From glycemic instability to neuropathic risk: a propensity score-matched retrospective cohort study

Background and Objective: Diabetic peripheral neuropathy (DPN) is a prevalent and debilitating complication of type 2 diabetes mellitus (T2DM). Although glycated hemoglobin (HbA1c) is a primary metric for glycemic control, many patients develop or experience progression of DPN despite achieving HbA1c targets, suggesting the importance of other dynamic glycemic parameters. Glycemic variability (GV) may contribute to nerve injury via mechanisms such as oxidative stress, inflammation, and neurotrophic factor dysregulation. However, clinical evidence linking GV to DPN remains inconsistent, and rigorous studies controlling for confounders are scarce. This study aimed to determine whether GV is independently associated with DPN beyond HbA1c in a propensity score-matched (PSM) cohort and to explore the potential mediating roles of inflammatory cytokines and neurotrophic factors. Methods: This single-center retrospective cohort study screened T2DM patients hospitalized between January 1, 2020, and December 31, 2024. Patients with complete 72-hour continuous glucose monitoring (CGM) data and bilateral nerve conduction studies (NCS) were included. DPN was diagnosed according to the Chinese Diabetes Society guidelines. Propensity score matching (PSM, 1:1, caliper=0.02) was used to balance the DPN and non-DPN groups on age, sex, BMI, diabetes duration, HbA1c, systolic blood pressure, LDL-C, and estimated glomerular filtration rate. Primary outcomes included GV parameters (mean amplitude of glycemic excursions [MAGE], coefficient of variation [CV], standard deviation [SD]) and a composite nerve conduction velocity (NCV) Z-score. Serum inflammatory cytokines (IL-6, TNF-α) and neurotrophic factors (NGF, IGF-1) were measured in a nested subcohort. Data were analyzed using multivariable linear regression, dose-response analysis, causal mediation analysis, and receiver operating characteristic (ROC) curve analysis. Results: After PSM, 256 well-matched patients (128 in each group) were included, with excellent covariate balance (all standardized mean differences