Table 6.
Incremental model performance, internal validation, and sensitivity analyses
| Analysis | Estimate | 95% CI | p-value |
|---|---|---|---|
| Panel A. Model performance and internal validation | |||
| Clinical model apparent AUC | 0.772 | 0.661–0.861 | — |
| Clinical model optimism-corrected AUC | 0.742 | — | — |
| Clinical model Brier score | 0.171 | — | — |
| Clinical model calibration slope | 0.81 | — | — |
| Clinical model calibration intercept | −0.06 | — | — |
| Clinical model Hosmer–Lemeshow p-value | — | — | 0.517 |
| Clinical model + ΔWINA apparent AUC | 0.903 | 0.827–0.958 | — |
| Clinical model + ΔWINA optimism-corrected AUC | 0.868 | — | — |
| Clinical model + ΔWINA Brier score | 0.118 | — | — |
| Clinical model + ΔWINA calibration slope | 0.84 | — | — |
| Clinical model + ΔWINA calibration intercept | −0.04 | — | — |
| Clinical model + ΔWINA Hosmer–Lemeshow p-value | — | — | 0.684 |
| Change in AUC after adding ΔWINA | 0.131 | 0.044–0.218 | 0.003 |
| Likelihood-ratio χ² | 18.74 | — | < 0.001 |
| Panel B. Sensitivity analyses for ΔWINA | |||
| Excluding complications diagnosed on or before POD3: adjusted OR per 100 units | 1.96 | 1.32–2.91 | 0.001 |
| Excluding complications diagnosed on or before POD3: AUC | 0.856 | 0.758–0.930 | — |
| SSI-only outcome: adjusted OR per 100 units | 2.04 | 1.31–3.17 | 0.002 |
| SSI-only outcome: AUC | 0.861 | 0.764–0.933 | — |
| Elective surgery subgroup: unadjusted OR per 100 units | 1.88 | 1.18–2.99 | 0.008 |
| Elective surgery subgroup: AUC | 0.844 | 0.730–0.921 | — |
| Major abdominal surgery subgroup: adjusted OR per 100 units | 2.03 | 1.31–3.15 | 0.002 |
| Major abdominal surgery subgroup: AUC | 0.858 | 0.754–0.930 | — |
The clinical model included diabetes mellitus and operative time. The extended model included diabetes mellitus, operative time, and ΔWINA. Internal validation was performed using 1,000 bootstrap resamples, with the complete model-fitting procedure repeated in each resample. Calibration slope and intercept values represent optimism-corrected estimates. The analysis excluding early events included 82 patients and 20 wound complications diagnosed after the POD3 laboratory measurement. The SSI-only analysis included 17 events. These two analyses and the major abdominal surgery analysis were adjusted for diabetes mellitus and operative time. Because of the limited number of events in the elective surgery subgroup, the corresponding odds ratio was obtained from a univariable model. Abbreviations: AUC area under the receiver operating characteristic curve, CI confidence interval, OR odds ratio, PO postoperative day, SSI surgical site infection, WINA Wound Inflammation–Nutrition Index