Prevalence, risk factors, and early prediction of cognitive impairment in patients with diabetes mellitus: a systematic review and meta-analysis

Abstract

Introduction – Diabetes mellitus (DM) substantially increases the risk of cognitive impairment (CI). Currently, systematic evidence regarding its prevalence, risk factors, and early predictive models remains scarce, hindering the development of targeted preventive strategies. Thus, this meta-analysis aimed to comprehensively estimate the global CI prevalence in individuals with DM, identify associated risk factors, and assess the performance and potential application of early predictive models. Methods – Embase, Web of Science, the Cochrane Library, and PubMed were searched up to May 2025 to include observational studies reporting on the prevalence, risk factors, and predictive models of CI in individuals with DM. The Agency for Healthcare Research and Quality scale and the Newcastle-Ottawa Scale were employed to appraise the quality of included studies. Meta-analyses of prevalence, risk factors, and c-indices of predictive models were carried out using R 4.5.0. This study was registered in PROSPERO (CRD420250632808). Results – In total, 41 studies involving 18, 768 patients with DM were included. After excluding three case–control studies, 38 studies were eligible for prevalence synthesis. Meta-analysis demonstrated an overall CI prevalence of 40.80% (95% CI: 33.91%–47.87%) among DM patients. Identified risk factors included demographic factors: age (odds ratio [OR]=2.18 95% confidence interval [CI]: 1.37–3.48), low income (OR = 2.25, 95%CI:1.28–3.96), and unmarried status (OR = 1.66, 95% CI:1.05–2.64); disease-related factors: Hemoglobin A1c (OR = 1.27, 95% CI:1.06–1.53), and hypoglycemia (OR = 1.87, 95% CI:1.30–2.71); and complications: diabetic nephropathy (OR = 1.64, 95% CI:1.19–2.25), diabetic retinopathy (OR = 1.71, 95% CI:1.27–2.42), depression (OR = 2.32, 95%CI: 2.32-5.33), and stroke (OR = 2.62, 95%CI:1.27–2.31). 8 studies constructed predictive models for CI, reporting a c-index of 0.83 (95%CI: 0.76–0.90) in the training sets and 0.81 (95%CI:0.74–0.89) in the validation sets. Conclusions – CI is highly prevalent among individuals with DM and is closely associated with multiple factors. These risk factors provide potential targets for early intervention. Although existing predictive models demonstrate encouraging performance, their clinical applicability remains limited owing to the small number of included studies and requires further validation. Future research should incorporate large, multicenter, and multiethnic cohorts and develop machine learning–based predictive models with broader applicability.