Association of depression with glycaemic control in people living with diabetes in low- and middle-income countries: a systematic review and meta-analysis

Abstract

Background

Depression has been found to be associated with poor diabetes control, which contributes to diabetes complications. However, the association between depression and glycaemic control remains understudied in low- and middle-income countries (LMICs) where the greatest burden of uncontrolled diabetes and diabetes complications exists. This meta-analysis examined the association of depression with glycaemic control in adults with diabetes mellitus in LMICs.

Methods

We performed comprehensive searches in PubMed-Medline, Scopus, Web of Science and Cumulative Index to Nursing and Allied Health Literature (CINAHL) databases to identify studies that examined the association of depression with glycaemic control. Study quality was assessed using the Newcastle-Ottawa Scale. Pooled effect estimates were expressed as ORs and mean differences (MDs) using random effects meta-analysis. Heterogeneity of effects was tested using Cochran’s Q test.

Results

A total of 39 studies comprising 22 456 adults with diabetes, of whom 21% had depression, were included in the meta-analysis. Depression was associated with poor glycaemic control (OR: 2.01, 95% CIs 1.41 to 2.86; I²: 90.8%; p<0.001; AOR: 1.52; 1.20 to 1.92; I²: 93%; p<0.001; MD: 0.56; 0.27 to 0.84; I²: 82%; p<0.001), with difference in effect sizes by depression diagnostic criteria (p<0.001). Age, diabetes duration, marital status and publication year had no effect on the association (all p≥0.096); while inconsistent effects on the association were observed for body mass index, male gender, sample size and region where studies were conducted. Observed publication bias (all p≤0.007 for the Egger’s test) was likely spurious.

Conclusion

This meta-analysis found a positive association of depression with poor glycaemic control in adults with diabetes in LMICs. The findings emphasise the importance of incorporating mental healthcare in diabetes management in low-resource settings.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Research in high-income countries has shown a bidirectional relationship between depression and poor glycaemic control, but this association is understudied in low- and middle-income countries (LMICs).

WHAT THIS STUDY ADDS

• A meta-analysis of 39 studies from 24 LMICs (including 22 456 individuals with type 2 diabetes, 21% of whom had depression) revealed a positive association between depression and poor glycaemic control.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• The findings underscore a lack of longitudinal studies investigating this association in LMIC settings.

• Incorporating the screening and management of depression into diabetes care is essential to improve glycaemic outcomes in LMICs.

Introduction

Type 2 diabetes mellitus (hereafter referred to as diabetes) is a leading cause of morbidity and mortality worldwide.^{1 2} According to the 2019 Global Burden of Disease study, approximately 437.9 million people were living with diabetes worldwide, and 1.5 million deaths and 66.3 million disability-adjusted life years (DALYs) lost were due to the illness.² In low- and middle-income countries (LMICs), the prevalence of diabetes among individuals aged 20–79 years ranged from 5% to 12%, varying by region.³ Globally, substantial increases were observed in age-standardised prevalence, death and DALY rates of 49%, 10.8% and 27.6%, respectively, between 1990 and 2019, with LMICs bearing the greatest burden.²

Comorbidities with diabetes are common and, together with other cardiometabolic diseases of hypertension, dyslipidaemia, obesity, etc, include poor mental health. Depression, a common mental disorder, affects 20%–30% of people with diabetes globally^{4 5} and 25%–45% in LMICs.^{6 7} These rates are significantly higher than the 8%–11% prevalence reported in general populations,^{4 5 8} and the 11% demonstrated among >50 years old in LMICs.⁹ Moreover, several studies have suggested that people with diabetes and comorbid depression often exhibit poorer glycaemic control than those without depression.⁴ This is of concern, seeing that diabetes morbidity and mortality are attributable to suboptimal glycaemic control.10,12 Other key risk factors contributing to poor glycaemic control include younger age, lower education level, longer duration of diabetes, etc.13,16

Further, poor glycaemic control in diabetes may lead to the development of depression compared with those with optimal control, which in turn worsens glycaemic control.¹⁵ A recent meta-analysis of 11 studies from high-income countries (HICs) suggested a bidirectional association between depression and glycated haemoglobin (HbA1c) levels, the measure of glycaemic control. Higher baseline depressive symptoms correlated with subsequent increases in HbA₁c, and higher baseline HbA₁c levels were associated with an increased risk of developing depression.¹⁶

Despite these significant health threats, the relationship between depression and diabetes, particularly glycaemic control, remains understudied in LMICs where socioeconomic status, traditions and cultures of people with diabetes differ from their counterparts in HICs.⁸ Stigma surrounding both mental health and diabetes is also often more pronounced in LMICs.¹⁷ Additionally, it is well documented that people with diabetes in LMICs experience a greater burden of complications and mortality attributable to poorer glycaemic control compared with their counterparts in HICs.¹⁸ These influences underscore the need to investigate the association between depression and glycaemic control among people with diabetes in LMICs.⁶ Such an understanding will contribute to the existing body of evidence and inform prevention and treatment strategies for individuals with diabetes. Therefore, the aim of the present systematic review and meta-analysis is to examine the association between depression and glycaemic control in adults with diabetes living in LMICs.

Methods

Search strategies

This review was performed in accordance with the Preferred Reporting Items for Systematic Review and Meta-Analyses guidelines. The review was registered on the International Prospective Register of Systematic Reviews (registration number: CRD42024507682).¹⁹ A comprehensive search was conducted in PubMed-Medline, Scopus, Web of Science and Cumulative Index to Nursing and Allied Health Literature (CINAHL) databases using keywords relating to glycaemic control, HbA₁c, blood glucose, depression and diabetes. The search was filtered for research articles published till December 2024 in English and French languages. The initial search was up to 31 January 2024 and updated to 6 December 2024, with the restriction to articles published in English and French languages.

Selection criteria

This review included cross-sectional, case-control and cohort studies in adults (aged ≥18 years or as defined by the study author) living with diabetes in LMICs based on the World Bank criteria.²⁰ Studies that were eligible for this review reported depression as a categorical variable (depression: yes/no) and HbA₁c as a continuous variable or a binary variable (glycaemic control: yes/no) or an estimate of the association between them (OR and 95% CIs). Studies that included children, gestational diabetes or people living in HICs were excluded. Two reviewers (KAN and DS) independently screened the titles, abstracts and full texts using predefined inclusion and exclusion criteria. Disagreement was resolved by consensus or by consulting a third reviewer (APK or NP).

Data extraction

Data extraction was done in parallel by KAN and DS using the data extraction form designed for this review. The form was piloted with five studies to minimise errors before being used in the main review. The following information was extracted from each study: (1) author names and year of publication; (2) study characteristics (study design, setting and analytic approach of the association of depression with glycaemic control); (3) sample characteristics such as sample size, age, sex, marital status, lifestyle behaviours (smoking and alcohol use), comorbidity (hypertension, obesity and other chronic diseases), diabetes (diagnosis, treatment, duration and complications), country; (4) depression (diagnosis, diagnostic scales/tools used and treatment) and (5) glycaemic control outcome: effect sizes included mean and SD of HbA₁c (depression vs no depression), number of depression cases (glycaemia uncontrolled vs glycaemia controlled as defined by the included studies) and adjusted OR (95% CI) for the association of depression with glycaemic control (AOR from the final model was chosen).

Quality assessment

The methodological quality of the included studies was assessed using the Newcastle-Ottawa Assessment Scale.^{21 22} Each study was assessed by three domains, namely selection, comparability and outcomes. These were categorised as good (≥7 stars), moderate or fair (4–6 stars) or low (0–3 stars) quality based on the total number of stars allocated to each of the three domains. One reviewer (KAN) conducted the quality assessment, which was confirmed by the second reviewer (DS), who assessed 30% of the included studies.

Statistical analysis

All statistical analyses were done in R statistical software (V.4.3.3) and the package ‘metafor’. Exploratory meta-analyses were conducted to understand the magnitude and direction of the effect estimate; three sets of meta-analyses were performed. First, among studies that reported the binary outcome of glycaemic control (poor glycaemic control: yes/no), we calculated OR with 95% CI for each included study, then pooled the data using Mantel-Haenszel random-effects meta-analysis. Second, among studies that reported AOR as an effect measure for the association of depression with poor glycaemic control, we pooled all AORs using Mantel-Haenszel random-effects meta-analysis. Third, among studies that compared mean HbA₁c levels in diabetes individuals with and without depression, we calculated the mean difference (MD) with 95% CI for each included study and then pooled the estimates using a random effects model. The Cochrane Q test and inconsistency index (I²) were used to assess heterogeneity characterised as low (I²<25%), moderate (25%<I²>50%) and high (I²>70%) heterogeneity. We conducted a subgroup analysis to explore the sources of heterogeneity. The subgroups of interest were defined based on differences in the major characteristics of the included studies. For example, these included categorical characteristics (setting, region and diagnostic criteria) and the use of median values of summary estimates for continuous characteristics (age, diabetes duration, body mass index (BMI) level, percentage of male, percentage of married, sample size and year of publication). The role of the latter as potential moderators was further assessed using univariable meta-regression models; multiple meta-regression models could not be done due to the insufficient number of studies. Additionally, we examined the robustness of the pooled estimates using leave-one-out analysis to investigate whether any particular study influenced the pooled estimates. The risk of publication bias was explored using egger’s test and by visual inspection of funnel plots, with a p value <0.05 indicating a significant asymmetry of the funnel plot and evidence of publication bias. The Duval and Tweedie trim-and-fill was used to adjust estimates for the effects of publication bias. The certainty of the evidence for the association of depression with glycaemic control was rated using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach.²³

RESULTS

In all, 37 articles met the selection criteria and were included in the review. Two articles that reported examining two different subgroups counted as two separate studies, translating into a total of 39 studies included in the meta-analyses. The details of the study selection process are shown in figure 1.

Figure 1. PRISMA flow diagram of the study selection procedure. LMICs, low- and middle-income countries; PRISMA, Preferred Reporting Items for Systematic Review and Meta-Analyses. From: Page et al⁸⁸ 2021. For more information, visit: http://www.prisma-statement.org/.

The characteristics of the included studies are summarised in table 1. Overall, the included studies consisted of 22 456 individuals with diabetes, 20.8% of whom had depression (n=4667) defined by different criteria. The sample size ranged from 70 to 2538 diabetes participants with a median age of 54 years (46–65 years) across 26 studies and a median diabetes duration of 10 years (4.3–15 years) across 17 studies. The median BMI was 27.6 kg/m² (23.9–34) across 17 studies, and the median HbA₁c level was 7.8% (6.1–9.9) across 23 studies.

Table 1. Characteristics of the included studies.

Study	Country	Year of data collection	Study design	Setting	Sample size	Age (mean (SD)) (year)	Depression measurement	Depression %	HbA1c (mean (SD))/median (IQR) %	Outcome-diabetes control %
Ali et al24 2023	Ethiopia	2022	Cross-sectional	Hospital-based	263	50.2 (14.8)	PHQ-9	47	8.8 (2.7)	67
Akpalu et al25 2018	Ghana	NR	Cross-sectional	Hospital-based	400	52.7 (8.7)	PHQ-9	31	9.9 (2.9)	NR
Mansori et al51 2019	Iran	2016	Cross-sectional	Hospital-based	514	>30	BDI-II	46	NR	74
Al-Ozairi et al52 2023	Kuwait	2018–2019	Cross-sectional	Hospital-based	446	55.5 (10.5)	PHQ-9	34	8.1 (1.7)	NR
Abdulah et al 2020	Malaysia	NR	Cross-sectional	Hospital-based	300	63 (16)	BDI-II	20	7.6 (2.7)	69.3
Vuuren et al 2019	South Africa	2017	Cross-sectional	Hospital-based	176	54.4 (14.8)	PHQ-9	47	76.9 (25.3)	NR
Sun et al38 2016	China	2013	Cross-sectional	PHC	893	63.9 (10.2)	Zung	44	NR	76
Ahmed et al26 2022	Egypt	2021	Cross-sectional	PHC	403	46 (11.5)	PHQ-9	09	7.8 (0.7)	93
Crispin-Trebejo et al56 2015	Peru	2014	Cross-sectional	Hospital-based	277	59 (4.8)	PHQ-9	11	NR	75
Bessel et al57 2016	Brazil	NR	Cross-sectional	Community-based	1096	NR	CIS-R	30	7.27 (1.78)	NR
Bessel et al57 2016	Brazil	NR	Cross-sectional	Community-based	1336	NR	CIS-R	26	6.07 (1.26)	NR
Woon et al 2020	Malaysia	2018	Cross-sectional	Hospital-based	176	61.5 (52–69)	BDI-II	07	8.5 (7.6–10)	50
Abdulai et al32 2020	China	2015–2017	Cross-sectional	Community-based	1173	55.4 (12.3)	PHQ-2	08	NR	25
Alzughbi et al53 2020	Saudi Arabia	NR	Cross-sectional	PHC	300	52.7 (11.4)	PHQ-9	20	8.9 (1.8)	74
Azeze et al27 2020	Ethiopia	2019	Cross-sectional	Hospital-based	410	47.4 (9.6)	PHQ-9	29	NR	37
Bahety et al39 2017	India	2017	Case-control	Hospital-based	100	56.09 (5.92)	PHQ-9	63	NR	NR
Bawadi et al54 2021	Qatar	2012–2018	Cross-sectional	Community-based	2448	NR	PHQ-9	15	7.4 (1.8)	51
Edah et al28 2020	Nigeria	2018	Cross-sectional	Hospital-based	310	54 (12)	MINI	11	NR	52
Fung et al33 2018	China	2013	Cross-sectional	Hospital-based	325	>65	GDS	13	7.2 (1.1)	52
Hargittay et al49 2022	Hungary	2018–2020	Cross-sectional	PHC	338	64 (11.5)	BDI-II	14	7.23 (1.26)	NR
Le et al48 2022	Vietnam	NR	Cross-sectional	Hospital-based	231	52 (45–57)	PHQ-9	17	7.6 (6.6–9.3)	55
Liu et al46 2023	Taiwan	2009–2013	Cross-sectional	Community-based	1527	>55	CES-D	07	NR	81
Mathew et al40 2012	India	NR	Cross-sectional	Hospital-based	80	54.5 (10.2)	MDI	39	9.5 (1.8)	95
Otieno et al29 2017	Kenya	NR	Cross-sectional	Hospital-based	220	57 (8.6)	PHQ-9	32	NR	70
Papelbaum et al58 2011	Brazil	NR	Cross-sectional	Hospital-based	70	30–65	SCID	19	7.72 (1.91)	NR
Rahman et al47 2020	Bangladesh	2018	Cross-sectional	Hospital-based	500	21–85	DASS-Bangla	43	NR	72
Sharif et al44 2019	Pakistan	NR	Cross-sectional	Hospital-based	100	58.3 (12.4)	PHQ-9	40	8.1 (1.88)
Zuberi et al45 2011	Pakistan	NR	Cross-sectional	Diabetic clinic	286	52	HADS	50	NR	72
Shirey et al30 2015	Kenya	NR	Retrospective chart review	Diabetic clinic	253	57.6	PHQ-2	21	NR	NR
Singh et al41 2014	India	NR	Case-control	Hospital-based	109	NR	CESD	42	8.16 (1.45)	NR
Stankovic et al50 2011	Serbia	NR	Cross-sectional	Diabetic clinic	90	NR	BDI-II	51	9.11 (1.65)	NR
Sweileh et al55 2014	Palestine	NR	Cross-sectional	PHC	294	NR	BDI-II	41	NR	82
Tellez-Zenteno and Cardiel59 2002	Mexico	NR	Cross-sectional	Diabetic clinic	189	61.7 (12.5)	BDI-II	13	NR	NR
Wang et al34 2018	China		Cross- sectional	Hospital-based	808	NR	PHQ-9	12	NR	28
Zhang et al35 2013	China	2010–2011	Cross-sectional	Hospital-based	586	55.1 (9.5)	PHQ-9	09	7.50 (1.40)	60
Zhang et al36 2015	China	2010–2011	Cross-sectional	Hospital; diabetic clinic	2538	56.4 (10.5)	PHQ-9	06	7.7 (2.0)	NR
Zhang et al37 2015	China	2010–2011	Cross-sectional	Hospital; diabetic clinics	545	55.4 (9.9)	PHQ-9	18	7.5 (1.4)	60
Mahmoud 2024	Saudi Arabia	2021–2022	Cross-sectional	Hospital-based	846	40.4	PHQ-9	35	NR	54.6
Mahmoud 2024	Egypt	2021–2022	Cross-sectional	Hospital-based	1500	54.8	PHQ-9	18	NR	26.1

Hospital Anxiety Depression Scale (HADS)-adapted version for Pakistanis.

BDI-II, Beck Depression Inventory-II; CES‑D, Center for Epidemiologic studies Depression; CIS-R, Clinical Interview Schedule-Revised; GDS-15, Geriatric Depression Scale 15; MDI, Major Depression Inventory; MINI, Mini International Neuropsychiatric Interview; modified DASS, Depression, Anxiety and Stress Scale: modified for Bangladeshis; NR, not report; PHC, primary healthcare clinic; PHQ-2, Patient Health Questionnaire 2; PHQ-9, Patient Health Questionnaire 9; SCID-P, Structured Clinical Interview for Depression, patient edition; Zung, Zung Self-Rating Depression Scale.

Most of the included studies (n=35) were cross-sectional in design; two were prospective cohorts, and a single study each was a retrospective chart review and a case-control study (table 1). Of the included studies, five were community-based while the rest were hospital- or clinic-based. Nine studies were from Africa,24,30 including one from South Africa;³¹ 17 studies were from Asia (seven from China,32,38 three from India,39,41 two from Malaysia,^{42 43} two from Pakistan^{44 45} and a single study each from Taiwan,⁴⁶ Bangladesh⁴⁷ and Vietnam⁴⁸); two studies were from Eastern Europe (Bulgaria and Serbia);^{49 50} six studies were from the Middle East region51,55 and five studies were from South American countries.56,59

Depression was diagnosed using variable criteria, with the Patient Health Questionnaire (PHQ-9) used in half of the included studies (20 studies) followed by the Beck Depression Inventory (BDI) (six studies),^{42 43 49 51 55 59} Center for Epidemiologic Studies Depression (CES-D, two studies),^{41 46} Clinical Interview Schedule-Revised (two studies)⁵⁷ and PHQ-2 (two studies).^{30 32} Other screening tools for depression included the Mini International Neuropsychiatric Interview (MINI),²⁸ Geriatric Depression Scale,³³ Major Depression Inventory⁴⁰ and Structured Clinical Interview for Depression, patient edition (SCID-P),⁵⁸ which were used in a single study each and grouped together under ‘mixed criteria’. Three studies used locally adapted questionnaires from the Zung Self-Rating Depression Scale,³⁸ Depression, Anxiety and Stress Scale (DASS-21)-modified Bangla questionnaire⁴⁷ and Hospital Anxiety Depression Scale (Pakistan)⁴⁵ and were categorised as ‘local criteria’.

Glycaemic control outcomes included the binary outcome of glycaemic control, which was reported in 21 studies (table 1); AOR of poor glycaemic control was analysed in 20 studies (online supplemental table S1), and mean HbA₁c values were presented in 16 studies (table 1).

Quality assessment of the included studies

All the included studies were of a high or moderate quality based on the Newcastle-Ottawa Scale. 21 studies (54%) were classified as good quality and the rest (18 studies) as moderate quality. No study was rated as low quality (table 2).

Table 2. Quality assessment of the included studies using the Newcastle-Ottawa Quality Assessment Scale.

Study	Selection	Comparability	Outcome/Exposure	Scores	Overall rating
Ali et al24 2023	3	1	3	7	Good
Akpalu et al25 2018	4	1	3	8	Good
Mansori et al51 2019	4	1	3	8	Good
Al-Ozairi et al52 2023	4	1	3	8	Good
Abdullah et al43 2020	2	1	3	6	Fair
Vuuren et al32 2019	2	0	2	4	Fair
Sun et al38 2016	4	0	2	6	Fair
Ahmed et al26 2022	4	1	1	6	Fair
Crispin-Trebejo et al56 2015	3	1	3	7	Good
Bessel et al57 2016	4	1	1	6	Fair
Bessel et al57 2016	4	1	1	6	Fair
Woon et al44 2021	2	1	3	6	Fair
Abdulai et al32 2020	4	1	3	8	Good
Alzughbi et al53 2020	4	0	3	7	Good
Azeze et al27 2020	4	1	3	8	Good
Bahety et al39 2017	3	1	2	6	Fair
Bawadi et al54 2021	3	1	3	7	Good
Edah et al28 2020	1	1	3	6	Fair
Fung et al33 2018	2	1	3	6	Fair
Hargittay et al49 2022	3	1	3	7	Good
Le et al48 2022	2	0	3	5	Fair
Liu et al46 2023	2	0	2	4	Fair
Mathew et al40 2012	2	0	3	5	Fair
Otieno et al29 2017	3	0	2	5	Fair
Papelbaum et al58 2011	2	0	2	4	Fair
Rahman et al47 2020	4	1	3	8	Good
Sharif et al44 2019	3	1	3	7	Good
Zuberi et al45 2011	3	1	3	7	Good
Shirey et al30 2015	4	0	2	6	Fair
Singh et al41 2014	4	1	2	7	Good
Stankovic et al50 2011	1	0	3	4	Fair
Sweileh et al55 2014	4	0	3	7	Good
Tellez-Zenteno and Cardiel59 2002	2	0	3	5	Fair
Wang et al34 2018	5	0	3	8	Good
Zhang et al35 2013	5	1	3	9	Good
Zhang et al36 2015	5	1	3	9	Good
Zhang et al37 2015	5	1	2	8	Good
Mahmoud et al89 2024	4	0	3	7	Good
Mahmoud et al89 2024	4	0	3	7	Good

Association of depression with poor glycaemic control

Depression was associated with higher odds of poor glycaemic control with significant heterogeneity (OR: 2.01, 95% CIs 1.41 to 2.86; I²: 90.8%; p-het <0.001) (figure 2; online supplemental table S2). The OR effect sizes varied by criteria used to define depression ranging from 0.55 (95% CI 0.35 to 0.87) in a single study that applied CES-D to 3.02 (95% CI 1.50 to 6.09) in three studies that used locally adapted criteria. The leave-one-out analysis indicated that no individual study had a significant influence on the overall estimates, confirming the robustness of the findings (online supplemental figure S1).

Figure 2. Forest plot for OR overall based on studies reporting the binary outcome of glycaemic control.

When considering OR effects drawn from adjusted models of the included studies, effects were slightly attenuated but remained significant (AOR: 1.52; 1.20 to 1.92; I²: 93%; p-het <0.001) (figure 3; Online supplemental table S2). The AOR effect sizes ranged between 0.47 (95% CI 0.22 to 1.00) and 1.84 (95% CI 1.31 to 2.58) in a single study that applied MINI criteria and two studies that used PHQ-9, respectively. The leave-one-out analysis indicated that no individual study had a significant influence on the overall effect estimates (online supplemental figure S2).

Figure 3. Forest plot for OR overall based on studies reporting adjusted OR of depression with poor glycaemic control.

When considering MD effects drawn from the differences in mean HbA₁c between diabetes individuals with and without depression, depression was associated with higher mean HbA₁c (MD: 0.56; 0.27 to 0.84; I²: 82%; p-het <0.001) (figure 4; online supplemental table S2), suggesting poorer glycaemic control with depression. Again, the effect sizes varied by the criteria used to diagnose depression, ranging from 1.60 (1.07 to 2.13) for CES-D criteria in one study to 0.30 (−0.63 to 1.23) for PHQ-2 criteria in another study. The leave-one-out analysis indicated that no individual study had a significant influence on the overall MD (online supplemental figure S3).

Figure 4. Forest plot for glycated haemoglobin (HbA₁c) mean difference between depression and no-depression groups.

Sources of heterogeneity assessment

Subgroup analyses yielded inconsistent results across the effect measures as shown in online supplemental table S2. The estimated effects of depression on poor glycaemic control appeared stronger in the subgroup having BMI <28 kg/m² versus BMI ≥28 kg/m² (AOR: 1.70; 1.24 to 2.32; I²: 0%; p-het=0.214 vs AOR: 1.10; 0.91 to 1.33; I²: 0%; p-het=0.372; p-subgroup=0.021). The impact of BMI was not evident in the effects measured with OR or MD.

Similarly, the effects of depression on glycaemic control were greater in studies that had sample sizes <325 participants (MD: 0.86; 0.43 to 1.29; I²: 64%; p-het=0.004 vs MD: 0.26; 0.01 to 0.52; I²: 79%; p-het <0.001; p-subgroup=0.019); in hospital-based settings (OR: 2.64; 1.73 to 4.02; I²: 85.4%; p-het <0.001 vs OR: 0.79; 0.51 to 1.21; I²: 70%; p-het=0.034; p-subgroup=0.001) and in studies using PHQ-9 in Africa (MD: 1.03; 0.48 to 1.57; I²: 00%; p-het=0.327; p-subgroup=0.019).

The effect sizes were not different in terms of age, diabetes duration, marital status and publication year. These were observed consistently across the OR, AOR and MD effect measures and depression diagnostic criteria (all p-subgroups ≥0.096).

Univariable meta-regression models revealed that age, diabetes duration and BMI significantly affected the overall AOR effects but not the overall OR nor overall MD effects. Percentage of married, sample size and year of publication significantly affected the overall MD but not the overall OR nor the overall AOR (onlinesupplemental figure S4S6).

Certainty of evidence

The certainty of the pooled effect estimates in this meta-analysis was rated as moderate. According to the GRADE approach, the certainty started as very low due to the inclusion of observational studies and the presence of heterogeneity. However, it upgraded to moderate based on the moderate-to-high methodological quality of the included studies and the consistency and robustness of the pooled effect estimates.

Publication bias

There was some evidence of publication bias among studies reporting AOR and mean HbA₁c (all p≤0.007 for the Egger’s test) (figure 5), and among studies using PHQ-9 (p≤0.018) and BDI criteria (p=0.007) overall (online supplemental table S2). There was also evidence of bias in studies that had sample sizes larger than 325 participants (p=0.014), including participants older than 54 years (all p≤0.034), had more than 49% men in the sample (p≤0.017), were published before 2018 (p=0.002) and were conducted in Asia (all p<0.001) and South America (p=0.032). Results of the Trim and Fill analysis are shown in online supplemental table S3. Overall, the effect estimates from imputed studies yielded massive reductions in HbA₁c levels which were favourable for reaching glycaemic control (onlinesupplemental table S3 figure S7). The effect sizes of imputed studies appeared to be implausible, indicating that publication bias is unlikely.

Figure 5. Funnel plots for publication bias.

Discussion

To our knowledge, this is the first systematic review and meta-analysis to report on the association of depression with glycaemic control in people with diabetes in LMICs. The key findings of the current review include the positive association of depression with poor glycaemic control in diabetes in LMICs. The significant associations were demonstrated in both the unadjusted and adjusted models. The substantial heterogeneities were not fully explained by major study-level characteristics, while suggestions of publication bias were likely spurious findings.

Our findings are consistent with earlier meta-analyses that included studies mostly from HICs.^{60 61} The meta-analysis by Lustman and colleagues included 24 cross-sectional studies and both type 1 and type 2 diabetes (n=2817 subjects) and demonstrated that depression was significantly associated with hyperglycaemia.⁶⁰ Another meta-analysis of 34 studies demonstrated that HbA₁c levels were higher in diabetes with depression compared with without depression. Further, HbA₁c levels remained higher in people using antidiabetes drugs, with shorter duration of diabetes and who had diabetes complications;⁶¹ this supports the findings in the present meta-analysis.

One suggested pathway linking depression with poor diabetes control is through self-care impairment.⁶² A review by Gonzalez et al demonstrated that people with diabetes and depression might encounter challenges in their self-care such as missing medical appointments and poor adherence to diabetes treatment, lifestyle advice, glucose monitoring and foot care.⁶² This non-adherence to self-care was associated with poor glycaemic control⁶⁰ and diabetes complications in people with diabetes and depression.^{63 64} On the other hand, uncontrolled diabetes could be a stressor triggering depression which in turn worsens glycaemic control.¹⁶ The relationship between depression and diabetes control, supported by the current literature, is likely bidirectional, with one condition exacerbating the other.¹⁶ Further research is required to elucidate this association and to ascertain any potential biological mechanisms linking depression with poor diabetes control and complications.

In this review, the effects of depression on glycaemic control varied by the criteria used to diagnose depression. Various cut-off points for the same depression diagnostic criteria could partly explain the different effects among studies using the same criteria. Although optimal cut-off points have not been established for various populations, all the criteria used have been validated in different populations.65,69 Several studies have suggested that the optimal cut-off points for the depression screening tools differed from those recommended for certain populations.^{66 70} For example, the cut-off point of 16 has been recommended for the CES-D scale, but some studies reported much higher cut-off points (≥20, 22 or 25) to identify clinical depression in different South African or elderly populations.^{66 67} These differences, explained by the authors, might be attributed to the variation in the perception, experience and idioms about depression among different populations.^{66 67} Therefore, there is a need to identify the optimal cut-off point for specific depression screening tools in different populations.

The effects of characteristics such as BMI, gender, sample size, setting or region could not be fully justified in this meta-analysis due to the insufficient number of studies for meaningful subgroups and meta-regression analyses. However, meta-analyses from HICs suggested that comorbid depression and diabetes were more common with older age and obesity and in women.^{7 71}

The modest number of studies, mostly cross-sectional in design, included in the current review (39 studies) underscores the need for more research in LMICs where the burden of comorbid depression and diabetes is high and compounded by large proportions of undiagnosed and undertreated cases.68 72,74 Longitudinal studies with larger sample sizes are required for a better understanding of the complex relationship between depression and diabetes care outcomes in LMICs. Such studies would allow for more robust examinations of causal relationships, possible mechanisms apart from behavioural influences linking depression with poor glycaemic control and the bidirectional impact of depression and diabetes over time.

However, in the present meta-analysis, the significant associations observed in both the unadjusted and adjusted models indicate that the effect of depression on glycaemic control was largely independent of potential confounding factors (adjusted in the included studies), such as age, gender, BMI, socioeconomic status, diabetes duration, comorbidities, etc, as well as independent of individual study influences. This robust finding underscores the need for early screening and timely treatment of depression comorbidity in people with diabetes in LMICs. It is important for optimising glycaemic control as well as for appropriately treating comorbid depression in diabetes management.^{2 18} The findings have substantial implications for people with diabetes, healthcare providers and healthcare systems in low-resource settings. For people with diabetes, understanding the connection between mental health and glycaemic control may encourage them to prioritise their mental well-being alongside their physical health. This knowledge could motivate early recognition of depressive symptoms and prompt individuals to seek timely help, potentially mitigating the impact of depression on diabetes management and overall quality of life. For healthcare providers, the recognition of this comorbidity may lead to routine mental health screening and appropriate treatment as part of diabetes care. At the healthcare system level, these findings highlight the need to allocate resources for the integrated care of diabetes and mental health. Emerging evidence from studies in the USA and Europe reports that psychosocial interventions, especially cognitive-behavioural therapy and antidepressant medications, are associated with achieving glycaemic control in the short term.75,77 Holistic interventions incorporating mental health into diabetes care are needed to enhance disease management and improve quality of life.

The present review could not determine the influence of contextual characteristics (social, economic, traditional and healthcare-related factors) on the association of depression with glycaemic control in different LMIC settings. However, existing literature suggests a link between poverty and depression.^{78 79} A systematic review of 115 studies in LMICs found that education, food insecurity, housing, social class and financial stress were significantly associated with common mental disorders, including depression, anxiety and somatoform disorders.⁷⁸ Similarly, a study of 4393 adults with non-communicable diseases (diabetes, hypertension or chronic respiratory disease) across 37 primary care clinics in two health districts of the Western Cape, South Africa, identified a bidirectional relationship between socioeconomic disadvantage and depression.⁷⁹

Notably, LMIC health services were characterised by limited resources, overcrowding and long waiting times, which are substantial barriers to accessing healthcare.^{80 81} There are limited healthcare services for chronic diseases, including depression and diabetes. These factors, together with barriers faced by patients such as a lack of finances and transportation difficulties, likely hinder the utilisation of health services by people with diabetes and may exacerbate the challenges of managing comorbid depression and diabetes.^{82 83} The stigma associated with mental health issues in LMICs further compounds these difficulties, often discouraging individuals from seeking care and so delaying their diagnoses and treatment.^{84 85} Comprehensively, any intervention strategy on comorbid depression and diabetes in LMICs should account for contextual challenges by integrating mental health support into chronic disease management programmes, fostering community awareness, and addressing stigma at the societal level.

Strengths and limitations

The use of a review protocol with comprehensive search strategies, including searches of multiple databases, and the consideration of various types of glycaemic control outcomes (binary, AOR and mean HbA₁c) to pool the effects of depression with glycaemic control are strengths of this study. The use of leave-one-out sensitivity analysis confirmed the robustness of the pooled effects estimated, and the overall quality of the included studies was high to moderate. These factors increase the level of certainty of the pooled evidence and are other strengths of this meta-analysis.

Limiting the selection to articles published in English and French might have omitted studies published in other languages. The inclusion of cross-sectional studies prevented the inference of direction and causality. The effect measures are derived from various diagnostic criteria of depression, but likely revolving around the PHQ-9 criteria. The relatively small number of studies included for each type of glycaemic control outcome limited the statistical power of the subgroup analyses and meta-regression models. The results of this meta-analysis should be interpreted in the context of heterogeneity and some evidence of publication bias.

Meta-analysis is intended to synthesise the findings across diverse contexts, and as such, some degree of heterogeneity is expected and acceptable. High levels of heterogeneity, which are potentially attributable to major study-level characteristics such as age, gender, BMI levels, diabetes duration, sample sizes, setting, region and diagnostic criteria, as well as unmeasured contextual factors, can lower the overall certainty of the observed association. It may raise some concern about the magnitude and generalisability of the observed association. Despite this, the positive association of depression with poor glycaemic control was consistently demonstrated across both unadjusted and adjusted models, as well as in sensitivity analyses. This consistency in the direction of effect supports the robustness of the association and strengthens the overall conclusion, even in the presence of heterogeneity.

It is known that publication bias can influence the conclusion of meta-analyses, and the best approach to address publication bias is to retrieve related unpublished results as advised by Turner et al.⁸⁶ Unfortunately, this approach is beyond the capacity of the present review. Nonetheless, the suggestions of publication bias in this meta-analysis were likely minimal because research has shown that when the studies are heterogeneous, trim-and-fill may inappropriately adjust for publication bias where none exists.⁸⁷ Additionally, in this meta-analysis, the funnel plot asymmetry might result from the heterogeneity among studies and larger effects in smaller studies.

Conclusion

This systematic review and meta-analysis demonstrate the positive association of depression with poor glycaemic control in people with diabetes in LMICs. While longitudinal studies are needed, the findings emphasise the importance of an integrated care approach incorporating the screening and management of depression in people with diabetes to optimise glycaemic control. To translate these findings into practice, healthcare systems in LMICs must prioritise resource allocation, workforce training and the development of care models that address both physical and mental health needs. Health practitioners should be sensitised to the relationship between depression and glycaemic control, particularly in patients with persistently poor diabetes outcomes. By potentially improving glycaemic control, the early screening and timely treatment of comorbid depression in people with diabetes in LMIC settings may not only improve mental health but also reduce diabetes-related complications.

Data availability statement

All data relevant to the study are included in the article or uploaded as supplementary information.