Related Experiment Videos
Cox-Based Prediction of 28-Day Mortality in Necrotizing Soft Tissue Infections: Development and External Validation
Jianghui Dong1,2, Xiannian Shao1, Shuliang Hua1
1Department of Joint Surgery, Baise People's Hospital, Affiliated Southwest Hospital of Youjiang Medical University for Nationalities, Baise, Guangxi, People's Republic of China.
Purpose:
Necrotizing soft tissue infections (NSTIs) are rapidly progressive and associated with high mortality. We aimed to develop and externally validate a parsimonious Cox-based model for predicting 28-day all-cause mortality using routinely available admission variables.
Patients And Methods:
This two-center retrospective cohort study included 496 adults with NSTIs: 265 in the development cohort and 231 in an independent external validation cohort. Candidate predictors were selected using 10-fold cross-validated least absolute shrinkage and selection operator Cox regression. Model performance was assessed using Harrell's C-index, time-dependent area under the receiver operating characteristic curve (AUC) at days 7, 14, and 28, calibration, Brier scores, decision curve analysis, and 1,000 bootstrap resamples. Static and web-based nomograms were developed.
Results:
Platelet count, all-cause circulatory shock, activated partial thromboplastin time, and serum albumin were retained in the final model. The C-index was 0.871 (95% CI, 0.822-0.920) in the development cohort and 0.864 (95% CI, 0.821-0.908) in the external validation cohort. The time-dependent AUCs were 0.914, 0.888, and 0.845 at days 7, 14, and 28 in the development cohort and 0.897, 0.922, and 0.906, respectively, in the external cohort. In external validation, the calibration slope was 0.711; mortality was overpredicted at days 7 and 14, whereas mean calibration was close to ideal at day 28. Exploratory comparisons showed higher observed discrimination than retrospectively reconstructed LRINEC, APACHE II, and SOFA scores, although APACHE II and SOFA were not assessed within their conventional ICU-based windows.
Conclusion:
A four-variable Cox model based on routinely available initial clinical data showed good discrimination in both cohorts. Mean calibration was close at the 28-day horizon, but the external calibration slope and early-horizon overprediction indicate that further validation and potential recalibration are needed before routine clinical implementation.