An immune-inflammatory dysregulation score for risk stratification of 28-day all-cause mortality in critically ill patients with severe community-acquired pneumonia.

BACKGROUND: Severe community-acquired pneumonia (SCAP) is associated with high mortality and substantial heterogeneity in host immune responses among critically ill patients. Conventional prognostic scores primarily reflect comorbidity burden and organ dysfunction but do not directly capture immune dysregulation. METHODS: This retrospective cohort study included 223 critically ill patients with SCAP admitted to the respiratory intensive care unit (RICU). Six prespecified immune-inflammatory biomarkers were evaluated as candidate predictors. The Immune-Inflammatory Dysregulation Score (IIDS) was developed using bootstrap-based stability selection (1,000 bootstrap resamples) followed by multivariable logistic regression. The discriminative performance of IIDS was compared with conventional clinical severity scores, and its incremental prognostic value was assessed using sequential prediction models. A final prediction model incorporating IIDS was internally evaluated using bootstrap optimism correction and presented as a nomogram. RESULTS: IIDS comprised CD4+ T-cell count, monocyte count, and the neutrophil-to-lymphocyte ratio. IIDS showed moderate apparent discrimination for 28-day mortality (AUC: 0.721; 95% CI: 0.654-0.787). Adding IIDS to the CCI + SOFA model increased the apparent AUC from 0.773 to 0.819 (DeLong P = 0.014) and the bootstrap optimism-corrected AUC from 0.767 to 0.809. Optimism-corrected reclassification estimates were imprecise, with confidence intervals for NRI and IDI including zero. The final model showed close agreement between predicted and observed risks after optimism correction and provided potential net clinical benefit across the evaluated threshold-probability range. CONCLUSIONS: IIDS provided prognostic information complementary to conventional clinical predictors and improved apparent discrimination when incorporated into a sequential prediction model. Independent external validation is required before clinical implementation.