The association between diabetes mellitus and postoperative cognitive dysfunction: a systematic review and meta-analysis

Abstract

Background:

Postoperative cognitive dysfunction (POCD) is a typical consequence following surgery, particularly in cardiac surgeries. Despite its high incidence, the underlying etiology remains unclear. While diabetes mellitus (DM) has been associated with cognitive impairment, its specific function in POCD development remains unidentified. This study aims to evaluate the connection between DM and the risk of POCD.

Methods:

We conducted a comprehensive search of PubMed, Embase, Web of Science, and the Cochrane Library databases for studies of DM and risk with POCD, collecting data up to 14 September 2023. We assessed publication bias, heterogeneity, and study quality, adhering to PRISMA and AMSTAR guidelines.

Results:

Our study comprised 38 trials involving 8748 individuals, with 7734 patients undergoing follow-up. The pooled results showed that individuals with DM had an increased incidence of POCD compared to nondiabetic individuals (RR: 1.44, 95% CI: 1.26–1.65). The incidence of POCD was significantly higher in the group of patients with an average age older than 65 years (RR: 1.69, 95% CI: 1.30–2.20) compared with diabetic patients with an average age younger than 65 years (RR: 1.29, 95% CI: 1.09–1.64). Compared with diabetic patients undergoing cardiac surgery (RR: 1.33, 95% CI: 1.15–1.53), patients receiving non-cardiac surgery showed a greater incidence of POCD (RR: 2.01, 95% CI: 1.43–2.84).

Conclusion:

Current evidence underscores that diabetic patients face a significantly higher risk of POCD compared to their nondiabetic counterparts. Further research is warranted to clarify the precise mechanisms of this relationship and explore potential preventive strategies for diabetic patients.

Graphical abstract. The association between diabetes mellitus and postoperative cognitive dysfunction: A systematic review and meta-analysis. The graphical abstract was created with BioRender (www.BioRender.com)

Introduction

Highlights.

• A meta-analysis of 38 research involving 1790 diabetic patients indicated the DM’s (diabetes mellitus) impact on postoperative cognitive dysfunction (POCD) risk.

• Diabetic people are connected with a 44% higher risk of POCD than nondiabetic patients.

• The risk of POCD significantly increases by 40% compared to diabetic patients under the age of 65.

• Patients undergoing non-cardiac surgery also showed a similar increase in risk (68%), where advanced age might be the potential driving factor behind this outcome.

Postoperative cognitive dysfunction (POCD) is a common complication after surgery, characterized by a new onset of cognitive dysfunction following surgical intervention of any kind^[1]. It is particularly prominent in patients undergoing cardiac surgery^[2]. Reports indicate that the incidence of POCD in high-risk surgical populations ranges from 30% to 80%^[3]. POCD manifests as a series of changes in neurocognitive conditions and behaviors, including declines in learning and memory abilities, language skills, attention, and executive functions^[4]. In addition, POCD is closely associated with dementia, increased mortality, and decreased quality of life, posing a significant burden on the healthcare system^[5,6].

In the past decade, the global incidence of diabetes mellitus has significantly increased. Diabetes can elevate the risk of disability and complications affecting various organs, leading to a substantial economic burden^[7]. Cognitive dysfunction is a crucial comorbidity and complication of diabetes, potentially linked to diabetes-induced cerebrovascular and brain metabolic changes^[8]. A large cohort study has found an independent association between diabetes and accelerated cognitive decline^[9]. Furthermore, elevated blood sugar levels in individuals without diabetes are also associated with an increased risk of dementia^[10]. Increasing evidence suggests that higher A1C levels are associated with diabetes-related cognitive decline^[11]. Additionally, dose-response curves show a nonlinear positive correlation between FPG and the risk of cognitive impairment^[12]. A recent prospective cohort study also demonstrated that patients with type 2 diabetes mellitus (T2DM) have an increased risk of perioperative neurocognitive disorders within 9 months following elective non-cardiac surgery under general anesthesia^[13].

Although the etiology of POCD is not fully understood, the inflammatory process is currently believed to be central to its occurrence. There is no clear evidence that anesthetic and surgical components affect POCD. However, some patient-related factors, such as advanced age, are associated with an increased risk of cognitive decline^[14-16]. Some studies have shown that the diabetic state increases the risk of POCD (RR: 1.26, 95% CI: 1.12–1.42)^[17]. Meanwhile, similar studies have not found this correlation^[18]. However, the relationship between DM and POCD is not clear^[19]. Therefore, this study aims to systematically review the available epidemiological evidence and conduct a meta-analysis on the relationship between diabetes and the risk of POCD.

Methods

Study design

This study follows the recommendations of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) (Supplementary Digital Content, Table S1, http://links.lww.com/JS9/D593) and the Assessing the Methodological Quality of Systematic Reviews (AMSTAR)^[20,21]. The protocol for this meta-analysis has been registered with PROSPERO.

Information sources and search strategy

A comprehensive search was conducted in the PubMed, Embase, Web of Science, and Cochrane Library databases, covering the period from inception to 14 September 2023. The search strategy was formulated based on a combination of keywords and mesh terms related to diabetes and POCD. Detailed information, including the specific search terms used, can be found in Supplementary Digital Content, Table S2 (http://links.lww.com/JS9/D593). Additionally, a manual search was conducted through the reference lists of eligible articles to identify any potentially missed studies.

Study selection and inclusion criteria

The studies included in this research were eligible based on the PICOS principles (Population, Intervention/Exposure, Comparator, Outcome, and Study Design). Articles were included in the review if a full-text assessment determined that the following criteria were met: (1) adult patients (aged ≥ 18) undergoing surgery under general or local anesthesia, (2) diagnosed as diabetes on admission, (3) surgical patients without diabetes, (4) reporting on the occurrence of postoperative events in POCD or reporting their correlation with cognitive changes in a standardized form, such as RR, OR, or HR, and the corresponding 95%CI (5) observational study (i.e., cohort study, case-control study, or cross-sectional study) and available randomized clinical trials. Since POCD is defined as a decline in cognitive test scores pre- and postoperatively and lacks a standardized definition^[22], there is no strict requirement to use the term “POCD” when screening identified literature.

The exclusion criteria for studies were as follows: (1) reviews, conference abstracts, letters, consensus statements and guidelines, case reports, or animal studies; (2) non-English studies; (3) studies where insufficient data details could be extracted; (4) studies focusing solely on POD; (5) studies with entirely negative results.; (6) duplicate reports of the same study. Two authors independently conducted an initial screening of titles and abstracts. To ensure the accuracy of the included studies, the research team resolved any disagreements encountered during the process through group discussion. Although both delirium and POCD are common postoperative neurocognitive complications, they differ in their underlying mechanisms, clinical manifestations, and duration^[14]. This study focuses on POCD to clarify the long-term association between diabetes and POCD without the confounding influence of delirium.

Data extraction

The independently extracted data is presented in standardized tables. The extracted data included the following information: first author, publication year, type of study design, geographical region, type of surgery, type of anesthesia, sample size, number of patients who completed follow-up, mean age, percentage of female participants, follow-up duration, method of neurocognitive function assessment, incidence of POCD in patients with diabetes, proportion of exposure, degree of cognitive impairment, definition of POCD, associations of exposure with POCD (RR) and cognitive outcomes, and adjustment variables, in cases where the study reported results from multiple follow-up periods, the most prolonged follow-up duration was preferentially extracted. Alternatively, (odds risk) extracted from the studies were treated as RR or converted to RR format in this review^[23]. The study with more complete data was included for articles with potential duplicate reporting. Data were obtained from multivariate-adjusted models unless unadjusted data were available. We emailed the author to obtain the missing critical information in the article but did not receive a response. The accuracy of the data was cross-checked, and any discrepancies were resolved through group discussion. Given the diversity in the assessment methods and definitions of neurocognitive function for POCD, extracting neurocognitive function scores before and after surgery posed significant challenges. Therefore, we did not effectively extract cognitive function scores.

Risk of bias assessment

Two authors independently assessed the risk of bias in randomized clinical trials using Cochrane Collaboration’s tool^[24]. For cross-sectional studies, we adopted the Agency for Healthcare Research and Quality (AHRQ) criteria for assessing the risk of bias in observational research^[25]. The total score for each study ranges from 0 to 11, with scores of 0-7 indicating a low risk of bias and 8-11 indicating a high risk of bias. However, due to considerations related to establishing control groups in some observational studies, we chose a more reasonable tool for assessment: the Methodological Index for Non-Randomized Studies (MINORS)^[26] and the Newcastle-Ottawa Scale (NOS)^[27]. The MINORS tool comprises 12 items, with four additional items for comparative studies. The ideal scores for comparative and non-comparative studies on the MINORS scale are 24 and 16, respectively. The NOS evaluates population selection, comparability between groups, and outcome measurement. The total score is 9, and studies with an overall NOS quality score of 6 or above are considered high-quality, while those scoring below 6 are classified as low-quality.

Data synthesis and statistical analysis

The meta-analysis used Review Manager software (version 5.4.1, Cochrane Collaboration) and StataSE 12.0 (College Station, TX, USA). Forest plots for subgroup analyses were generated using the “forestplot” package in R (version 4.2.3) and RStudio (R Foundation for Statistical Computing, Vienna, Austria, 2023.12.1 + 402). RR and 95% CI for dichotomous outcomes were calculated using the inverse-variance method. Heterogeneity among studies was assessed using the Q statistic. Given the potential heterogeneity arising from demographic and measurement differences, a random-effects model was employed to pool the data. A two-sided P-value < 0.05 was considered statistically significant.

Visual inspection of funnel plots was conducted to identify any potential publication bias, which was further evaluated using Egger’s test^[28] and Begg’s test^[29]. The presence of publication bias was assumed if asymmetry was observed in the funnel plot or if either test (or both) yielded a statistically significant bias coefficient. In cases where different tests yielded conflicting results, the more powerful Egger’s test findings were prioritized. If publication bias was detected, trim and fill methods were applied to detect and correct for the bias.

Pre-specified subgroup analyses were subsequently conducted to assess potential heterogeneity among studies and further investigate the impact of these factors on the association between diabetes and POCD. Subgroups included: (1) follow-up duration, (2) study quality, (3) sample age, (4) sample size, (5) male proportion, (6) geographical region, (7) type of surgery (cardiac surgery, non-cardiac surgery, mixed); (8) anesthetic technique and (9) study design.

Meta-regression was performed concurrently. Finally, a meticulous sensitivity analysis was conducted to assess the robustness of the meta-analysis results. The certainty of evidence was evaluated using the GRADE (Grades of Recommendation, Assessment, Development, and Evaluation) system^[30].

Result

Study selection

The study screening and selection process flowchart is as follows (Fig. 1). Initially, a total of 5760 articles were registered. Before screening, duplicate records (n = 218), non-English articles (n = 874), case reports, letters, clinical guidelines, conference proceedings (n = 241), meta-analyses, and review articles (n = 284), animal experiments (n = 39), and records excluded for other reasons (n = 50) were removed, leaving 4054 articles.

Figure 1.

Flowchart of the study selection process of the meta-analysis between diabetes mellitus (DM) and postoperative cognitive dysfunction (POCD).

At the title and abstract level, 546 studies were screened. A total of 508 studies were excluded for not meeting our inclusion criteria, including 266 that did not address the research question. Among these, 171 studies focused on postoperative delirium (POD), 1 reported exclusively negative results, 69 provided insufficient data, and 1 was a suspected duplicate report. Ultimately, 38 articles were included in this review.

Study characteristics

The detailed clinical characteristics of the included studies are summarized in Table 1.

Table 1.

Summary of included studies on diabetes status and POCD.

Author, year of study	Sample size, Follow-up	Participants,Region (Enrolled/Completed)	Study design and anesthesia	Type of surgery	Mean age (y), female (%)	DM proportion, POCD in DM	Definition of POCD	Neurocognitive Assessment	Degree of impairment	Measures of Association (95% CI)	Adjustment
Yong and Jiang, 2023 [37]	312,1 mo	312/312, Asia	Prospective Cohort	Laparoscopic radical gastrectomy; General anesthesia.	73,30	9.6%, 37%	POCD defined as the composite Z-score ≥ 1.96 or Z-score ≥ 1.96 on at least two tests.	MMSE;VFT;DSST; DSF and DSB; TMT-A.	None.	RR 2.05 (95%CI 0.93 - 4.55)	None.
Wang et al., 2023 [53]	154,1 mo	188/154, Asia	Prospective Cohort	Thoracic surgery; General anesthesia.	68,56	19%, 30%	POCD defined as the t-MoCA score decreased > 1.96 points at 1 mo postoperatively.	MoCA.	None.	RR 6.51 (95%CI 2.05 - 20.66)	History of cerebro-vascular disease;Previous major surgery history; Intravenous inhalation combined;anesthesia; Perioperative hypothermia.
Van Zuylen et al., 2023 [31]	102,6 mo	162/102,Western	Prospective Cohort	Cardiac surgery, nocardiac surgery; Mixed-anesthesia.	72,48	22%, 9%	Mild POCD defined as RCI decrease by 1 to 2 SD and major POCD defined as RCI decrease by≥ 2 SD.	TICS-M.	4 patients (3.9%) were ranked as major POCD, 10 (9.8%) as mild POCD.	RR 0.94 (95%CI 0.19 - 4.76)	None.
Pu et al., 2023 [27]	114,3 mo	114/114, Asia	Retrospective Cohort	Neurointervention and intravenous thrombolysis; NA..	NA.,41	45%, 53%	POCD defined as the MMSE score ≤ 27.	MMSE.	None.	RR 5.906 (95%CI 2.00 - 17.44)	Age;Time from to admission;Hyperlipidemia; Lesions at critical site.
Oyoshi et al., 2023 [51]	71,1 w	110/71, Asia	Retrospective Cohort	CPB,Valve repair and replacement; General anesthesia.	58,32	45%, 38%	POCD defined as a decrease of 1 SD population means in at least two out of six variables in the test battery.	DSF and DSB; DST2;TMT-A and B; Kana Pick-out test.	None.	RR 1.20 (95%CI 0.45 - 3.19)	None.
Luo et al., 2022 [58]	222,3 d	229/222, Asia	Prospective Cohort	TURP; Intrathecal anesthesia.	73,0	14%, 27%	POCD defined as the |Z-score| > 1.96.	MMSE.	None.	RR 2.55 (95%CI 1.02 - 6.36)	None.
Herman et al., 2023 [33]	99,9 mo	115/99,Europe	Prospective Cohort	Percutaneous pulmonary vein isolation,hybrid ablation for atrial fibrillation; Regional anesthesia.	64,40	24%, NA.	POCD defined as an individual RCI score ≤ -1.96.	TMT-A and B; BSF and BSB; DSF and DSB; Design fluency test;COWAT; Stroop test.	11 patients (11%) as major POCD, 11 (11%) as minor POCD and 13 (13%) as combined POCD.	RR 1.66 (95%CI 0.43 - 6.30)	None.
Florido-Santiago et al., 2023 [47]	41,12 mo	44/41,Europe	Prospective Cohort	CABG;AVR NA..	74;25	85%, NA.	POCD defined as a percentile rank equal or ≤ 10% (a score deterioration in the performance ≥1.5 SD) of any test.	TMT;SCWIT;FCSRT; JLOT;Stroop Test; SVFT and PVFT.	None.	RR 1.7 (95%CI 1.1 - 3.1)	Age;History of smoking; Arterial Hypertension; Heart failure;Cognitive dysfunction;EuroSCorE; Operation time.
Scrimgeour et al., 2022 [54]	28,4 d	30/28,Western	Prospective Cohort	CABG,Elective or urgent valvular procedure; General anesthesia.	67,27	54%, 47%	POCD defined as a decrease from baseline RBANS score.	RBANS.	None.	RR 1.40 (95%CI 0.31 - 6.33)	None.
Ren et al., 2022 [34]	103,3 d	103/103,Asia	Retrospective Cohort	TKA; General anesthesia.	67,68	11%, 27%	POCD defined as a as a postoperative MOCA score < 26.	MOCA.	40 patients (39%) were ranked as major POCD,33 (32%) as mild POCD.	RR 1.11 (95%CI 0.27 - 4.50)	None.
Zhang et al., 2021 [46]	190,2 d	190/190,Asia	Prospective Cohort	Ablation for atrial fibrillation; Regional anesthesia.	67,41	17%, 9%	POCD defined as the Z-score< -1.96 on≥2 tests or the combined z-score<-1.96.	COWAT;GPND;DSST: GPD;CERAD Auditory-Verbal Learning;Semantic Fluency Test.	None.	RR 0.58 (95%CI 0.16 - 2.07)	None.
Kadoi and Goto, 2006 [30]	88,6 mo	95/88,Asia	Prospective Cohort	CABG,CPB; General anesthesia.	62,20	24%,71%	NA.	MMSE;RAVLT;DST1; TMAT-A and B.	None.	RR 1.8 (95%CI 1.2 - 2.4)	None.
Heyer et al., 2015 [6]	585,1 d	756/585,Western	Prospective Cohort	CEA; General anesthesia.	NA.,35	21%, 23%	POCD defined as the ≥ 2 SD worse performance in 2 or more cognitive domains or ≥1.5 SD worse performance in all 4 cognitive domains.	COWAT;GP;FTT;RCFT; HVLT Halstead-Reitan Trials A and B.	None.	RR 0.90 (95%CI 0.56 - 1.43)	None.
Pérez-Belmonte et al.,2015 [50]	36,12 mo	36/36,Western	Prospective Cohort	CABG. NA..	66,31	53%, NA.	POCD defined as percentile ranks of scores ≤18% on at least 1 test.	TMT-A and B;FCSRT; Stroop Test;VFT.	None.	RR 1.7 (95%CI 1.1 - 2.5)	None.
Halazun et al., 2014 [28]	432,1 d	551/432,Western	Retrospective Cohort	CEA; General anesthesia.	NA.,36	21%, NA.	POCD defined as a decline performance in ≥2 cognitive domains or ≥1.5 SD worse performance in all 4 cognitive domains.	COWAT;GP;HVLT;FTT; RCFT;SCSRT; Halstead-Reitan Trials A and B.	None.	RR 2.03 (95%CI 1.08 - 3.75)	Sex; statin use; NLR ≥5;y of education;
Joudi et al., 2014 [36]	171,1 d	171/171	Prospective Cohort	Off-pump open-heart surgery; General anesthesia.	64,NA.	34%, 79%	NA.	MMSE	None.	RR 1.39 (95%CI 0.65, 2.96)	None.
Mathew et al., 2009 [60]	182,6 w	241/182,Western	RCT	CABG,open chamber procedure with CPB; General anesthesia.	62,30	20%, 44%	POCD defined as the ≥1 SD or more in at least 1 of the 4 domains.	RANDT;TMT-B;DST1 DSF and DSB.	None.	RR 0.94 (95%CI 0.45 - 1.96)	None.
Heyer et al., 2005 [56]	75,1 mo	75/75,Western	Prospective Cohort	CEA; General anesthesia.	69,39	25%,26%	POCD defined as a decline of ≥ 2 SD in a total deficit score compared with control group.	BNT;COWAT;Halstead-Reitan;Trails parts A and B;The copy portion of the RCFT.	None.	RR 51.42 (95%CI 1.94 - 1363)	Age;Obesity;. APOE-ε4+
H. Zhang et al., 2019 [48]	287,1 w	287/287,Asia	Prospective Cohort	Hip replacement; Intrathecal anesthesia.	70,59	13%, 23%	POCD defined as the Z-score ≥ 1.96 on at least two tests.	DST1;DST2;MMSE;VFT; Word recognition memory tests;TMA-A.	None.	RR 1.22 (95%CI 0.58 - 2.51)	Age;CRP; pNF-H positivity. Preoperative MMSE;
Wu et al., 2019 [39]	198,1 w	198/198,Asia	Prospective Cohort	Hip fracture surgery; Mixed-anesthesia.	72,60	16%, 40%	POCD defined as the composite Z-score ≥ 1.96 or Z-score ≥ 1.96 on at least two tests.	MMSE;VFT;DST1; TMT-A;Symbol digit test;Word recognition memory tests.	None.	RR 1.05 (95%CI 0.95 - 1.17)	Age;Type of anesthesia; CRP on POD1; MDA on POD1; SOD on POD1.
Scott et al., 2018 [40]	389,3 mo	437/389,Australia	Prospective Cohort	LHC Regional anesthesia.	61,43	24%a, NA.	POCD defined as the composite Z-score ≤ -1.96 or an individual showed cognitive decline on at least two tests.	Maze Test; Computerized playing cards Test; One-back test.	None.	RR 2.31 (95%CI 1.09 - 4.90)	Prior stent;APOE-ε4+; MCI multidomain;BMI.
Soenarto et al., 2018 [29]	54,5 d	54/54,Asia	Prospective Cohort	Cardiac surgery; General anesthesia.	NA.,30	39%, 52%	POCD defined as the 20% decrease in at least one of the cognitive function tests.	RAVLT;DSF and DSB	None.	RR 2.20 (95%CI 0.72 - 6.75)	None.
Ziyaeifard et al., 2017 [35]	99,3 d	99/99,Asia	Prospective Cohort	CABG; Valve repair or replacement and elective operation; NA..	53,59	20%, 45%	Mild POCD defined as the score is 20 - 24, moderate cognitive dysfunction is 11 - 19, severe cognitive dysfunction is 0 - 10.	MMSE	0 patients (0%) were ranked as severe POCD, 9 (45%) as moderate POCD and 11 (55%) as no POCD.	RR 1.03 (95%CI 0.38 - 2.76)	None.
Mocco et al., 2006 [55]	153,1 mo	186/153,Western	Prospective Cohort	CEA; General anesthesia.	70,32	25%,21%	POCD defined as a decline of ≥ 2 SD in performance compared with a similarly aged control group.	COWAT;BNT;Halstead-Reitan;Trails Parts A and B;The copy portion of the RCFT.	None.	RR 4.26 (95%CI 1.15 - 15.79)	Age;BMI ≥ 30; Midazolam.
Xiao et al., 2021 [45]	450,1 w	463/450,Asia	Prospective Cohort	Valve replacement, valvuloplasty, ventricular and atrial septal defect repair, CABG; General anesthesia.	51,58	18%, 35%	POCD defined as two scores in individual tests or the combined Z score ≥ 1.96.	MMSE;SDMT;DST; TMT-A;CDT.	None.	RR 1.81 (95%CI 1.07 - 3.03)	None.
Soenarto et al., 2021 [49]	60,5 d	70/60,Asia	Prospective Cohort	Open-heart surgery; General anesthesia.	54,37	35%, 62%	POCD defined as a decrease of 20% or more was determined on ≥2 cognitive tests.	RAVLT;DST1; TMT-A and B.	None.	RR 1.90 (95%CI 0.64 - 5.60)	None.
Ngcobo et al., 2020 [59]	28,6 w	28/28,Africa	Cross-sectional survey	CABG; NA..	59,25	46%, 62%	POCD defined as a MOCA score of ≤ 25.	MOCA.	None.	RR 1.07 (95%CI 0.23 - 4.89)	None.
Y. Zhang et al., 2019 [38]	77,1 w	80/77,Asia	Prospective Cohort	Colorectal surgery; General anesthesia.	70,45	10%, 88%	POCD defined as the Z-score ≥ 1.96 or composite Z-score ≥ 1.96.	MMSE;VVLT; DST1;DST2.	None.	RR 8.391 (95%CI 2.21 - 31.88)	Gender, body mass index, tumor location and stage, hypertension, hemoglobin, hematocrit, total protein, serum albumin, days in hospital waiting for surgery, operation time, bleeding, transfusion, ΔHb, urine, hypotension, duration of Intensive Care Unit stay, and in-hospital stay.
Toeg et al., 2013 [61]	652,3 mo	696/652,Western	Prospective Cohort	CABG; NA..	64,12	31%, 21%	POCD defined as the ≥ 1 SD worse in more than 1 domain of interest.	SCSRT;TMT-A and B; Letter and Category Fluency;SDMT;FTT.	None.	RR 1.95 (95%CI 1.39 - 2.73)	CPB time;Preop creatinine;
Medi et al., 2013 [44]	120,3 mo	120/120	Prospective Cohort	RFA; General anesthesia.	56,28	8%, NA.	POCD defined as the RCI score ≤ 1.96 on at least 2 tests and/or the combined z-score ≤ 1.96.	RAVLT;TMT-A and B; DSST;COWAT;SVFT;GP.	None.	RR 0.15 (95%CI 0.03 - 0.79)	None.
Stewart et al., 2013 [52]	155,12 mo	155/155,Western	Prospective Cohort	CABG; NA..	67,10	30%, 28%	POCD defined as the ≥ 1 SD worse in Z-score.	RAVLT;RVDLT;TMT-A and B;GP;DSF and DSB; VFT;Choice and Simple Reaction Time Tests.	None.	RR 1.91 (95%CI 0.85 - 4.32)	None.
Evered et al., 2011 [41]	443,1 w	443/443,Australia	Prospective Cohort	CA,CABG,THJR; Mixed-anesthesia.	68,37	21%a, NA.	POCD defined as the RCI < ‐1.96 on ≥2 tests and/or composite RCI < ‐1.96.	WLT;TMT-A and B; COWAT;CERAD; GPD;GPND;DSST.	None.	RR 1.52 (95%CI 0.89 - 2.59).	Age;BMI;β-blockers;Statins; Hypercholesterolemia;Peri- pheralvascular diseas;History acute myocardial infarction; THJR versus CABG..
Gaudet et al., 2010[57]	64,1 d	73/64,Western	Prospective Cohort	CAE; General anesthesia.	72,27	17%, 27%	POCD defined as the an average z-score of ≤ -1.5 worse than the control group.	COWAT;GP;BNT;RCFT Halstead-Reitan Trails A and B.	None.	RR 1.83 (95%CI 0.41 - 8.28)	None.
McDonagh et al., 2010 [62]	350,6 w	394/350,Western	Prospective Cohort	Vascular, thoracic, major orthopedic surgery; Mixed-anesthesia.	68,50	16%a, NA.	POCD defined as a decline of ≥1SD on ≥ 1 of 4 cognitive domains.	RANDT;TMT-B;FIGM; DST1 and DST2.	None.	RR 2.34 (95% CI 1.22 - 4.51).	Age;y of education; Preoperative cognitive index; orthopedic surgery;Female sex;General anesthesia;APOE4.
Dieleman et al., 2009 [42]	240,5 y	281/240,Western	ProspectiveObservational	CABG; General anesthesia.	61, 28	13%, 45%	POCD defined as the composite RCI ≤ ‐1.96 or RCI ≤ ‐1.96 in at least cognitive tests,a stroke or a developed severe dementia.	RAVLT;GP;TMT-A and B; SDMT;JLOT;Sternberg memory comparison; Continuous performance task;SCWIT;Self-ordering tasks;Visuospatial working memory.	None.	RR 1.71 (95%CI 0.80 - 3.67)	None.
Puskas et al., 2007 [63]	525,6 w	703/525,Western	Prospective Cohort	CABG,CPB; General anesthesia.	61,28	28%,40%	POCD defined as a decline of ≥ 1 SD calculated from the baseline scores in at Leas tone of the four cognitive domains.	RANDT;DST1 and DST2; TMT-A and B;FIGM.	None.	RR 1.14 (95%CI 0.77 - 1.69)	None.
Di Carlo et al., 2001 [32]	110,6 mo	123/110,Western	Prospective Cohort	CABG,Intracardiac surgery; General anesthesia.	64,29	21%,22%	NA.	MMSE;RMT;Token Test;Test of abstract. thinking.	10 patients (9.1%) were ranked as severely deteriorated, 22 (20%) as mildly– Moderately deteriorated, and 78 (70.9%) as unchanged–improved.	RR 0.62 (95%CI 0.21 - 1.84)	None.
Liu et al., 2009 [43]	169,3 mo	227/169,Asia	Prospective Cohort	CABG; General anesthesia.	61,10	31%,NA.	POCD defined as two scores in individual tests or the combined Z score were 1.96 or more.	GP;DSF and DSB;DST2; TMT-A;Paired-associate verbal learning;Visual retention Test;Mental control Test.	None,	RR 3.024 (95% CI 1.04 - 8.791)	Age; MI history; CPB use; Cerebral microemboli; Ventilation duration in ICU; Post-Operative hospital stay

Abbreviations: AVR, Aortic valve replacement; BMI, Body mass index; BNT, Boston Naming Test; BSF/BSB, Block Span forward; CA, Coronary angiography; CABG, Coronary artery bypass graft; CDT, Clock Drawing Test; CEA, Carotid endarterectomy; CERAD, Consortium to Establish a Registry for Alzheimer’s Disease; COWAT, Controlled Oral Word Association Test; CPB, Cardiopulmonary bypass; CRP, C-reactive protein; DSF/DSB, Digit span forward and backward test; DSST, Digit symbol substitution test; DST1, Digital span test; DST2, Digital symbol test; FCSRT, Free and Cued Selective Reminding Test; FIGM, Figural memory from Modified Visual Reproduction Test in the Wechsler Memory Scale; FTT, Finger Tapping Test; GP, Grooved pegboard; GPD, Grooved Pegboard Test Dominant; GPND, Grooved Pegboard Test Nondominant; HVLT, Hopkins Verbal Learning Test; JLOT, Judgment of Line Orientation Test; LHC, Left heart catheterization; MCI, Mild cognitive impairment; MDA, Malondialdehyde; MI, Myocardial infarction; MMSE, Mini-mental state examination; MoCA, Montreal Cognitive Assessment; NIHSS, National Institutes of Health Stroke Scale; PVFT, Phonological verbal Tests; RANDT, Randt Short Story Memory Test; RAVLT, Rey Auditory Verbal Learning Test; RBANS, Repeatable Battery Assessment of Neuropsychological Status; RCFT, Rey Complex Figure test; RCI, Reliable change index; RFA, Radiofrequency ablation; RR, Relative risk; RVDLT, Rey Visual Design Learning Test; SCWIT, Stroop Color-Word Interference Test; SD, Standard deviation; SDMT, Symbol Digit Modalities Test; SOD, Superoxide dismutase; SVFT, Semantic verbal fluency test;; THJR, Total hip joint replacement surgery; TICS-M, Modified telephone interview for cognitive status; TKA, Total Knee Arthroplasty; TMT-A/TMT-B, Trail making test-part A and B; TURP, Transurethral resection of prostate; VFT, Verbal fluency test; VVLT, Visual verbal learning test; WLT, Word Learning Test. ^aData from the study is based on the sample population (The sample information for completed follow-up is unclear.

The publication dates span from 2001 to 2023, with studies originating from countries in Europe, Asia, Africa, and Australia. The number of patients in each study ranges from 28 to 652. Except for 4 studies where the mean age is unknown^[6,31-33], the average age of the samples ranged from 51 to 74 years (with an overall average of 65 years), and the proportion of female patients ranged from 0% to 68%.

Among the 38 studies, there is 1 cross-sectional study^[34], 34 cohort studies^{[6,31-33,35-64]}, 2 case-control studies^[65,66], and 1 RCT^[67]. 22 studies focused on cardiac surgery^{[33-35,37,39-41,45-51,55-57,59-61,65,67]}, 14 on non-cardiac surgery^{[6,31,32,36,38,42,43,52-54,58,63,64,66]}, and 2 on mixed surgeries^[44,62]. All patients underwent baseline cognitive assessment before surgery and were reassessed postoperatively during follow-up. The follow-up duration ranged from 1 day to 5 years, with 11 studies having a follow-up period of less than 1 week^{[6,32,33,42,48,50,55,57-59,66]}. All 38 studies investigated POCD based on diabetes status, with only 2 studies reporting detailed information on diabetes type^[37,62], and 5 studies classifying the severity of cognitive dysfunction^{[35,50,61,62,66]}. N = 8639 patients were included, of which N = 7628 were retained for follow-up. Among these studies, the total number of patients with diabetes was 1790. Nine studies did not mention the proportion of patients who developed POCD during follow-up, while the reported rates in other studies ranged from 9% to 88%.

Three studies did not report the diagnostic rule itself^[35,37,48]. The 18 studies were based on the composite scores calculated from all cognitive tests in the battery or a combination of several of them^{[33,40,41,44-46,49,51-57,60,61,64,65]}. Seven studies relied on simple analytical methods, with changes compared with baseline in patients or study populations without considering natural variability and learning effects^{[36,38,42,50,58,59,63]}. Four deviations from the population norm were used to define relevant changes^{[31,34,62,66]}. Component analysis of decline based on one or more predefined cognitive domains in six studies^{[6,32,39,43,47,67]}. Seven studies used cognitive screening tools, including 4 studies that used the Mini-Mental State Examination (MMSE) to assess cognitive function, 3 studies that used the Montreal Cognitive Assessment (MoCA), and the remaining studies employed detailed neuropsychological testing. Notably, only 14 studies had covariates that were adjusted, extracted, or available for analysis.

Diabetes and risk of POCD

Of the 38 studies included, 15 demonstrated a significant association between diabetes and POCD, while the remaining 23 studies did not show a significant correlation. The pooled analysis (Fig. 2) revealed that patients with diabetes had a 44% increased risk of developing POCD compared to nondiabetic patients (RR: 1.44, 95% CI: 1.26–1.65), and this difference was statistically significant (P < 0.001). Moreover, a notable heterogeneity is observed in studies investigating the association between two distinct patient cohorts and the development of POCD.

Figure 2.

Forest plot for fixed‐effects meta‐analysis of diabetes and risk of POCD.

Risk of bias assessment

The quality and risk of bias were evaluated for the included studies, comprising 34 cohort studies (Tables S3-S5, Supplementary Digital Content, http://links.lww.com/JS9/D593), of which 5^{[39,53,55,59,63]} were deemed low quality, and the rest were classified as high quality. Within the cohort studies, 15^{[6,32,33,42,44,48,50,52-59]} exhibited minor deficiencies in follow-up time, whereas 11^{[6,32,38-41,43,55,61-63]} showed substantial loss to follow-up. Regarding the case-control studies (Table S7, Supplementary Digital Content, http://links.lww.com/JS9/D593), 2^[65,66] were assessed as high quality. Similarly, the cross-sectional study was rated high quality (Table S6, Supplementary Digital Content, http://links.lww.com/JS9/D593). The sole RCT demonstrated notable attrition bias attributed to loss to follow-up, while the remaining studies did not reveal significant bias risks, thus suggesting superior quality.

Publication bias

Visual inspection of the funnel plot revealed a relatively small publication bias (Fig. 3). This was further supported by the Egger test (P = 0.001) and Begg test (P = 0.269). Given the higher power of the Egger test, it was considered strong evidence for publication bias. Therefore, the trim and fill method assessed the number of missing studies (Supplementary Digital Content, Figure S1, http://links.lww.com/JS9/D593). By adding virtual studies and reassessing the pooled RR, we aimed to correct for any potential bias.

Figure 3.

Funnel plot for meta‐analysis of diabetes and risk of POCD.

After the final re-pooling, it was observed that after four iterations, three missing studies were added. However, the correlation between diabetes and the risk of POCD did not change significantly. The measured effect size did not significantly change after trimming and filling, suggesting that publication bias had no substantial impact on the results, further indicating that the findings are robust and stable.

In conclusion, although publication bias was detected, it did not significantly alter the association between diabetes and the occurrence of POCD. The trim and fill method provided a more comprehensive and unbiased estimate of the actual effect, confirming the stability and reliability of our findings.

Subgroup analysis, meta-regression analysis, and sensitive analysis

We conducted subgroup analyses to investigate the sources of heterogeneity (Fig. 4). Detailed results of the subgroup analyses can be found in Supplementary Digital Content, Figure S6–14 (http://links.lww.com/JS9/D593). In the included studies, four studies^[6,31-33] omitted age information for diabetic and nondiabetic patients. Subgroup analysis by age reaffirms age as a risk factor for POCD, with no significant age difference observed between the two groups overall (RR: 1.42, 95% CI: 1.24–1.64, P = 0.09). Follow-up time subgroup analysis reiterates diabetes as a risk factor for POCD. Although heterogeneity decreased within subgroups, it persists. Notably, non-cardiac surgeries displayed a significant effect (RR: 2.01, 95% CI: 1.43–2.84, P < 0.001), while cardiac surgeries showed a moderate effect (RR: 1.33, 95% CI: 1.15–1.53, P < 0.001). Moreover, studies in Asia (RR: 1.51, 95% CI: 1.23–1.84, P < 0) exhibited a more robust effect compared to non-Asian regions (RR: 1.40, 95% CI: 1.51–1.69, P < 0.001). Supplementary Digital Content, Table S8 (http://links.lww.com/JS9/D593) presents a subgroup analysis of other study characteristics. Regrettably, no significant impact of predefined factors on subgroup heterogeneity was identified.

Figure 4.

Diabetes mellitus (DM) and risk of postoperative cognitive dysfunction (POCD) in subgroup analyses. All results are from random effect models. (*Indicates that some articles have incomplete data).

To further explore possible sources of heterogeneity, we conducted a meta-regression analysis encompassing nine covariates covered in the subgroup analysis: sample age, follow-up duration, study quality, sample size, male proportion, region, type of surgery, anesthesia method, and study design. The summary results can be found in Supplementary Digital Content, Table S8 (http://links.lww.com/JS9/D593). Again, no significant influencing factors for heterogeneity were identified. Finally, the sensitivity analysis was performed by excluding each study individually, and the combined effect did not change significantly, demonstrating the stability and reliability of our results (Supplementary Digital Content, Figure S2-S5 http://links.lww.com/JS9/D593). The quality of evidence ranges from very low to moderate. Table 2 summarizes individual studies’ quality, bias assessment, and evidence strength.

Table 2.

Grading of recommendations assessment, development, and evaluation assessment.

Quality assessment							No. of patients		Effect		Quality	Importance
No of studies	Design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Cognitive decline	Control	Relative (95% CI)	Absolute
Cross-sectional study
1	Observational studies	Very seriousa	No serious inconsistency	Seriousb	Very seriousc	None	–	–	RR 1.03 (0.57 to 1.86)	–	ÅOOO VERY LOW	Critical
								0%		–
Cohort study
35	Observational studies	No serious risk of bias	No serious inconsistency	No serious indirectness	No serious imprecision	Reporting biasd strong associatione	–	–	RR 1.55 (1.33 to 1.82)	–	ÅÅOO LOW	Critical
								0%		–
RCT
1	Randomized trials	Seriousf	No serious inconsistency	No serious indirectness	No serious imprecision	None	–	-	RR 0.97 (0.65 to 1.45)	–	ÅÅÅOMODERATE	Critical
								0%		–
Case-control study
2	Observational studiesg	Seriousa	No serious inconsistency	No serious indirectness	Serioush	Reporting biasi	–		OR 1.17(0.53 to 2.61)	–	ÅOOO VERY LOW	Critical
								0%		–

^aThere are potential confounding biases, small sample sizes, and deviations from established biases caused by research design.

^bDiseases and factors coexist, and the exact causal relationship cannot be determined.

^cThe sample size is small, and the merged evidence has great inaccuracy.

^dEgger’s test may indicate publication bias.

^eEleven studies have shown that RR values are greater than 2.

^fData loss caused by missing visits.

^gCase-control.

^hWide confidence interval.

ⁱNo explanation was provided.

Discussion

Major findings

This meta-analysis provides an assessment of the relationship between diabetes and the risk of POCD among surgical patients. The study included 7734 surgical patients, with follow-up durations ranging from one day to five years. The results indicate that, overall, people with diabetes have a 44% increased risk of developing POCD compared to nondiabetics. Compared to a previous review by Feinkohl et al^[17] on the same topic, our review reports consistent results with a slightly higher risk estimate. The lack of extensive, robust, randomized clinical trials precludes definitive inference of a causal link between the two variables from existing epidemiological evidence. Nonetheless, the clinical implications of our study emphasize the necessity of proactive diabetes screening preoperatively and continuous monitoring of cognitive function changes in diabetic patients. Our results imply the increased vulnerability of diabetic individuals to perioperative and anesthetic influences.

Comparison with previous research

In comparison with previous similar studies (RR: 1.26, 95% CI: 1.12–1.42, P < 0.001), as mentioned above, our meta-analysis yielded consistent results (RR: 1.44, 95% CI: 1.26–1.65, P < 0.001)^[17]. Meanwhile, a larger sample size was included in the final pooled analysis. In addition to subgroup analyses based on similar baseline characteristics, we conducted subgroup analyses by surgical and anesthesia types to provide clinical evidence. Differently, our study found that while people with diabetes showed an increased risk of POCD across all surgical types, the risk was particularly pronounced among patients undergoing non-cardiac surgeries. This finding contrasts with data from a recent systematic review, which reported a range of 2.2% to 31.5% for non-cardiac surgeries and 11.8% to 35.7% for cardiac surgeries^[68].

However, the primary driver of this difference seems to be the advanced age of patients, as we observed that most studies involving non-cardiac surgeries had a mean patient age exceeding 65 years. Among the 14 studies focusing on the non-cardiac surgery group, 7 specifically targeted the elderly population^{[42,43,52,53,63,64,66]}, confining the inclusion criteria to older individuals. Secondly, upon examining the remaining 7 studies^{[6,31,32,36,38,54,58]}, we found that 4 studies involved carotid endarterectomy procedures^[6,32,36,38]. In contrast, the others addressed intravenous thrombolysis^[31], standard colorectal cancer open surgery^[54], and transurethral resection of the prostate^[58]. These surgeries were geared toward patients with age-related conditions such as cerebrovascular disease, cancer, and adenomas. Considering the follow-up period, except for two studies^[31,43], the maximum follow-up duration in the remaining studies involving non-cardiac surgeries was less than one month^{[6,32,36,38,42,52-54,58,63,64,66]}. This short timeframe makes capturing short-term cognitive impairments, such as postoperative delirium, possible.

Meanwhile, cardiac surgery patients often receive more rigorous preoperative monitoring and postoperative care, creating favorable conditions for preventing POCD. Although age has been established as an independent risk factor for POCD, diabetes, as a potential risk factor, may further exacerbate cognitive dysfunction in elderly populations. Lastly, heterogeneity among the studies cannot be overlooked, and insufficient sample sizes may contribute to these results. Future research should consider the interaction between follow-up time, age, and diabetes, primarily to determine whether diabetes still has a significant impact on POCD in non-cardiac surgery patients after adjusting for multiple factors.

Nevertheless, the role of anesthesia in the complex mechanism of POCD remains unclear. Studies by Tzimas et al^[69] and a systematic review^[70] have suggested no correlation between anesthesia methods and the risk of POCD. In light of the predominantly employed general anesthesia in the selected studies, we only found a 39% increased risk in diabetic patients undergoing general anesthesia. Additionally, a more significant effect size was observed in non-Asian regions, which may be attributed to differences in ethnic groups. Different ethnicities can exhibit variations in intracranial vascular and cardiovascular pathologies, such as atherosclerosis and coronary stenosis^[71,72]. These factors have been proven to be risk factors for POCD^[40,73,74]. A possible explanation is that these parametric variations may increase the risk of cerebral hemodynamic instability and microembolism formation in these populations^[75], thereby increasing their susceptibility to POCD. Surgical procedures can alter hemodynamic patterns, potentially leading to cerebral ischemia/reperfusion and direct cellular damage^[76]. The systemic inflammatory response triggered by surgery can also contribute to plaque instability and accelerated microembolism formation, ultimately leading to POCD. The neuroinflammatory response triggered by surgery gradually improves after one month. In previous studies, most patients were observed to recover cognitive function after 30 days^[38,59]. There is limited research on the correlation between preoperative glycemic control and the occurrence of POCD in patients. Strict glycemic control (119 ± 18 mg/dL equivalent to 6.6 ± 1.0 mmol/L) seems to impact neurocognitive function 78 negatively or unrelated^[59], while higher blood glucose maintenance is beneficial for clinical prognosis^[77]. Based on the correlations we have observed, there is a need to address the requirement for risk screening. In clinical practice, it is helpful to develop appropriate neurocognitive protection strategies for diabetic patients with different backgrounds during the perioperative period.

Potential mechanisms

Although postmortem neuropathological studies have established a link between type 2 diabetes mellitus (T2DM) and cerebrovascular diseases, it is still unclear whether T2DM-related cognitive impairment or dementia is solely related to the effects of cerebrovascular diseases, ageing, or neurodegeneration^[78]. Insulin resistance has been identified as a potential risk factor^[79], but animal studies have also shown that diabetes exhibits non-Alzheimer’s disease (AD) processes in the development of cognitive dysfunction^[80]. Common comorbidities of diabetes, such as acute hyperglycemia, advanced glycation end products^[81], blood-brain barrier permeability impairment^[82], oxidative stress^[83], and inflammation^[84], have complex effects on brain function through mechanisms independent of insulin signaling, leading to varying degrees of damage to cognitive function.

Strength and limitations

The main strengths of this meta-analysis include having a larger sample population from all over the world and using a systematic approach, including the risk of bias, grading of evidence specifically for the association of DM and POCD, and some sensitivity analyses to assess the stability of the results. Our research has some potential limitations. The moderate heterogeneity in our meta-analysis warrants cautious interpretation. We could not fully explain the heterogeneity despite using a random effect model and subsequent subgroup analysis and meta-regression. We attribute this to various factors, including the diversity in neurocognitive function assessment methods and diagnostic criteria^[85]. Most tests focus on visual and verbal memory, spatial abstraction, language comprehension, and attention, with fewer assessing learning ability^[86]. As highlighted in previous research, the lack of a standardized diagnostic criterion for postoperative cognitive dysfunction (POCD) is notable^[85]. Despite a consensus statement advocating for a “core battery” of assessments in 1995, widespread adoption has not occurred^[87]. Differences in patients, diagnostic criteria, and control groups contribute to potential heterogeneity, as described in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5). Shorter follow-up periods may also confound the distinction between POCD and other cognitive-related conditions. While we obtained the most adjusted relative risk (RR) from original studies, it is essential to acknowledge variations in adjustment levels.

Additionally, residual confounding factors, such as diabetes type and glycemic control, should be considered due to limitations in available data. Existing evidence predominantly focuses on type 2 diabetes, aligning with our meta-analysis results^[19],^[88]. Secondly, the diagnosis of some diabetes patients may come from the form of self-report, which makes us underestimate the correlation between the two. Thirdly, the grading system results indicate that the quality of evidence in the study is not high, making us more cautious. Fourthly, the existence of publication bias cannot be ignored. We assumed that smaller studies may have reported significant results while ignoring insignificant ones, although the results did not change significantly after trimming and filling. Delirium may affect POCD development, but inconsistent assessment in the included studies prevented us from controlling for it. This limitation could introduce bias, so future systematic reviews and original research should consider including delirium in their analyses. Finally, conducting a meta-analysis combining observational studies and randomized clinical trials is a methodological limitation. However, the only randomized controlled trial in this analysis does not negate the significant effects of observational studies. These limitations emphasize the need for further research to better understand the relationship between diabetes and postoperative cognitive dysfunction.

Conclusion

In summary, our systematic review and meta-analysis have revealed an increased risk of postoperative cognitive dysfunction (POCD) among diabetic patients. However, due to various limitations, this evidence is considered preliminary despite the robustness of the results having been validated through multiple approaches. Further, well-designed studies are needed to discuss appropriate strategies to reduce the incidence of POCD and actively explore the impact of diabetes type and glycemic control on the risk of onset.

Ethical approval

Not applicable.

Consent

Not applicable.

Sources of funding

This work was supported by the Natural Science Foundation of China (No. 82160371 to J.Z., No. 82100869 and No. 82360162 to P.Y., Nos. 81760150 and 82160162 to J.P.L.); Natural Science Foundation in Jiangxi Province grant [20224ACB216009 to J.Z]; the Jiangxi Province Thousands of Plans (No. jxsq2023201105 to P.Y.); Young Elite Scientists Sponsorship Program by JXAST (No. 2023QT05 to J.Z.) and the Hengrui Diabetes Metabolism Research Fund (No. Z-2017-26-2202-4 to P.Y.).

Author’s contribution

L.H.B.: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, resources, software, supervision, validation, visualization, writing; C.J.L.: conceptualization, investigation, methodology, project administration, resources, software-supervision, validation, visualization, writing; L.J.T., W.Y.T., Y.P.P., L.X., L.J.P., Z.D.J., Y.X.P.: conceptualization, investigation, methodology, project administration, resources, software, supervision, validation, writing; Y.P., Z.J.: conceptualization, data curation, formal analysis, funding acquisition resources, supervision.

Research registration unique identifying number (UIN)

The protocol for this meta-analysis has been registered with PROSPERO (CRD42023490602) at https://www.crd.york.ac.uk/PROSPERO/#recordDetails.

Conflicts of interest disclosure

All authors declare that they have no conflicts of interest.

Guarantor

Peng Yu and Jing Zhang.

Provenance and peer review

Not commissioned, externally peer-reviewed.

Availability of data and materials

All data generated or analyzed during this study are included in this published article [and its supplementary information files].

Assistance with the study

None.

Presentation

None.