Influencing factors and predictive model construction of perioperative blood glucose fluctuations in patients with coronary heart disease complicated with type 2 diabetes mellitus undergoing percutaneous coronary intervention

The study aimed to explore key influencing factors of perioperative blood glucose fluctuations in patients with coronary heart disease (CHD) complicated with type 2 diabetes mellitus (T2DM) undergoing percutaneous coronary intervention (PCI), and construct/validate a clinical risk prediction model for precise blood glucose management. A retrospective cohort study enrolled 457 eligible patients (282 in the blood glucose fluctuation group, 175 in the normal group; defined by SDBG ≥ 2.0 mmol/L, PPGE ≥ 2.2 mmol/L, or LAGE ≥ 4.4 mmol/L). Patients were randomly divided into training (n = 320, 70%) and validation cohorts (n = 137, 30%). After Lasso regression dimensionality reduction, a nomogram was built via multivariate Logistic regression. Model performance was assessed by receiver operating characteristic (ROC) curve, calibration curve, Hosmer-Lemeshow test, and curve analysis (DCA); stability was verified by 1000 bootstrap samples and multiple sensitivity analyses. Univariate analysis identified 21 associated indicators (P < 0.05), and Lasso regression optimized variables. Multivariate Logistic regression confirmed 11 independent risk factors: age, diabetes duration, SBP, HbA1c, LDL-C, HDL-C, FBG, PSQI score, surgical start time window, surgical duration, and preoperative last meal-to-surgery interval (P < 0.05). The 11-variable full variable model achieved AUCs of 0.973 (95% CI 0.958–0.988) in the training cohort and 0.982 (95% CI 0.966–0.998) in the validation cohort; DCA confirmed that the model yielded significant clinical net benefits within the risk threshold of 0–65%. The 5 high-impact predictor model constructed after excluding 6 high-influence variables still maintained high predictive performance, with AUCs of 0.868 in the training cohort and 0.880 in the validation cohort, and all variables are clinically easily accessible indicators. The nomogram model constructed with the 11 identified indicators in this study exhibits excellent predictive performance for perioperative blood glucose fluctuations in patients with CHD complicated with T2DM undergoing PCI. In contrast, the 5 high-impact predictor model balances predictive accuracy and clinical convenience. These two models can be respectively applied to refined risk stratification of inpatients and rapid preoperative screening in primary care settings/outpatient clinics, providing a practical predictive tool for clinical individualized management of perioperative blood glucose and a scientific basis for reducing the risk of perioperative adverse cardiovascular events.