Global and regional level association between depression and glycemic control in type 2 diabetes mellitus: A systematic review and meta-analysis

Abstract

Background:

Depression and type 2 diabetes mellitus (T2DM) are major global public health concerns. Depression worsens glycemic control through behavioral and biological pathways, increasing insulin resistance and poor adherence to diabetes care.

Aim:

This systematic review and meta-analysis (SRMA) examined the association between depression and glycemic control among individuals with T2DM worldwide.

Materials and Methods:

Following PRISMA guidelines, studies published from January 2000 to May 2024 were retrieved from Cochrane, Embase, ProQuest, Web of Science, and PubMed. Thirty-eight studies were included to assess mean differences (MDs) in HbA1c and fasting blood sugar (FBS), proportions with poor glycemic control, and medication adherence among those with and without depression. Random-effects models in R Studio were used, with subgroup and meta-regression analyses addressing heterogeneity.

Results:

Depressed individuals showed a significant mean increase of 0.30% in HbA1c compared to nondepressed counterparts (95% CI: 0.09–0.51; n = 20). Stronger associations were observed in North American studies, larger sample sizes, and those using Patient Health Questionnaire-9 (PHQ-9) cutoff scores >10. Duration of diabetes significantly influenced the HbA1c difference. No significant association was found for poor glycemic control (OR = 1.74, 95% CI: 0.81–3.73; n = 12), FBS (MD = 24.39, 95% CI: −17.29–66.06; n = 4), or medication adherence (OR = 1.86, 95% CI: 0.57–6.10; n = 5). High heterogeneity (I² > 90%) and publication bias were noted, with GRADE rating the evidence as very low certainty.

Conclusion:

Depression modestly affects glycemic outcomes in T2DM, with heterogeneity among studies. Higher HbA1c among depressed patients was found with low certainty and to be minimally clinically significant. Methodologies in the included studies emphasize the need for standard approaches and better study designs (prospective cohort studies).

Prospero registration number:

CRD42024543104, https://www.crd.york.ac.uk/PROSPERO/view/CRD42024543104.

INTRODUCTION

In 2021, it was estimated that over half a billion adults worldwide had diabetes, with more than three-quarters of them living in low- and middle-income countries (LMICs). Diabetes mellitus was linked to approximately 6.7 million deaths during that year. The majority of diabetes cases, exceeding 90%, were due to type 2 diabetes mellitus (T2DM). However, around half of the estimated DM were undiagnosed and prone to complications, including mortality.[1,2] Complications of T2DM include cancers, multiorgan infections and damage, liver disease, functional and cognitive disability, and affective disorders.[3] The American Diabetes Association recommends glycemic control targets of Hemoglobin A1c (HbA1C) levels under 7%, fasting blood glucose below 130 mg/dL, and postprandial blood glucose less than 180 mg/dL.[4,5] Among the individuals aware of the T2DM status, around half achieved glycemic control. The glycemic control varied in high-income countries like the United States (50.5%),[6] upper-middle-income countries like China (16.9%),[7] and LMICs like Tehran (31.7%),[8] Nepal (21%),[9] and India (15.7%).[10] Despite T2DM being preventable, largely controllable, and manageable, the control of diabetes remains a challenge.

Globally, 3.8% of the population, 280 million people (approx.), were reported to have depression.[11] The coexistence of T2DM and depression significantly heightens the risk of disability. A large-scale study across more than 50 countries revealed that depression affects 9.3% of persons with diabetes compared to 3.2% of those without diabetes.[12] Additionally, people with T2DM face a more than 1.2 times increased incidence of depression than the general population.[13,14,15]

Depression has been documented to impact glycemic control through both behavioral and biological pathways. Irregular adherence to diabetic medications and monitoring, unhealthy dietary choices, and a lack of physical activity led to suboptimal glycemic control. Physiologically, in depression, elevated cortisol levels contribute to insulin resistance, impaired glucose uptake, and increased hepatic glucose production, exaggerating impaired glycemic control. Conversely, poor glycemic control can worsen depressive symptoms, creating a vicious cycle. Integrated care models and biopsychosocial approaches that simultaneously address both conditions through pharmacological interventions, psychological/psychosocial therapies, and lifestyle modifications have demonstrated superior outcomes. Recognizing and addressing the interconnected nature of depression and glycemic control is essential for optimizing the health and wellbeing of individuals with diabetes.[16] The goal of achieving glycemic control among individuals with depression is challenging in comparison to other conditions. In addition, depression among females was documented to be higher, and rural–urban disparities across the regions were documented.[17,18]

With an estimated 1.3 billion individuals with T2DM by 2050, addressing the associated conditions like depression remains crucial.[19] Given the variations and inconsistencies in diabetes prevalence, the burden of depression among individuals with T2DM, and the association of depression and glycemic control, there are substantial gaps in understanding how this association varies across global and regional contexts, particularly in relation to cultural, economic, and healthcare differences. Furthermore, synthesizing data on the depression and its relationship with glycemic control at both global and regional levels is critical to generate evidence that informs targeted and effective interventions. We aimed to investigate the global and regional level association between depression and glycemic control in T2DM.

MATERIALS AND METHODS

The current “Systematic Review and Meta-analysis (SRMA)” adhered to the “Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 (PRISMA)” guidelines. [Table S1]. It was registered on PROSPERO (CRD42024543104) (https://www.crd.york.ac.uk/PROSPERO/view/CRD42024543104). The Nested-Knowledge was employed to ensure a structured, rigorous, transparent, and replicable approach to conducting the literature review and meta-analysis.[20]

Supplementary Table S1.

PRISMA Checklist (2020)

Section and Topic	Item #	Checklist item	Location where item is reported
Title
Title	1	Identify the report as a systematic review.	Page 1
Abstract
abstract	2	Made as per the Journal guidelines	Page 1
Introduction
Rationale	3	Describe the rationale for the review in the context of existing knowledge.	Page 2, para 3
Objectives	4	Provide an explicit statement of the objective(s) or question(s) the review addresses.	Page 3, para 2
Methods
Eligibility criteria	5	Specify the inclusion and exclusion criteria for the review and how studies were grouped for the syntheses.	Page3, para 4
Information sources	6	Specify all databases, registers, websites, organisations, reference lists and other sources searched or consulted to identify studies. Specify the date when each source was last searched or consulted.	Page 3, para 5
Search strategy	7	Present the full search strategies for all databases, registers and websites, including any filters and limits used.	Page 3, para 5
Selection process	8	Specify the methods used to decide whether a study met the inclusion criteria of the review, including how many reviewers screened each record and each report retrieved, whether they worked independently, and if applicable, details of automation tools used in the process.	Page 4, para 1,2
Data collection process	9	Specify the methods used to collect data from reports, including how many reviewers collected data from each report, whether they worked independently, any processes for obtaining or confirming data from study investigators, and if applicable, details of automation tools used in the process.	Page 4, para 3,4
Data items	10a	List and define all outcomes for which data were sought. Specify whether all results that were compatible with each outcome domain in each study were sought (e.g. for all measures, time points, analyses), and if not, the methods used to decide which results to collect.	Page 4, para 5
	10b	List and define all other variables for which data were sought (e.g. participant and intervention characteristics, funding sources). Describe any assumptions made about any missing or unclear information.	Page 4, para 5
Study risk of bias assessment	11	Specify the methods used to assess risk of bias in the included studies, including details of the tool(s) used, how many reviewers assessed each study and whether they worked independently, and if applicable, details of automation tools used in the process.	Page 4, para 6
Effect measures	12	Specify for each outcome the effect measure(s) (e.g. risk ratio, mean difference) used in the synthesis or presentation of results.	Page 4, para 5
Synthesis methods	13a	Describe the processes used to decide which studies were eligible for each synthesis (e.g. tabulating the study intervention characteristics and comparing against the planned groups for each synthesis (item #5)).	Page 3, para 4
	13b	Describe any methods required to prepare the data for presentation or synthesis, such as handling of missing summary statistics, or data conversions.	-
	13c	Describe any methods used to tabulate or visually display results of individual studies and syntheses.	Table 1
	13d	Describe any methods used to synthesize results and provide a rationale for the choice(s). If meta-analysis was performed, describe the model(s), method(s) to identify the presence and extent of statistical heterogeneity, and software package(s) used.	Page 5, para 1
	13e	Describe any methods used to explore possible causes of heterogeneity among study results (e.g. subgroup analysis, meta-regression).	Page 5, para 1
	13f	Describe any sensitivity analyses conducted to assess robustness of the synthesized results.	Results
Reporting bias assessment	14	Describe any methods used to assess risk of bias due to missing results in a synthesis (arising from reporting biases).	Results
Certainty assessment	15	Describe any methods used to assess certainty (or confidence) in the body of evidence for an outcome.	Page 4, para 7
Results
Study selection	16a	Describe the results of the search and selection process, from the number of records identified in the search to the number of studies included in the review, ideally using a flow diagram.	Figure 1
	16b	Cite studies that might appear to meet the inclusion criteria, but which were excluded, and explain why they were excluded.	Supplemetary table S6
Study characteristics	17	Cite each included study and present its characteristics.	Table 1
Risk of bias in studies	18	Present assessments of risk of bias for each included study.	Page 6, para 2
Results of individual studies	19	For all outcomes, present, for each study: (a) summary statistics for each group (where appropriate) and (b) an effect estimate and its precision (e.g. confidence/credible interval), ideally using structured tables or plots.	Results
Results of syntheses	20a	For each synthesis, briefly summarise the characteristics and risk of bias among contributing studies.	Results
	20b	Present results of all statistical syntheses conducted. If meta-analysis was done, present for each the summary estimate and its precision (e.g. confidence/credible interval) and measures of statistical heterogeneity. If comparing groups, describe the direction of the effect	Figures 2-5
	20c	Present results of all investigations of possible causes of heterogeneity among study results.	Results
	20d	Present results of all sensitivity analyses conducted to assess the robustness of the synthesized results.	Results
Results
Reporting biases	21	Present assessments of risk of bias due to missing results (arising from reporting biases) for each synthesis assessed.	-
Certainty of evidence	22	Present assessments of certainty (or confidence) in the body of evidence for each outcome assessed.	Page 6, para 3
Discussion
Discussion	23a	Provide a general interpretation of the results in the context of other evidence.	Page 25, para 1
	23b	Discuss any limitations of the evidence included in the review.	Page 27, para 1
	23c	Discuss any limitations of the review processes used.	Page 27, para 1
	23d	Discuss implications of the results for practice, policy, and future research.	Page 27, para 2
Other information
Registration and protocol	24a	Provide registration information for the review, including register name and registration number, or state that the review was not registered.	Page 3, para 3
	24b	Indicate where the review protocol can be accessed, or state that a protocol was not prepared.	Page 1
	24c	Describe and explain any amendments to information provided at registration or in the protocol.	-
Support	25	Describe sources of financial or non-financial support for the review, and the role of the funders or sponsors in the review.	Title page
Competing interests	26	Declare any competing interests of review authors.	Title page
Availability of data, code and other materials	27	Report which of the following are publicly available and where they can be found: template data collection forms; data extracted from included studies; data used for all analyses; analytic code; any other materials used in the review.	Table 1, Supplementary Table S 6

Selection criteria

The research question posed was, “Among adults with type 2 diabetes mellitus, does the presence of depression, compared to those without depression, affect glycemic control as measured by HbA1c or equivalent indicators?” Inclusion and exclusion criteria, detailed in Table S2, guided the screening process. Eligible studies were those published between January 1, 2000 and May 5, 2024 to focus on the literature from the 21^st century. Conference abstracts, and thesis papers and gray literature were not included, and only English-language articles were selected. Search results generated using a predefined strategy were imported into the Nested-Knowledge platform, which automatically removed duplicate entries.

Supplementary Table S2.

Inclusion and exclusion criteria for the systematic review and meta-analysis

Research Question: “What is the global and regional association between depression and glycemic control in type 2 diabetes mellitus?”
	Inclusion	Exclusion
Population	Type 2 diabetes patient Any age Any gender	Type 1 diabetes Gestational diabetes Medication induced diabetes
Cases	With depression Self-reported instrument/scales Structured or semi structured clinical interviews Clinically diagnosed with diagnostic criteria (DSM/ICD)	Substance use disorder
Controls	Without depression	-
Primary outcome	Glycemic control statuss Poor glycemic control Standard mean differences in the glycemic indices (HbA1c, FBS, PPBS)	-
Secondary outcome	Quality of life Diabetic medication adherence Self-efficacy and self-management behaviors	-
Measures of outcome	Odds ratio Risk difference Standard mean difference	-
Study design	Observational analytical studies, case-control, cohort, cross-sectional, RCT	Qualitative studies, case series, case reports, reviews and opinions
Geography	Global	-
Language	English language	-

Search strategy

A comprehensive search was conducted across five databases—”Cochrane, Embase, ProQuest, Web of Science, and PubMed”—on May 5, 2024. Details of the search strategy are included in the supplementary materials [Table S3]. No gray literature or additional source search was done.

Supplementary Table S3.

The adjusted search terms as per searched electronic databases [as of May 5 2024]

Database	No	Search Query	Results
PubMed
	#1	(“diabetes mellitus, type 2”[MeSH Terms] OR “type 2 diabet*”[Title/Abstract] OR “non insulin dependent diabetes mellitus”[Title/Abstract] OR “NIDDM”[Title/Abstract])	245905
	#2	(“depressive disorder”[MeSH Terms] OR “depression”[MeSH Terms] OR depressi*[tiab])	563672
	#3	((“glycemic control”[MeSH Terms] OR glycemic control [tiab])) OR (disease control[Title/Abstract])	98371
	#4	(((“diabetes mellitus, type 2”[MeSH Terms] OR “type 2 diabet*”[Title/Abstract] OR “non insulin dependent diabetes mellitus”[Title/Abstract] OR “NIDDM”[Title/Abstract])) AND ((“depressive disorder”[MeSH Terms] OR “depression”[MeSH Terms] OR depressi*[tiab]))) AND (((“glycemic control”[MeSH Terms] OR glycemic control [tiab])) OR (disease control[Title/Abstract]))	574
	#5	(((“diabetes mellitus, type 2”[MeSH Terms] OR “type 2 diabet*”[Title/Abstract] OR “non insulin dependent diabetes mellitus”[Title/Abstract] OR “NIDDM”[Title/Abstract])) AND ((“depressive disorder”[MeSH Terms] OR “depression”[MeSH Terms] OR depressi*[tiab]))) AND (((“glycemic control”[MeSH Terms] OR glycemic control [tiab])) OR (disease control[Title/Abstract]))	453
	#6	(((“diabetes mellitus, type 2”[MeSH Terms] OR “type 2 diabet*”[Title/Abstract] OR “non insulin dependent diabetes mellitus”[Title/Abstract] OR “NIDDM”[Title/Abstract])) AND ((“depressive disorder”[MeSH Terms] OR “depression”[MeSH Terms] OR depressi*[tiab]))) AND (((“glycemic control”[MeSH Terms] OR glycemic control [tiab])) OR (disease control[Title/Abstract]))	438
	#7	((“diabetes mellitus, type 2”[MeSH Terms] OR “type 2 diabet*”[Title/Abstract] OR “non insulin dependent diabetes mellitus”[Title/Abstract] OR “NIDDM”[Title/Abstract]) AND (“depressive disorder”[MeSH Terms] OR “depression”[MeSH Terms] OR “depressi*”[Title/Abstract]) AND (“glycemic control”[MeSH Terms] OR “glycemic control”[Title/Abstract] OR “disease control”[Title/Abstract])) AND (humans[Filter]) AND (2000/1/1:2024/5/5[pdat]) AND (english[Filter]))	423
Embase
	#1	‘non insulin dependent diabetes mellitus’/de OR ‘type 2 diabetes’:ab,kw,ti OR ‘type 2 diabet*’:ab,kw,ti OR ‘non insulin dependent diabetes’:ab,ti,kw OR ‘niddm’:ti,ab,kw	412532
	#2	‘depression’/de OR ‘depression’:ab,kw,ti OR ‘depressi*’:ab,kw,ti	870072
	#3	‘glycemic control’/de OR ‘glycemic control’:ab,kw,ti	88820
	#4	#1 AND #2 AND #3	1662
	#5	#1 AND #2 AND #3 AND [humans]/lim AND [english]/lim	1564
	#6	#1 AND #2 AND #3 AND [humans]/lim AND [english]/lim AND [01-01-2000]/sd NOT [06-05-2024]/sd	1548
Web of Science
	#1	(((TS=(“type 2 diabetes mellitus”)) OR TS=(“type 2 diabet*”)) OR TS=(“non insulin dependent diabetes mellitus”)) OR TS=(“NIDDM”)	223712
	#2	(TS=(“depressive disorder”)) OR TS=(“depression”)	519845
	#3	TS=(“glycemic control”)	49326
	#4	#1 AND #2 AND #3	996
	#5	#1 AND #2 AND #3 Timespan: 2000-01-01 to 2024-05-05	985
	#6	#1 AND #2 AND #3 and English (Languages) Timespan: 2000-01-01 to 2024-05-05	961
Cochrane
	#1	(“type 2 diabetes” OR “non insulin dependent diabetes mellitus” OR “NIDDM”):ti,ab,kw	48520
	#2	MeSH descriptor: [Diabetes Mellitus, Type 2] explode all trees	26302
	#3	#1 OR #2	53267
	#4	(“depressive disorder” OR “depressive disorders” OR “depression”):ti,ab,kw	10683
	#5	MeSH descriptor: [Depression] explode all trees	18240
	#6	#4 OR #5	107683
	#7	(“glycemic control”):ti,ab,kw	17161
	#8	MeSH descriptor: [Glycemic Control] explode all trees	452
	#9	#7 OR #8	17161
	#10	#3 AND #6 AND #9	344
ProQuest
	#1	abstract(“type 2 diabet*” OR “non insulin dependent diabetes mellitus” OR NIDDM) OR title(“type 2 diabet*” OR “non insulin dependent diabetes mellitus” OR NIDDM) OR mainsubject(type 2 diabetes mellitus)	45510
	#2	summary(“depressive disorder” OR “depression”) OR title(“depressive disorder” OR “depression”) OR mainsubject(depression)	147564
	#3	summary(“glycemic control”) OR title(“glycemic control”) OR mainsubject(Glycemic Control)	7309
	#4	[S1] AND [S2] AND [S3]	149
	#5	([S1] AND [S2] AND [S3]) AND pd(20000101-20240505)	148
	#6	([S1] AND [S2] AND [S3]) AND (la.exact(“ENG”) AND pd(20000101-20240505))	139

Screening process

• Title and Abstract Screening: Two authors (BN and AL) independently screened the imported studies by reviewing titles and abstracts against the inclusion and exclusion criteria. Discrepancies were resolved by a third author (APG).

• Full-Text Screening: Studies shortlisted during the initial screening underwent a detailed full-text evaluation. This step was conducted independently by four authors (BN, AL, NR, and FK), with final decisions reviewed by a fifth author (APG).

• Data extraction: Relevant data were extracted independently by two authors (BN and AL) for each study and tagged using an Excel form, including author details, publication year, study location, methodology, participant demographics, and glycemic control outcomes. Any disagreements were resolved by third reviewer (APG). Authors were not contacted for seeking clarifications.

• No automation was used in any of the stages except for deduplication.

Study outcomes

Data were extracted mainly for age, study design, total population, depression cutoff and instrument used, duration of diabetes, cutoff for poor glycemic control, FBS, postprandial blood sugar (PPBS) and other study outcomes like medication adherence and quality of life. Primary outcomes included the mean difference (MD) in HbA1c levels and the proportion of patients with poor glycemic control. Secondary outcomes encompassed medication adherence. These metrics were analyzed to assess the relationship between depression and glycemic control in individuals with T2DM.

Risk of bias and study quality

Study quality was evaluated using design-specific tools such as the “Newcastle-Ottawa Scale (NOS)” for cohort studies[21] and “Joanna Briggs Institute (JBI)” critical appraisal tools for cross-sectional studies.[22] Two reviewers worked independently and completed the risk of bias per study (AL and NR), and the conflicts were adjudicated by a third reviewer (AGP).

Certainty in the evidence

The certainty of pooled estimates for clinically important outcomes was assessed and summarized according to the “Grading of Recommendations, Assessment, Development, and Evaluations (GRADE)” methodology, utilizing GRADEpro software.[23] GRADEpro is freely available online software (https://www.gradepro.org/) which enables systematic reviewers to ascertain the certainty in the outcomes reported in the meta-analysis. It requires basic training on how to make decisions for each of the five domains of GRADE within GRADEpro tool, for which tutorials are provided in the website. Reviewers can choose between serious, very serious, and not serious judgements for each of the domains. Reasons need to be added whenever the reviewer chooses ‘Serious, Very Serious’ judgement for a domain. Based on these decisions and the study design, GRADEpro automatically generates final certainty levels as ‘High, Moderate, Low, Very Low,’ which is downloadable.

Statistical analysis

R studio software[24] was used for conducting meta-analysis,[25,26] and the “meta” package 7.0.0 was used. Forest plots were generated to present odds ratios (ORs) for pooled categorical data and mean differences (MDs) for pooled continuous data, with 95% confidence intervals (95% CIs). Since HbA1c was measured on the same scale across studies, MD was used instead of the standardized mean difference (SMD). Crude OR was used, which was derived from the crude counts. Heterogeneity was assessed using the I² test, and a random-effects model was used to assess pooled estimates, which included prediction intervals. A random-effects model with the Hartung–Knapp adjustment was used to obtain more accurate pooled estimates and confidence intervals. For outcomes with at least 10 studies, meta-regression and stratified analyses were performed. Publication bias was assessed using a funnel plot (if more than 10 studies were eligible for meta-analysis), Egger’s regression test and Doi plots (if more than five studies were eligible for meta-analysis), and Luis Furuya–Kanamori (LFK) indices (method to assess asymmetry, which indicates possible publication bias. Values above +1 and below -1 indicated major asymmetry). Subgroup analyses were performed based on geographical location and depression assessment tools.

RESULTS

Study selection

A total of 1551 articles were retrieved through database searches, including EMBASE (n = 25), PUBMED (n = 423), Web of Science (n = 959), Cochrane Library (n = 6), and ProQuest (n = 138). Duplicate entries were removed using Nested Knowledge, and the remaining articles underwent title and abstract screening. Following this, a full-text review was conducted to assess eligibility. No additional studies were identified through bibliography screening (manual searching). Several articles were excluded during the full-text review due to inappropriate outcomes [Figure 1 and Table S4 ^{(193.9KB, pdf)} ]. Manual citation searching of the reference list of the included studies (backward chasing) was carried out.

Figure 1.

PRISMA flow diagram

Ultimately, 38 studies met the inclusion criteria and were incorporated into the SRMA. The PRISMA flow chart illustrating the study selection process is presented in Figure 1.

Study characteristics

The individual study demographics are enumerated in Table 1. The sample sizes ranged from 23 to 73,739. Out of the total studies, 34 were cross-sectional, two were cohort studies,[27,28] and one was a case-control study.[29] The study design was not mentioned for one study. Eight different tools were used to assess depression in these studies. Studies were different countries like China (n = 4),[30,31,32,33] USA (n = 10),[27,34,35,36,37,38,39,40,41,42] India (n = 1)[29] Ethiopia (n = 2),[32,33] Canada (n = 2),[28,34] Pakistan (n = 1),[35] Indonesia (n = 1),[36] Qatar (n = 1),[37] Peru (n = 1),[38] Tunisia (n = 1),[39] Saudi Arabia (n = 2),[40,41] Hungary (n = 1),[42] Mexico (n = 2),[43,44] UAE (n = 1),[45] Malaysia (n = 1),[46] Turkey (n = 1),[47] Vietnam (n = 1),[48] Palestine (n = 1),[49] Serbia (n = 1),[50] and Kuwait (n = 2).[51,52] Variables included in the adjusted analysis are enumerated in Table S5.

Table 1.

Characteristics of the studies included for SRMA

Author, Year	Study design	Country	Study population (n)	Duration of diabetes	Mean age	Female (n)	Male (n)	Tool	Depression cutoff	Cutoff for HbA1c	Outcome
Abuhegazy et al., 2014[41]	Cross-sectional	Saudi Arabia	172	10.7 years	51.3	120	152	BDI-II	>16	higher HbA1c	Glycemic control
Ali et al., 2023[32]	Cross-sectional	Ethiopia	263	Not given	50.21	129	134	PHQ-9	>=6	>7	Glycemic control
Al-Ozairi et al., 2020[52]	Cross-sectional	Kuwait	465	12.5 years	55.3	224	241	PHQ-9	>=10		Glycemic control
Al-Ozairi et al., 2023[51]	Cross-sectional	Kuwait	446	13years	55.3	228	218	PHQ-9	>=10		Glycemic control
Amelia and Yunanda, 2018[36]	Cross-sectional	Indonesia	100	5-10 years	-	-	-	CES-D	Not given	>=8	Glycemic control
Azeze et al., 2020[33]	Cross-sectional	Ethiopia	418	63.7% had <=8 years	-	-	-	PHQ-9	>=12	An average fasting blood glucose level from three recent visits of ≤130 mg/dL was consid-ered good glycemic control.	Glycemic control
Badedi et al., 2016[40]	Cross-sectional	Saudi Arabia	288	median- 7 years	54.58	143	145	PHQ-9	Not given	HbA1c <7%	Glycemic control
Bawadi et al., 2021[37]	Cross-sectional	Qatar	2448	Not given	51.6	-	-	PHQ-9	>=10		Glycemic control
Chiu et al., 2010[31]	Ongoing biennial survey	USA	998	12.5 years	-	-	-		>=3/8		Glycemic control
Crispín-Trebejo et al., 2015[38]	Cross-sectional	Peru	277	7.1 years	59	93	184	PHQ-9	>=15	≥7% HbA1c	Glycemic control
De la Roca-Chiapas et al., 2013[44]	Not mentioned (cross-sectional)	Mexico	65	12.5 years	55.8	39	26	GDS	>=16		Glycemic control
Dinh Le et al., 2022[48]	Cross-sectional	Vietnam	231	Not given		109	122	PHQ-9	Not given		Glycemic control
Egede and Ellis, 2010[53]	Cross-sectional	USA	201	11.1 years	56.6	146	55	CES-D	>=10		Glycemic control, Quality of life
Ellouze et al., 2017[39]	Cross-sectional	Tunisia	100	12 years	-	60	40	HADS	>=7		Glycemic control
Fung et al., 2018[30]	Cross-sectional	China	325	Not given	69*	-	-	GDS	>=3		Glycemic control
Golden et al, 2017[54]	Cross-sectional	USA		Not given	-	-	-	-	multiple cutoff		Glycemic control
Hargittay et al, 2022[42]	Cross-sectional	Hungary	338	Not given	64	-	-	HAM-D	>=11		Glycemic control
Hasan et al., 2015[46]	Cross-sectional	Malaysia	611	not given	56.25*	-	-	-	>=16	HbA1C levels >7.5	Glycemic control
Hawamdeh et al., 2013[45]	Cross-sectional	UAE	92	Not given	-	-	-	BDI	>=8		Glycemic control
Juárez-Rojop et al., 2023[43]	Cross-sectional	Mexico	376	Not given	55.42	263	113	HAM-D	Not given		Glycemic control
Kalsekar et al., 2006[27]	Retrospective cohort design	USA	1326	6.3 years	-	-	-	ICD-9 CM	Not given	≥7% HbA1c	Medication nonadherence
Kayar et al., 2017[47]	Cross-sectional	Turkey	154	Not given	54.8	82	72	SCID-I	>=16		Glycemic control
Kendzor et al, 2014[55]	Cross-sectional	USA	486	multiple categories	51.3	315	171	CES-D	>=5	≥7.5% HbA1c	Glycemic control
Lee et al., 2009[56]	Cross-sectional	USA	55	11.9 years	58.9	23	32	BDI-II	>=13		Glycemic control
Lin et al., 2004[57]	Cross-sectional	USA	4463	Not given	-	-	-	-	Not given		Self-care activities
Lunghi et al., 2017[28]	Population-based cohort study	Canada	73739	not given	-	36194	37545	-	Not given		Medication nonadherence
Sue Penckofer et al., 2012[58]	Cross-sectional	USA	23	3 years	52.61*	23	-	CES-D	>=11		Glycemic control
Richardson et al., 2008[59]	Cross-sectional	USA	11,525	Not given	66	-	-	ICD-9	>=10		Glycemic control
Sidhu et al., 2017[34]	Cross-sectional	Canada	41	mean 4.28 in cases, 4.85 in controls	67	30	11	PHQ-9	>=16		Glycemic control
Singh et al., 2014[29]	Cross-sectional case-control design	India	109	mean- 11.96 in cases, 12.11 in controls	51	56	53	CES-D	>=16		Glycemic control
Stanković et al., 2011[50]	Cross-sectional	Serbia	90	median- 10 years	56	59	31	BDI	>=16		Glycemic control
Sweileh et al., 2014[49]	Cross-sectional	Palestine	294	mean 12 in cases, 10 in controls	-	164	130	BDI	>=16	HbA1c ≥7	Glycemic control
Téllez-Zenteno et al., 2002[60]	Cross-sectional	Mexico	189	Not given	61.7	109	80	BDI	>=14	Blood glucose 200 mg/dL or more as an average of the last five blood glucose readings	Glycemic control
W Zhang et al., 2015[61]	Cross-sectional	China	412	8.93 years		207	205	BDI	>=14		Glycemic control
Werremeyer et al., 2016[62]	Cross-sectional	USA	517	Not given	-	345	172	PHQ -9	>=15	Patients not at HbA1c goal of <7%	Glycemic control
Y Zhang et al., 2015[63]	Cross-sectional	China	2538	median- 6 years	56.4	1192	1346	PHQ-9	>=10		Glycemic control, medication nonadherence
Zhang et al., 2013[64]	Cross-sectional	China	586	7 years	55.1	239	347	PHQ-9	>=7	HbA1c level, and were less likely to achieveHbA1c target (7.0%)	Glycemic control
Zuberi et al., 2011[35]	Cross-sectional	Pakistan	286	64.7% had more than 3 years	52	158	128	HADS	>=8	≥7% HbA1c	Glycemic control, medication nonadherence

*Median age converted to mean age. FBS, Fasting blood sugar; PHQ, Patient Health Questionnaire; HADS, Hospital Anxiety and Depression Scale; BDI, Beck’s Depression Inventory; CES-D, Centre for Epidemiological Studies Depression; GDS, Geriatric depression scale; SCID, Structural clinical interview for DSM; ICD-CM, International classification of Diseases-Clinical Modification; HAM-D, Hamilton Depression rating scale

Supplementary Table S5.

Variables adjusted for the final analysis in the articles

Author	Adjusted variable
Crispín-Trebejo et al, 2015	Model 1: adjusted by gender, age, education level, time of disease, working currently, place of birth, and hospital setting, Model 2: Model 1 plus hospital status, and last year hospital admissions, Model 3: Model 2, plus hypertension diagnosis, retinopathy diagnosis, and diabetic foot diagnosis, Model 4: Model 3, plus total cholesterol and 24-hour proteins in urine.
Sweileh et al, 2014	gender, education, occupation, additional illness, BMI, medication adherence score
Zuberi et al, 2011	age, gender, BMI, family history of DM, family history of depression, smoker, duration of illness, complications, treatment
Kendzor et al, 2014	age, gender, years of education, assessment language, birth country
Roca-Chiapas et al, 2013	not given
Zhang et al, 2013	age, sex, and disease duration
Bawadi et al, 2021	model 1 was adjusted for age, gender, and poor glycemic control; Model 2 was further attuned for smoking, physical activity, education, and BMI; lastly, Model 3 was further adjusted for insulin use, utilization of diabetes medications other than insulin, and the use of hypertension medications
Al-Ozairi et al 2023	Mode 1: without a confounder. Model 2: confounder variables are age, body mass index, gender and insulin medication
Azeze et al, 2020	marital status, had child, smoking, glycemic control, having HTN, retinopathy, duration of DM, duration of DM treatment, family history of DM
Golden et al, 2017	age, sex, race, and diabetes duration
Egede et al 2010	age, diabetes duration, education, body weight, sex, employment status, insurance status, income, number of comorbid conditions, and insulin use
Hasan et al, 2015	depression or anxiety, age, comorbidities, physical activity, BMI, and physical health
Hargittay et al, 2022	HAM-A score, age, gender, BMI, alcohol consumption, physical exercise, smoking, and levels of education
Le et al, 2022	age, gender, treatment therapy, microvascular complications
Kayar et al, 2017	age, gender, duration of DM, HTN, nephropathy, retinopathy, cardiovascular disease, neuropathy, dietary compliance, physical exercise, self-monitoring of blood glucose, drug compliance
Zhang et al.2013	age, gender, employment status, BMI, duration of DM, social support, diabetes complications, and others
Juárez-Rojop et al, 2023	age, gender, marital status, ischemic heart disease, diabetic neuropathy, fatty liver
Fung et al, 2018	Model 1: Adjusted for age, gender, and use of anti-hypertensive drugs. Model 2: Adjusted for age, gender, use of anti-diabetic drugs, insulin, lipid-lowering drugs and anti-hypertensive drugs. Model 3: Adjusted for age, gender, use of anti-diabetic drugs, and use of insulin. Model 4: adjusted for age, gender, and duration of diabetes.
Al-Ozairi et al 2020	Model 1: regression included gender, past history of depression, hypertension, insulin user, exercise, body mass index (BMI), Model 2: included model 1 plus psychological measures. ∗Psychological measure was PAID,∗∗Psychological measure was PHQ-9

Risk of bias assessment

Two independent authors (AL and NR) assessed the risk-of-bias assessment for cross-sectional and cohort studies using the JBI and Newcastle-Ottawa assessment (NOS) tool. A total of 36 studies were evaluated for risk of bias using the JBI Critical Appraisal Checklist for Analytical Cross-Sectional Studies and the NOS for two cohort studies. The JBI tool evaluated domains including inclusion criteria, measurement of exposure and outcomes, identification and control of confounding factors, and appropriateness of statistical analysis. All cohort studies assessed using the NOS were of good methodological quality (scores 9). Overall, all 36 studies met the required quality standards and were retained for the final analysis [Table S6A and B].

Supplementary Table S6A.

Quality assessment of included cohort studies with the use of Newcastle-Ottawa quality assessment tool

Author (YOP)	S1	S2	S3	S4	C1	O1	O2	O3	Overall Quality
Kalsekar et al.	*	*	*	*	**	*	*	*	Good
Lunghi et al.	*	*	*	*	**	*	*	*	Good

S1 - Representativeness of the exposed cohort. S2 - Selection of the non-exposed cohort. S3 - Ascertainment of exposure. S4 - Demonstration that outcome of interest was not present at start of study. C1 - Comparability of cohorts on the basis of the design or analysis. O1 - Assessment of outcome. O2 - Was follow-up long enough for outcomes to occur. O3 - Adequacy of follow-up of cohorts

Supplementary Table S6B.

Quality assessment of included cross-sectional studies with the use of JBI – analytical - quality assessment tool

Author	1	2	3	4	5	6	7	8	Include/Exclude
Abuhegazy et al. 2014	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Include
Al Ozairi et al., 2023	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Include
Ali et al., 2023	Yes	Yes	Yes	Yes	No	Yes	Yes	Yes	Include
Al-Ozairi et al., 2020	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Include
Amelia and Yunanda et al., 2018	Yes	Yes	Yes	Yes	No	No	Yes	No	Include
Azeze et al., 2020	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Include
Badedi et al., 2016	Yes	Yes	Yes	Yes	No	Yes	Yes	Yes	Include
Bawadi et al., 2021	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Include
Chiu et al., 2010	Yes	Yes	Yes	Yes	No	No	Yes	Yes	Include
Crispín-Trebejo et al., 2015	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Include
De la Roca-Chiapas et al., 2013	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Include
Dinh Le et al., 2004	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Include
Egede Ellis et al., 2010	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Include
Ellouze et al., 2017	Yes	Yes	Yes	Yes	No	No	Yes	Yes	Include
Fung et al., 2018	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Include
Golden et al., 2017	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Include
Hargittay et al., 2022	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Include
Hasan et al., 2015	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Include
Hawamdeh et al., 2013	Yes	Yes	Yes	Yes	No	No	Yes	Yes	Include
JuárezRojop et al., 2023	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Include
Kayar et al., 2017	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Include
Kendzor et al., 2014	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Include
Lee et al., 2009	Yes	Yes	Yes	Yes	No	No	Yes	Yes	Include
Lin et al., 2004	Yes	Yes	Yes	Yes	No	Yes	Yes	Yes	Include
Penckofer et al., 2012	Yes	Yes	Yes	Yes	No	No	Yes	Yes	Include
Richardson et al., 2008	Yes	Yes	Yes	Yes	No	Yes	Yes	Yes	Include
Sidhu et al., 2017	Yes	Yes	Yes	Yes	No	No	Yes	Yes	Include
Singh et al., 2014	Yes	Yes	Yes	Yes	No	No	Yes	No	Include
Stankovic et al., 2011	Yes	Yes	Yes	Yes	No	Yes	Yes	Yes	Include
Sweilah et al., 2014	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Include
Téllez-Zenteno et al., 2002	Yes	Yes	Yes	Yes	No	No	Yes	Yes	Include
W Zhang et al., 2015	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Include
Werremeyer et al., 2016	Yes	Yes	Yes	Yes	No	No	Yes	Yes	Include
Y Zhang et al., 2015	Yes	Yes	Yes	Yes	No	Yes	Yes	Yes	Include
Zhang et al., 2013	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Include
Zuberi et al., 2011	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Include

1 - Were the criteria for inclusion in the sample clearly defined? 2 - Were the study subjects and the setting described in detail? 3 - Was the exposure measured in a valid and reliable way? 4 - Were objective, standard criteria used for measurement of the condition? 5 - Were confounding factors identified? 6 - Were strategies to deal with confounding factors stated? 7 - Were the outcomes measured in a valid and reliable way? 8 - Was appropriate statistical analysis used?

GRADE assessment

As per the GRADE[23] assessment, mean difference in HbA1c was rated to have very low certainty owing to substantial heterogeneity (inconsistency) between the included studies and poor glycemic control, and mean difference in FBS and poor medication adherence were also rated to have very low certainty due to high heterogeneity (inconsistency) and serious imprecision [Table S7].

Supplementary Table S7.

GRADE assessment of the mean difference in HbA1c between the groups

Certainty assessment							№ of patients		Effect		Certainty	Importance
№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Depression	No Depression	Relative (95% CI)	Absolute (95% CI)
Mean difference in HbA1c
20	non-randomised studies	not serious	seriousa	not serious	not serious	none	2277	17884	-	MD 0.3 percentage points higher (0.09 higher to 0.51 higher)	⨁◯◯◯ Very lowa	Critical
Poor glycemic control
12	non-randomised studies	not serious	very seriousb	not serious	seriousc	none	806/1304 (61.8%)	2317/4364 (53.1%)	OR 1.74 (0.81 to 3.73)	132 more per 1,000 (from 53 fewer to 278 more)	⨁◯◯◯ Very lowb,c	Critical
Mean difference in FBS
4	non-randomised studies	not serious	very seriousb	not serious	seriousd	none	313	544	-	MD 24.39 mg/dL higher (2.27 higher to 46.5 higher)	⨁◯◯◯ Very lowb,d	Critical
Poor medication adherence
5	non-randomised studies	not serious	very seriousb	not serious	seriousc	none	1808/3598 (50.3%)	34925/73448 (47.6%)	OR 1.86 (0.57 to 6.10)	152 more per 1,000 (from 135 fewer to 371 more)	⨁◯◯◯ Very lowb,c	Important

cI: confidence interval; MD: mean difference; OR: odds ratio. a. Heterogeneity is substantial (>50%). b. Heterogeneity is very high (>75%). c. 95% CI crosses the point of null effect. d. 95% CI is wide

Outcome

The main outcomes of the study were under as follows:

1. Mean difference in HbA1c

2. Proportion of poor glycemic control (HbA1c)

3. Mean difference in FBS

4. Proportion of poor glycemic control (FBS)

5. Medication adherence

6. Quality of life.

The pooled outcomes for a, b, c, and e are given in results section 3.5. Only one study had assessed the proportion of poor glycemic control using FBS.[33]

The study found that individuals with T2DM who had poor glycemic control had over six times higher odds of having untreated depression compared to those with good glycemic control (AOR = 6.23; 95% CI: 3.65–10.54). Similarly, only one study had found quality of life among depressed and nondepressed diabetic patients. Women with type 2 diabetes who were depressed (CES-D ≥16) reported a lower overall quality of life compared to nondepressed women (mean = 19.1 ± 5.0 vs. 23.3 ± 4.7; P = 0.054).[58] There were no studies that had assessed poor glycemic control among depressed and nondepressed based on post-prandial blood glucose.

Outcome based on the severity of depression

Among the included studies, two studies further divided depression cases into two subgroups. Zhang et al.[64] classified the depressed cases into mild (PHQ-9 < 10) and major (PHQ-9 >=10) and found the mean HbA1c levels were 7.43 (1.30) and 8.2 (2.19), respectively. The proportion of poor glycemic control was 58.1% and 76.9%, respectively. Lee et al.[56] classified depressed cases based on BDI-II into group: BDI-II <20 and group: BDI >=20. The groups’ respective mean HbA1c were 7.41 (1.22) and 7.80 (1.22).

Outcomes of pooled studies

The results of the meta-analyses are discussed under the following headings:

1. Mean difference in HbA1c among depressed and nondepressed diabetic patients.

2. Proportion of poor glycemic control (HbA1c) in depressed and nondepressed among the diabetes.

3. Mean difference in FBS among depressed and nondepressed diabetic patients.

4. Proportion of poor medication adherence in depressed and nondepressed diabetes patients.

Mean difference in HbA1c among depressed and nondepressed diabetic patients

Twenty studies assessed the mean difference in HbA1c among depressed and nondepressed diabetic patients. The pooled MD indicates a 0.30 percentage point increase in mean HbA1c in depressed patients as compared to nondepressed diabetic patients. (MD = 0.30, 95% CI: 0.09- 0.51). Heterogeneity was observed between the studies (I² = 67.4%, P < 0.0001) [Figure 2].

Figure 2.

Forest plot of mean difference in HbA1c among depressed and nondepressed diabetic patients

Subgroup analysis

Subgroup analysis was done based on continent, sample size, tool used, and depression cutoff used for different tools. Subgroup analysis for continent showed that the mean HbA1c difference between depressed and nondepressed diabetic patients was significant in North American studies (MD = 0.28; 95% CI: 0.01–0.54) but not in those from Asia or Europe [Figure S.1.A ^{(2.4MB, tif)} ]. Studies with larger samples showed significantly higher HbA1c among depressed patients (MD = 0.34; 95% CI: 0.12–0.57), whereas smaller studies did not [Figure S.1.B ^{(2.2MB, tif)} ]. Among depression scales, only studies using HAM-D reported a significant difference (MD = 0.65; 95% CI: 0.19–1.11) [Figure S.1.C ^{(3MB, tif)} ].

For PHQ-9, a cutoff ≥10 showed a higher HbA1c in depressed patients (MD = 0.21; 95% CI: 0.09–0.33), while other thresholds were not significant [Figure S.2.A ^{(1.6MB, tif)} -E ^{(963.4KB, tif)} ].

Meta-regression analysis based on mean age, year of publication, and mean duration of diabetes

Meta-regression analysis demonstrated a statistically significant association between the mean duration of diabetes and the mean difference in HbA1c between depressed and nondepressed patients (β = –0.11; 95% CI: –0.19 to –0.03; P = 0.01) [Figure S.3.A ^{(607.2KB, tif)} ]. In contrast, year of publication (β = 0.00; 95% CI: –0.04 to 0.04; P = 0.94) and mean age of participants (β = –0.01; 95% CI: –0.05 to 0.03; P = 0.54) were not significantly associated with the mean HbA1c difference [Figure S.3.B ^{(607.2KB, tif)} and C ^{(608.7KB, tif)} ].

Leave-one-out analysis

Leave-one-out sensitivity analysis demonstrated that sequential exclusion of individual studies did not alter the pooled mean difference in HbA1c between depressed and nondepressed diabetic patients [Figure S.4 ^{(3.3MB, tif)} ].

Publication bias

Assessment of publication bias revealed no major asymmetry in the funnel plot [Figure S.5 ^{(552.7KB, tif)} ]. The LFK index was 0.7, consistent with no asymmetry [Figure S.6 ^{(552.7KB, tif)} ]. Similarly, Egger’s regression test (t = 1.31; df = 18; P = 0.21; bias estimate = 1.08 ± 0.83) did not demonstrate significant funnel plot asymmetry, confirming that publication bias was unlikely to have influenced the meta-analysis results.

Proportion of poor glycemic control (HbA1c) in depressed and nondepressed among the diabetes

Twelve studies examined the proportion of patients with poor glycemic control (HbA1c) among individuals with diabetes, comparing those with and without depression. Depressed patients were found to have 1.74 times higher odds of poor glycemic control as compared to nondepressed patients (OR = 1.74, 95% CI: 0.81–3.73). The prediction interval, ranging from 0.13 to 23.31, reflected considerable variability across the studies. Substantial heterogeneity was detected (I² = 90.6%, P < 0.0001) [Figure 3].

Figure 3.

Forest plot of the proportion of poor glycemic control (HbA1c) in depressed and nondepressed among the diabetes

Subgroup analysis

Subgroup analysis showed that the proportion of patients with poor glycemic control (based on HbA1c) was significantly higher among depressed compared with nondepressed diabetic patients in studies from North America (MD = 1.57; 95% CI: 1.01–2.46), South America (MD = 5.54; 95% CI: 1.29–23.85), Asia (MD = 1.24; 95% CI: 0.45–3.43), and Africa (MD = 4.53; 95% CI: 0.00-470247.20) [Figure S.7.A ^{(2.7MB, tif)} ].

When stratified by sample size, both large (MD = 1.74; 95% CI: 0.98–3.08) and small studies reported a higher proportion of poor glycemic control among depressed patients [Figure S.7.B ^{(2.2MB, tif)} ].

By assessment tool, increased odds of poor glycemic control were observed in studies using HADS (MD = 6.86; 95% CI: 0.03–1448.45), BDI (MD = 0.63; 95% CI: 0.03–12.53), PHQ-9 (MD = 1.80; 95% CI: 0.93–3.46), CES-D (MD = 0.10; 95% CI: 0.02–0.45), and SCID-I (MD = 3.57; 95% CI: 1.17–10.92) [Figure S.7.C ^{(3MB, tif)} ].

For PHQ-9, significant associations were noted at cutoffs of ≥15 (MD = 2.42; 95% CI: 0.00–4732.00) [Figure S.7.D ^{(2.2MB, tif)} -F ^{(1.3MB, tif)} ].

Meta-regression

The meta-regression analysis showed no significant effect of publication year (β = –0.06, 95% CI: –0.26 to 0.13, P = 0.54), mean duration of diabetes (β = 0.13, 95% CI: –0.35 to –0.61, P = 0.60), or mean age of participants (β = 0.02, 95% CI: –0.18 to 0.23, P = 0.82) on the mean difference in HbA1c levels between depressed and nondepressed diabetic patients [Figure S.8.A ^{(473.6KB, tif)} -C ^{(473.6KB, tif)} ].

Leave-one-out analysis

The leave-one-out sensitivity analysis revealed that sequential exclusion of individual studies did not substantially change the pooled proportion of patients with poor glycemic control among depressed and nondepressed diabetic individuals [Figure S.9 ^{(1.5MB, tif)} ].

Publication bias

The LFK index was 3.05, indicating substantial asymmetry in the DOI plot and suggesting possible publication bias or small-study effects [Figure S.10 ^{(551.2KB, tif)} ]. However, Egger’s regression test (bias estimate = 3.11; SE = 1.71; t = 1.82; df = 10; P = 0.0986) did not reveal statistically significant evidence of publication bias. Consistently, the funnel plot demonstrated a roughly symmetric, inverted funnel-shaped distribution of studies around the pooled estimate, indicating that publication bias was unlikely to have materially influenced the meta-analysis results [Figure S.11 ^{(553.7KB, tif)} ].

Mean difference in FBS between depressed and nondepressed diabetes patients

Four studies compared the mean fasting blood sugar (FBS) levels between depressed and nondepressed diabetic patients. The pooled analysis using a random-effects model showed that depressed individuals had a higher mean FBS by 24.39 mg/dL compared to their nondepressed counterparts (MD = 24.39; 95% CI: −17.29 to 66.06). However, it was not statistically significant. Substantial heterogeneity was observed among studies (I² = 83.7%, P = 0.0004), suggesting notable variability in the mean differences reported [Figure 4].

Figure 4.

Forest plot of mean difference in FBS between depressed and nondepressed diabetic patients

Leave-one-out analysis

The plot shows that removal of individual studies does not substantially alter the pooled mean difference, indicating the robustness and stability of the meta-analytic findings for FBS [Figure S.12 ^{(955.4KB, tif)} ].

Proportion of poor medication adherence in depressed and nondepressed diabetes patients

Five studies evaluated the proportion of poor medication adherence among T2DM patients with and without depression. The pooled OR showed no significant difference in the likelihood of poor medication adherence between depressed and nondepressed individuals (OR = 1.86, 95% CI: 0.57–6.10). The prediction interval, from 0.11 to 31.27, indicated substantial variability across the studies due to a few studies. High heterogeneity was observed (I² = 93.7%, P < 0.0001) [Figure 5].

Figure 5.

Forest plot of proportion of poor medication adherence in depressed and nondepressed diabetes patients

Leave-one-out analysis

The leave-one-out sensitivity analysis for poor medication adherence showed that omitting any individual study had minimal impact on the pooled odds ratio [Figure S.13 ^{(1,007.6KB, tif)} ].

DISCUSSION

This SRMA provides a comprehensive synthesis of the association between depression and diabetes outcomes, particularly glycemic control and medication adherence. The pooled mean difference in HbA1c levels revealed that depressed diabetic patients exhibited a higher mean HbA1c by 0.29 percentage units compared to their nondepressed counterparts. This association was particularly pronounced in studies conducted in North America, which may reflect regional differences in healthcare system dynamics, health behavior, and lifestyle differences. However, the mean difference could not reach the minimal clinically important difference for HbA1c, which is taken as 0.5%.[65] Hence, the overall association between depression and glycemic control needs to be taken with caution from a clinical perspective.

Depression is associated with elevated systemic inflammation, marked by an increase in the levels of proinflammatory cytokines such as interleukin-6 (IL-6) and tumor necrosis factor-alpha (TNF-α). These cytokines impair insulin signaling, contributing to insulin resistance and subsequent hyperglycemia. Miller et al.[66] reported elevated inflammatory markers in individuals with depression, emphasizing their role in metabolic dysregulation. Dowlati et al.[67] reinforced these findings, demonstrating consistent cytokine elevations across depressed populations, which may exacerbate poor glycemic control. Another critical pathway involves dysregulation of the hypothalamic-pituitary-adrenal (HPA) axis. Chronic stress in depression leads to hyperactivation of the HPA axis, resulting in prolonged cortisol release. Cortisol, a glucocorticoid hormone, induces gluconeogenesis and inhibits peripheral glucose uptake, fostering hyperglycemia and insulin resistance. Gold et al.[68] elucidated this mechanism, linking chronic hypercortisolemia in depression to increased adiposity and impaired glucose homeostasis. Compounding these challenges are socioeconomic and healthcare disparities. Depression and diabetes often coexist within vulnerable populations facing barriers to care, including financial constraints and limited access to mental health or diabetes-specific services. These factors exacerbate the bidirectional relationship between the two conditions, highlighting the importance of addressing systemic inequities in healthcare delivery.[69] However, when analyzing the proportion of poor glycemic control among depressed and nondepressed patients, no statistically significant association was observed (OR = 1.74). Even though studies by Crispín-Trebejo et al.[38] and Ellouze et al.[39] had demonstrated higher odds, their moderate sample size limited the weight of these studies. Behavioral mechanisms also play a role in mediating this relationship. Depression adversely impacts self-care behaviors critical for diabetes management, including dietary adherence, physical activity, and medication compliance. It impairs executive functioning and motivation, leading to suboptimal diabetes self-management. Lustman et al.[70] identified depressive symptoms as significant predictors of poor adherence to diabetes treatment regimens, which indirectly contribute to worsening glycemic outcomes. However, while examining medication adherence in this meta-analysis, depressed diabetes patients did not demonstrate significantly higher odds of poor adherence compared to nondepressed patients. The high heterogeneity observed renders the pooled estimates less reliable. These findings also underscore the variability in methodologies employed across studies to define and measure medication adherence.

The significant association between depression and elevated HbA1c levels underscores the importance of routine screening for depression, early detection, and timely interventions. Implementing evidence-based treatments, such as cognitive-behavioral therapy (CBT) or antidepressant medications, can play a crucial role in improving both mental health and glycemic control. Regional variations in outcomes further emphasize the necessity of personalized interventions tailored to cultural and healthcare contexts, such as leveraging telemedicine or community health workers in low-resource settings. Although the link between depression and medication adherence was not statistically significant, behavioral strategies like motivational interviewing or digital tools can still promote adherence and optimize self-management. These findings call for healthcare systems to prioritize integrated care models and innovative interventions that address both the psychosocial and clinical needs of this high-risk population to achieve better long-term outcomes.

Strengths and limitations

SRMA included studies from multiple databases and geographic regions, ensuring broad representativeness. The use of standardized tools, such as the PRISMA guidelines and registration under PROSPERO, enhances the transparency of the findings. The quality assessment was done using a standard tool (JBI and NOS tool). The substantial heterogeneity in several analyses (e.g., I² > 90% for glycemic control and medication adherence) limits the generalizability of the findings. The DOI plot indicated potential asymmetry for poor glycemic control, suggesting possible publication bias. Small-study effects may have also influenced the results. Differences in tools for assessing depression (e.g., PHQ-9, BDI) and outcomes (e.g., glycemic control thresholds, adherence definitions) complicate cross-study comparisons. The review did not include gray literature and included only English-language only. Therefore, some relevant articles may have been missing. Finally, since most of the studies included being cross-sectional, it limits the causality inference of the association.

CONCLUSION

The existing evidence suggests weak effect of depression on glycemic outcomes. However, the observed heterogeneity and methodological inconsistencies across studies underscore the need for standardized approaches and better study designs (prospective cohort studies) in future research. The impact of treatment status of depression on HbA1c also needs to be evaluated as a potential factor for heterogeneity, in future studies. Although statistically high HbA1c levels were observed in patients with depression compared to patients without depression, the certainty was very low and the difference could not reach a clinically meaningful range. Future studies, such as prospective cohort studies, should aim to unravel causal pathways, explore effective interventions, and include diverse populations to ensure global applicability. Advancing our understanding of biological and contextual factors will help improve outcomes for individuals with these comorbid conditions.

Data availability statement

All data used in the analysis are available in the manuscript or the supplementary files attached.

Conflicts of interest

There are no conflicts of interest.

Supplementary Table S4

List of excluded studies at full text review stage and the reason for exclusion

Figure S.1.A

Subgroup analysis of mean difference in HbA1c among depressed and non-depressed diabetic patients based on Continent

Figure S.1.B

Subgroup analysis of mean difference in HbA1c among depressed and non-depressed diabetic patients based on sample size

Figure S.1.C

Subgroup analysis of the mean difference in HbA1c among depressed and non-depressed diabetic patients based on the Tool used to assess depression

Figure S.2.A

Subgroup analysis of the mean difference in glycemic control in HbA1c among depressed and non-depressed diabetic patients based on the PHQ-9 cut off

Figure S.2.B

Subgroup analysis of the mean difference in glycemic control in HbA1c among depressed and non-depressed diabetic patients based on the CES cut off

Figure S.2.C

Subgroup analysis of the mean difference in glycemic control in HbA1c among depressed and non-depressed diabetic patients based on the GDS tool cut off

Figure S.2.D

Subgroup analysis of mean difference in HbA1c among depressed and non-depressed diabetic patients based on the BDI-I cut off

Figure S.2.E

Subgroup analysis of mean difference in HbA1c among depressed and non-depressed diabetic patients based on the BDI-II cut off

Figure S.3.A

Meta-regression analysis of the mean difference in HbA1c between depressed and non-depressed diabetic patients by duration of diabetes

Figure S.3.B

Meta-regression analysis of the mean difference in HbA1c between depressed and non-depressed diabetic patients by year of publication

Figure S.3.C

Meta-regression analysis of the mean difference in HbA1c between depressed and non-depressed diabetic patients by overall mean age

Figure S.4

Leave one out analysis of the mean difference in HbA1c between depressed and non-depressed diabetic patients

Figure S.5

Funnel plot for mean difference in HbA1c

Figure S.6

DOI Plot for mean difference in HbA1C

Figure S.7.A

Subgroup analysis of the proportion of poor glycemic control in HbA1c among depressed and non-depressed diabetic patients based on the Continent

Figure S.7.B

Subgroup analysis of the proportion of poor glycemic control in HbA1c among depressed and non-depressed diabetic patients based on the sample size

Figure S.7.C

Subgroup analysis of the proportion of poor glycemic control in HbA1c among depressed and non-depressed diabetic patients based on the Tool used to assess depression

Figure S.7.D

Subgroup analysis of the proportion of poor glycemic control in HbA1c among depressed and non-depressed diabetic patients based on the PHQ-9 cut off

Figure S.7.E

Subgroup analysis of the proportion of poor glycemic control in HbA1c among depressed and non-depressed diabetic patients based on HADS cut off scores

Figure S.7.F

Subgroup analysis of the proportion of poor glycemic control in HbA1c among depressed and non-depressed diabetic patients based on BDI cut off scores

Figure S.8.A

Meta-regression analysis of the proportion of poor glycemic control in HbA1c between depressed and nondepressed diabetic patients by year of publication

Figure S.8.B

Meta-regression analysis of the proportion of poor glycemic control in HbA1c between depressed and nondepressed diabetic patients by duration of diabetes

Figure S.8.C

Meta-regression analysis of the proportion of poor glycemic control in HbA1c between depressed and nondepressed diabetic patients by overall mean age

Figure S.9

Leave one out analysis of the proportion of poor glycemic control in HbA1c between depressed and non-depressed diabetic patients

Figure S.10

DOI Plot proportion of poor glycemic control

Figure S.11

Funnel plot for proportion of poor glycemic control studies

Figure S.12

Leave one out analysis for mean difference in FBS between depressed and non depressed diabetic patients

Figure S.13

Leave-one-out analysis for medication adherence among depressed and non-depressed diabetic patients

Funding Statement

Nil.