.66). Jeong et al., 2017: 1.43 (1.41 - 1.46). Novak et al., 2016: 1.30 (1.28 - 1.31). Lynch et al., 2014: 1.05 (1.03 - 1.07). Ting et al., 2013: 1.07 (0.60 - 1.90). Winkley et al., 2012: 2.09 (1.34 - 3.25). Mai et al., 2011: 1.54 (1.37 - 1.73). Rikhardson et al., 2008: 1.60 (1.30 - 1.80). Katon et al., 2008: 1.36 (1.16 - 1.59). A pooled-effect row shows a diamond with: 1.30 (1.21 - 1.39). CVD-specific mortality section: The same right column header “RR (95 percent CI)” appears. The x-axis label remains “Estimate (95 percent CI)” with the same tick labels. The study rows and values shown are: Li et al., 2025: 1.26 (0.93 - 1.70). Fleetwood et al., 2023: 1.29 (1.17 - 1.42). Kim et al., 2022: 1.15 (1.09 - 1.21). Lughati et al., 2021: 1.19 (0.99 - 1.42). Wu et al., 2020: 0.96 (0.88 - 1.06). A pooled-effect row shows a diamond with: 1.15 (1.02 - 1.29).
Stratified analyses of all-cause mortality by geographic regions, study period and nature of diabetes are summarized in Table 2. Significant heterogeneity was observed in the risk of all-cause mortality and its stratified subgroups. In stratified analyses of all-cause mortality by geographic regions, the risk of all-cause mortality was higher in diabetes patients with depression than their non-depressed counterparts in Asia (1.27 [1.13–1.42]; Q 7 = 513.5, P < 0.001, I 2 = 98.7%, k = 8), in Europe (1.28 [1.16–1.41]; Q 4 = 15.6, P = 0.004, I 2 = 74.4%, k = 5), in North America (1.30 [1.14–1.48]; Q 4 = 366.7, P < 0.001, I 2 = 99.8%, k = 5) and in Australia (1.54 [1.37–1.73]) (Supplementary Figure S2). No significant group differences in mortality risk were detected for geographic regions (between-group P = 0.930). When stratified by study period, the risk in patients from studies conducted after 2010 (1.28 [1.16–1.41]; Q 7 = 316.7, P < 0.001, I 2 = 97.0%, k = 8) and from studies conducted before or in 2010 (1.32 [1.20–1.45]; Q 10 = 666.0, P < 0.001, I 2 = 98.7%, k = 11) were significantly higher in depression-diabetes group relative to the comparison group (Supplementary Figure S3). No significant group differences in mortality risk were detected for study period (between-group P = 0.298). When stratified by nature of diabetes, the risk of patients with prevalent diabetes (1.34 [1.24–1.46]; Q 13 = 815.0, P < 0.001, I 2 = 98.6%, k = 14) and those with incident diabetes (1.20 [1.09–1.32]; Q 4 = 61.4, P < 0.001, I 2 = 94.0%, k = 5) were significantly higher in the depression-diabetes group relative to the comparison group (Supplementary Figure S4). No significant group differences in mortality risk were detected for incident and prevalent diabetes status (between-group P = 0.116). Univariate meta-regression analyses showed a significant association between the percentage of male gender and the risk of depression-associated all-cause mortality in diabetes patients (β = 0.004 [0.003–0.008]; Qm = 4.44, P = 0.035), but not the study mean age at baseline of both groups (Qm = 0.19, P = 0.663) or the overall follow-up duration (Qm = 0.4, P = 0.517).
Random-effects meta-analytic models showing mortality outcomes and stratified analyses
Table 2 Long description
Random-effects meta-analysis estimates relative risk of mortality for people with both depression and diabetes compared with diabetes only, reporting study counts, confidence intervals, heterogeneity, and p values. All-cause mortality across 19 studies shows a relative risk of 1.30 with a confidence interval from 1.21 to 1.39 and very high heterogeneity at 98.3 percent, with a p value below 0.001. By region, relative risks are similar: Asia 1.27, Europe 1.28, and North America 1.30; the between-region comparison p value is 0.930, suggesting no meaningful difference by region. By study period, results are 1.32 for studies before or in 2010 and 1.28 for studies after 2010; the between-period p value is 0.298. By diabetes nature, incident diabetes shows 1.20 and prevalent diabetes shows 1.34; the between-group p value is 0.116. Cardiovascular disease-specific mortality across 5 studies shows a relative risk of 1.15 with a confidence interval from 1.02 to 1.29 and heterogeneity of 81.4 percent, with a p value of 0.018. Interpretation should note the consistently high heterogeneity, indicating substantial variation across studies even when pooled effects are statistically significant.
CI = confidence interval; CVD = cardiovascular disease; N/A = not applicable; RR = relative risk.
a P value for the comparison between depression-diabetes group versus diabetes-only group.
b P value for between-group comparison.
Meta-analysis and stratified analyses of complication outcomes
The synthesized estimate showed that the depression-diabetes group was associated with an increased risk of any complications (1.28 [1.18–1.40]) relative to the diabetes-only group, with significant heterogeneity (Q 15 = 686.4, P < 0.001, I 2 = 99.3%, k = 16) (Fig. 3). Publication bias was indicated by the asymmetrical funnel plot (Supplementary Figure S5) and a significant result of Egger’s test (z = 3.30, P = 0.001). The result remained robust after the trim-and-fill procedure was applied to correct for publication bias (1.28 [1.18–1.40]). Among the 16 studies assessing the complication risk, 14 used HRs and 2 used IRRs. The pooled estimate was (1.27 [1.17–1.38]; Q 13 = 434.5, P < 0.001, I 2 = 99.0%, k = 14) when the analysis was restricted to HR and remained consistent in leave-one-out analysis (1.25–1.31).
Forest plot for risk of any complications in diabetes patients with vs without depression. Studies were arranged in reverse chronological order (first author and their colleagues, year of publication).
Figure 3 Long description
Study A forest plot with a left column listing study names and years, a central plot area with square markers and horizontal confidence-interval lines and a right column labeled “RR (95 percent CI)”. A vertical dashed reference line is drawn at 1. The x-axis is labeled “Estimate (95 percent CI)”, ranging from 0.5 to 3.5 with tick marks at 0.5, 1, 1.5, 2, 2.5, 3 and 3.5. The plotted study estimates are shown as squares, each with a horizontal line for the confidence interval. Most squares are positioned to the right of the reference line at 1. The right column lists the following values: 1.30 (0.97 to 1.74) 1.23 (0.91 to 1.66) 1.23 (0.91 to 1.66) 1.23 (0.91 to 1.66) 1.32 (1.03 to 1.68) 1.23 (0.91 to 1.66) 1.32 (1.03 to 1.68) 1.23 (0.91 to 1.66) 1.39 (1.16 to 1.67) 1.35 (1.16 to 1.57) 1.35 (1.16 to 1.57) 1.35 (1.16 to 1.57) 1.21 (0.91 to 1.59) 1.80 (0.96 to 3.38) 1.20 (0.91 to 1.59) 2.10 (1.49 to 2.97) 1.28 (1.18 to 1.40) A diamond marker near the bottom represents the pooled estimate, aligned with 1.28 (1.18 to 1.40).
Stratified analyses of complication risk by geographic regions, study period, nature of diabetes and complication subtypes are summarized in Table 3. In stratified analyses by types of complications, the depression-diabetes group was associated with a higher rate of macrovascular complications (1.23 [1.15–1.31]; Q 13 = 469.6, P < 0.001, I 2 = 98.2%, k = 14) than the diabetes-only group, but not microvascular complications (1.08 [0.97–1.19]; Q 4 = 222.1, P < 0.001, I 2 = 98.3%, k = 5) (Supplementary Figure S6). Relative to diabetes-only group, depression-diabetes group was associated with a higher rate of metabolic complications (1.63 [1.33–1.99]; Q 2 = 10.1, P = 0.006, I2 = 83.7%, k = 3) and cardiovascular complications (1.20 [1.11–1.29]; Q 8 = 255.5, P < 0.001, I 2 = 97.7%, k = 9), and lower likelihood of retinopathy (0.84 [0.76–0.94]; Q 2 = 4.3, P < 0.001, I 2 = 54.7%, k = 3), although cerebrovascular complications (1.36 [0.99–1.87]; Q 5 = 141.5, P < 0.001, I 2 = 99.6%, k = 6), nephropathy (1.09 [0.93–1.27]; Q 3 = 94.4, P < 0.001, I 2 = 98.4%, k = 4) and peripheral vascular complications (0.97 [0.79–1.18]; Q 3 = 19.7, P < 0.001, I 2 = 86.3%, k = 4) were not associated with significant rate differences (Supplementary Figure S7). In stratified analyses by nature of diabetes, an increased risk of complications was observed in depression patients with incident diabetes (1.32 [1.11–1.58]; Q 5 = 327.2, P < 0.001, I 2 = 99.0%, k = 6) and prevalent diabetes (1.24 [1.15–1.34]; Q 9 = 354.6, P < 0.001, I 2 = 98.7%, k = 10) (Supplementary Figure S8). No significant group differences in complication risk were detected for incident and prevalent diabetes status (between-group P = 0.604). In stratified analyses by geographic regions, an increased risk of complications was observed in depression patients with diabetes in Asia (1.31 [1.15–1.48]; Q 8 = 420.8, P < 0.001, I 2 = 99.5%, k = 9), in Europe (1.29 [1.06–1.56]; Q 3 = 230.6, P < 0.001, I 2 = 98.1%, k = 4) and in North America (1.18 [1.11–1.25]; Q 2 = 6.3, P < 0.001, I 2 = 35.8%, k = 3) (Supplementary Figure S9). No significant group differences in complication risk were detected for geographic regions (between-group P = 0.850). In stratified analyses by study period, a higher risk of complications was observed in studies with study period after 2010 (1.22 [1.13–1.32]; Q 7 = 380.0, P < 0.001, I 2 = 98.2%, k = 8) and before or in 2010 (1.40 [1.16–1.69]; Q 7 = 292.4, P < 0.001, I 2 = 99.6%, k = 8) (Supplementary Figure S10). No significant group differences in complication risk were detected after 2010 and before or in 2010 (between-group P = 0.214). Univariate meta-regression analyses found no significant associations between risk of any complications and study mean age at baseline of both groups (Qm = 2.90, P = 0.09), percentage of male gender of both groups (Qm = 0.79, P = 0.375) and overall follow-up duration (Qm = 0.1, P = 0.722).
Random-effects meta-analytic models showing complication outcomes and stratified analyses
Table 3 Long description
Random-effects meta-analysis results compare complication risks in people with both depression and diabetes versus diabetes alone, reporting relative risk, confidence intervals, heterogeneity, and p values. Overall, any complication is higher across 16 studies, with a relative risk of 1.28 and very high heterogeneity. By region, risks are similar in Asia, Europe, and North America, and the between-region difference is not statistically significant. By study period, risk is higher in studies before or in 2010 than after 2010, but the between-period difference is not statistically significant. By diabetes type, incident and prevalent diabetes show similar elevated risks, with no significant between-group difference. By subtype, macrovascular, metabolic, and cardiovascular complications show statistically significant increases, while microvascular, cerebrovascular, peripheral vascular complications, and nephropathy are not statistically significant. Retinopathy shows a statistically significant lower risk, and several subtype estimates have high heterogeneity, so results may vary notably across studies.
Note: CI: confidence interval; RR: relative risk.
a P value for the comparison between depression-diabetes group versus diabetes-only group.
b P value for between-group comparison.
Discussion
To our knowledge, this is the largest systematic review and meta-analysis to summarize the risk of mortality and complications in patients with depressive disorder and co-occurring diabetes relative to diabetes-only patients, to investigate the associations between depression and a broad range of diabetes-related outcomes and to evaluate the impact of potential moderators on the differential mortality and complication risk using subgroup analyses and meta-regression on an array of study-level characteristics.
The main analyses of mortality outcome indicated that patients with depression and comorbid diabetes exhibited 30% increased risk of all-cause mortality relative to non-depressed counterparts. Our result was consistent with that of Hofmann et al., (Reference Hofmann, Köhler, Leichsenring and Kruse2013), which reported a 49% increased mortality risk in their meta-analysis. Albeit more modest, the all-cause mortality risks across studies were consistently elevated, with robust pooled estimates confirmed by the sensitivity analyses. Our findings also suggested that a greater percentage of males may increase the depression-associated all-cause mortality risk in patients with diabetes, suggesting that male patients could be at greater risk. Notably, such increased mortality risk was consistent across Asia, Europe and North America (i.e., 27%, 28% and 30% increased risk, respectively), in studies conducted before and after 2010 (i.e., 32% and 28% increased risk, respectively), and regardless of incident or prevalent diabetes status, suggesting that the mortality gap was present over distinct regions and have persisted over time. A meta-analysis focusing on schizophrenia-spectrum disorder (SSD) and bipolar disorder (BD) had demonstrated a 77% higher all-cause mortality risk (Ho et al., Reference Ho, Chan, Chiu, Tsang, Chan, Wong, Wong, Pang and Chang2025b) in this respect. As appeared across the diagnoses of depression, SSD and BD, the influence of mental disorders on the clinical outcomes of diabetes might be relatively transdiagnostic, and the differential mortality risk points to the importance of investigating the attenuating or aggravating factors related to diabetes outcomes in depression.
The analyses of complication outcomes indicated that patients with depression and comorbid diabetes exhibited significantly higher risk of any complications relative to non-depressed counterparts, regardless of incident or prevalent diabetes status. Importantly, such increased complication risk was demonstrated in various geographic regions and has persisted over time, suggesting the need for interventions to reduce the health disparity in depression. Analyses of the major specific complications observed higher rates of metabolic and cardiovascular complications, and lower rates of retinopathy in the depression-diabetes group than in the diabetes-only group, albeit comparable rates of nephropathy, peripheral vascular complications and cerebrovascular complications. The markedly elevated risk of metabolic complications in the depression-diabetes group likely reflects acute glycemic fluctuations as characterized by events such as diabetic ketoacidosis and hyperosmolar hyperglycemic state. Although the available data did not allow us to isolate the contribution of each specific event type, it highlights the need for closer monitoring of glycemic instability, early detection of treatment non-adherence and difficulties in diabetes self-management that might lead to rapid metabolic decompensation in patients with depression and diabetes. Conversely, a lower or similar rate of vascular complications could be due to a competing risk model from premature death that precluded the emergence of advanced vascular complications. Alternatively, it might be due to under-recognition of diabetes complications, driven by inadequate understanding at the patient level and by disparities in diabetes treatment and monitoring at the clinician and system levels (Wagner et al., Reference Wagner, Højlund, Fiedorowicz, Nielsen, Østergaard, Høye, Heiberg, Poddighe, Delogu, Holt, Correll, Cortese, Carvalho, Boyer, Dragioti, Du Rietz, Firth, Fusar-Poli, Hartman, Larsson, De Giorgi, Lehto, Lindgren, Manchia, Nordentoft, Skonieczna-żydecka, Veroniki, Marx, Campana, Mortazavi, Hasan, Stubbs, Taipale, Vancampfort, Vieta and Solmi2026). While patients might present with acute-onset and distressing symptoms (e.g., coma and myocardial infarction) warranting clinical attention, suboptimal glycemic control could silently drive any hidden chronic complications (e.g., progressive visual blurring and kidney failure), and ultimately the mortality gap in the long term.
It is acknowledged that the association between depression and excess mortality among patients with co-occurring diabetes is complex (Hackett and Steptoe, Reference Hackett and Steptoe2017). Depressive symptoms, such as reduced motivation and negative cognitions, may affect patients’ self-management of diabetes, leading to delayed help-seeking, treatment non-adherence and unfavourable lifestyle behaviours (Berk et al., Reference Berk, Kohler-forsberg, Turner, Penninx, Wrobel, Firth, Loughman, Reavley, McGrath, Momen, Plana-Ripoll, O’neil, Siskind, Williams, Carvalho, Schmaal, Walker, Dean, Walder, Berk, Dodd, Yung and Marx2023), thereby resulting in poor glycemic control and worse physical outcomes (Lustman et al., Reference Lustman, Anderson, Freedland, de Groot, Carney and Clouse2000; Gonzalez et al., Reference Gonzalez, Peyrot, McCarl, Collins, Serpa, Mimiaga and Safren2008). Patients’ diabetes outcomes might also be adversely affected by healthcare disparities, with lower receipt of equitable diabetes care and suboptimal treatment (Wagner et al., Reference Wagner, Højlund, Fiedorowicz, Nielsen, Østergaard, Høye, Heiberg, Poddighe, Delogu, Holt, Correll, Cortese, Carvalho, Boyer, Dragioti, Du Rietz, Firth, Fusar-Poli, Hartman, Larsson, De Giorgi, Lehto, Lindgren, Manchia, Nordentoft, Skonieczna-żydecka, Veroniki, Marx, Campana, Mortazavi, Hasan, Stubbs, Taipale, Vancampfort, Vieta and Solmi2026). Accumulating evidence has also indicated shared pathophysiological mechanisms underlying both conditions, including hypothalamic-pituitary-adrenal axis hyperactivity, immune-inflammatory dysregulation and overlapping genetic and epigenetic factors (Vreeburg et al., Reference Vreeburg, Hoogendijk, van Pelt, DeRijk, Verhagen, van Dyck, Smit, Zitman and Penninx2009; Stapelberg et al., Reference Stapelberg, Neumann, Shum, McConnell and Hamilton-craig2011). In addition, biological and behavioural factors that are connected with both depression and diabetes, such as obesity and sedentary lifestyles (Tabák et al., Reference Tabák, Akbaraly, Batty and Kivimäki2014), could be implicated in the poorer clinical outcomes. Notably, evidence indicated that successful treatment for depression is able to enhance glycemic control (Croatto et al., Reference Croatto, Vancampfort, Miola, Olivola, Fiedorowicz, Firth, Alexinschi, Gaina, Makkai, Soares, Cavaliere, Vianello, Stubbs, Fusar-Poli, Carvalho, Vieta, Cortese, Shin, Correll and Solmi2023), therefore highlighting that treating mental disorders through effective management can reduce their adverse effects on physical health outcomes (Dragioti et al., Reference Dragioti, Radua, Solmi, Gosling, Oliver, Lascialfari, Ahmed, Cortese, Estradé, Arrondo, Gouva, Fornaro, Batiridou, Dimou, Tsartsalis, Carvalho, Shin, Berk, Stringhini, Correll and Fusar-Poli2023). Optimization of antidepressants in patients with comorbid depression and diabetes may additionally contribute to the reduction of all-cause mortality (Ho et al., Reference Ho, Chan, Lo, Fang, Wong, Lee, Lai, Ng, Wong and Chang2026). However, not all existing treatment strategies for patients with comorbid depression and diabetes are considered effective for glycemic control (van der Feltz-cornelis et al., Reference van der Feltz-cornelis, Allen, Holt, Roberts, Nouwen and Sartorius2021). Identifying effective interventions is important, including integrated and targeted approaches (Petrak et al., Reference Petrak, Baumeister, Skinner, Brown and Holt2015), such as improving diabetes screening and management in psychiatric settings. Improving the provision of lifestyle advice and community support might mitigate the shared risk factors across depression and diabetes that drive the risk of adverse clinical outcomes. Precision psychiatry that aims to provide personalized diabetes care might identify patients with depression who are at risk of developing diabetes, and thus, pre-emptive interventions can be delivered to prevent the onset of diabetes in this at-risk population (Perry et al., Reference Perry, Osimo, Si, Hitchins, Lewis, Laws, Griffin, Khandaker, Murray, Shiers, Chew-graham, Jones, Denniston, Bardus, Jowett, Walsh, Arshad, Formanek, Pillinger, McCutcheon, Holt, Heyse, Rambousek, Whiteley, Upthegrove, Ensor, Vazquez-bourgon, Vandenberghe, Chan, Teasdale, Curtis, Ward, Keinänen, Brodeur, Quadackers, Solmi, Virani and Zavala2026).
Several limitations warrant consideration in interpreting the study results. First, the overall risk estimates should be interpreted with caution. To maximize data inclusion, we pooled ORs, HRs and IRRs into a single relative risk estimate, but these effect measures are not strictly equivalent, and ORs may overestimate relative risk when the outcomes are common. In the current analysis, the pooled risk estimates were dominated by HR, as extracted from the primary studies. The sensitivity analyses restricted to HRs are consistent with the pooled risk estimates and could minimize this concern. Second, significant heterogeneity was found across most meta-analytic estimates, and this heterogeneity could not be entirely accounted for by the stratified analyses based on study-level characteristics. The substantial heterogeneity likely reflects clinical and methodological diversity among the included studies. The presence of high heterogeneity as such could be attributed to different sociocultural contexts and health systems in distinct geographic regions with variable degrees of health literacy, health-seeking behaviour and medication prescribing practices in the diabetes population. Across studies, methodological variations with inconsistencies in the ascertainment of depression (e.g., differences in diagnostic thresholds in different health systems), different degrees of accuracy and representativeness of healthcare databases, differences in confounder adjustment (e.g., varying confounding variables including lifestyle factors, socioeconomic status and physical comorbidities), inconsistencies in the coding of diabetes or diabetes complications and the variations in depression and diabetes management over the study periods covering several decades might also account for the heterogeneity. At these high levels of heterogeneity, the pooled risks should be interpreted cautiously, noting that it represents a descriptive estimate of central tendency within a highly variable distribution of effects, rather than a precise population-level effect size. This indicates the need for future studies to standardize diagnostic and outcome ascertainment. Third, the vast majority of the included studies did not employ formal competing-risk regression methods to account for the competing risk of death when analysing outcomes of diabetes complications. This represents a systematic methodological limitation of the complication analyses, as Cox proportional hazards models used in the primary studies may overestimate complication incidence in the depression-diabetes group by failing to properly handle mortality as a competing event. The observed lower rates of retinopathy and comparable rates of nephropathy and peripheral vascular complications may therefore partly reflect this analytical limitation rather than true clinical differences. Future research should incorporate competing-risk methods to provide more robust estimates of complication risks in this population.
Fourth, there is a scarcity of data for subgroup analyses and meta-regression, and the results should be re-evaluated when more studies are conducted. Fifth, to minimize the confounding effects of other covariates on the associations between depression and diabetes outcomes, the most adjusted estimates were selected. However, such estimates might be overadjusted and no longer represent the real impact of depression on diabetes outcomes. Alternatively, although the effect of depression was isolated by selecting the fully-adjusted estimate, it might still be confounded by unmeasured or residual variables. Sixth, the included studies were geographically concentrated in high-income countries, predominantly the United States, the United Kingdom and South Korea, with the absence of evidence from low- and middle-income countries (LMICs) that bear a significant disease burden of depression and diabetes. Our findings may reflect health systems with relatively advanced infrastructure and greater capacity of diagnostic ascertainment, enabling the identification of depression and diabetes complications. Caution is warranted in generalizing the results to LMIC settings with variable patterns of healthcare services and quality of diabetes care. Future studies should incorporate findings from different geographic regions, particularly LMICs, to determine the extent to which these associations hold across regions. Seventh, the findings of several specific complications, including retinopathy, nephropathy, peripheral vascular complications and metabolic complications, were based only on a few studies and should be interpreted with caution. The limited number of contributing studies precludes robust conclusions about these specific complications and requires replications in larger pools of primary studies. Eighth, most studies focused on type 2 diabetes, with only two studies examining type 1 diabetes. Many studies included both type 1 and type 2 diabetes and did not delineate their analyses by diabetes subtypes, but their sample is likely to be predominantly type 2 diabetes. Due to the limited data on type 1 diabetes, we could not further evaluate the differential effect of depression on the physical health outcomes in relation to diabetes subtypes. Future studies should separately investigate type 1 diabetes, given its distinct pathophysiological mechanisms. Ninth, similar to many meta-analyses, publication bias was found, indicating the presence of unpublished literature. Notwithstanding, this meta-analysis confirms the adverse influence of depression on mortality and complications in patients with co-existing diabetes and provides preliminary targets to improve the clinical outcome in patients with depression in the context of diabetes.
In conclusion, our study demonstrated that patients with depression and co-occurring diabetes were associated with an elevated overall mortality risk compared to their non-depressed counterparts. Such an increased all-cause mortality risk was present across geographical regions and has persisted over time. Depression patients with comorbid diabetes were also associated with a higher risk of metabolic and cardiovascular complications, implying an overall poorer glycemic control that might eventually drive their earlier death. This study provided a comprehensive evidence synthesis that indicates an unmet treatment need for patients with depression to rectify their poorer diabetes-related outcomes. A multilevel interventional framework, including potential development of depression-specific risk prediction models for major diabetes complications, is needed for individualized risk estimation of diabetes-related outcomes, with consequent early interventions to minimize the avoidable physical morbidity and premature mortality in this vulnerable population.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S2045796026100882.
Availability of data and materials
The data that support the findings of this study are available from the corresponding author upon reasonable request.
Acknowledgements
The authors would like to thank Gary Cheuk Yin Lor and Sherry Ng for their kind assistance in the development of the search strategy.
Author contributions
M.T.H.H. conceptualized the study and registered the study protocol, with critical input from J.K.N.C. and W.C.C. on study protocol design. M.T.H.H. and W.C.Y.C. contributed to the screening of literature and quality assessment. L.L.W.T., J.K.N.C. and W.C.C. oversaw the data analysis and interpreted the results. M.T.H.H. contributed to the data analysis, interpretation of the results, wrote the first draft of manuscript and revised the manuscript. W.C.C., J.K.N.C. and H.K.Y.L. contributed substantially to the revision of manuscript drafts. W.C.C. and M.T.H.H. finalized the manuscript. All authors provided critical feedback to the manuscript drafts, reviewed and approved the final manuscript.
Financial support
This research received no specific grant from any funding agency, commercial or not-for-profit sectors.
Competing interests
The authors declared no conflicts of interest in relation to the subject of this study.
Ethical standards
As this study was based on existing literature, no ethical approval or patient consent was required.
Open access
