The relationship between changes in metabolic score for insulin resistance (METS-IR) and diabetes mellitus (DM) remains to be elucidated. The present study sought to explore whether changes in METS-IR were associated with incident DM.This study included 4031 individuals aged over 45 years from the China Health and Retirement Longitudinal Study. Utilizing the K-Means clustering method, participants were divided into four groups. Cumulative METS-IR was employed as a quantitative indicator reflecting changes in METS-IR. Multivariable logistic regression models and restricted cubic splines (RCS) were employed to assess the relationship between changes in METS-IR and DM.During the five-year follow-up, 338 (8.4%) incident DM cases were identified. After adjusting for potential confounders, compared to class 1, the incident DM was significantly higher in class 4 [odds ratio (OR): 2.731, 95% confidence interval (CI): 1.411-5.527, p = 0.003]. When cumulative METS-IR (per SD increase) was introduced as a continuous variable into the multivariate regression model, the association remained significant in Model 3 (OR: 1.393, 95% CI: 1.125-1.715, p = 0.003). Additionally, when cumulative METS-IR was categorized into quartiles, compared to the first quartile, the incident DM was significantly higher in the fourth quartile (OR: 2.022, 95% CI: 1.171-3.501, p = 0.011). In addition, cumulative METS-IR exhibited a nonlinear association with DM (poverall = 0.001, pnon linear = 0.038).Substantial changes in METS-IR were independently associated with the risk of DM in middle-aged and elderly Chinese people. Long-term METS-IR monitoring is significant for early identification and prevention of DM, with significant implications for clinical practice.