Abstract / Summary
Surgical-site infection (SSI) after instrumented spinal surgery causes substantial morbidity, reoperation, and healthcare burden. Although perioperative hyperglycemia is a recognized risk factor, the incremental predictive value of short-term glycemic variability beyond diabetes, glycated hemoglobin, and mean glucose remains uncertain. This study aimed to determine whether perioperative glycemic variability improves early postoperative prediction of SSI after instrumented spinal surgery. This single-center retrospective cohort study included 1,487 consecutive adults undergoing instrumented spinal surgery between 1 January 2019 and 31 December 2022. Predictors were fixed 24 h after surgery. The primary outcome was CDC/NHSN-defined SSI within 30 days, with 90-day ascertainment of deep or organ/space SSI. Glucose coefficient of variation (CV) was the primary variability measure. Random-forest and penalized logistic-regression models were evaluated by bootstrap optimism correction, calibration, and decision-curve analysis. The modeling pipeline, including hyperparameters and calibration, was then locked and applied unchanged to a later, non-overlapping cohort of 1,050 patients treated between 1 January 2023 and 31 December 2025 (63 events). Sensitivity analyses assessed the independent association of glucose CV with SSI. SSI occurred in 78 patients (5.2%) within 30 days and 99 (6.7%) within 90 days. Glucose CV was higher among patients with SSI than among those without (28.6%±6.4% vs. 19.6%±7.3%; Hedges’ g = 1.31). Adding glucose CV to conventional predictors increased the area under the curve from 0.790 to 0.849 (ΔAUC = 0.059, 95% CI 0.031–0.088; P < 0.001), and a comparable increment persisted against an expanded clinical model containing surgical approach, open versus minimally invasive technique, emergency status, ASA physical status, and drain use (ΔAUC = 0.049, 95% CI 0.024–0.075; P < 0.001). The optimism-corrected random-forest AUC was 0.846 (95% CI 0.808–0.881), and the locked model achieved 0.809 (95% CI 0.756–0.856) in the later cohort. Glucose CV remained independently associated with SSI (adjusted odds ratio 3.21 per standard deviation, 95% CI 2.10–4.91). Perioperative glucose CV provided predictive information beyond conventional glycemic indicators and beyond a comprehensive clinical model, and performed reproducibly over time within the same institution. The threshold-linked actions presented here are illustrative rather than prospectively evaluated, and external prospective validation is required before implementation.