A Multimodal Model Integrating Early Clinical and Biomarker Data to Predict 1-Year Outcomes after Pediatric Cardiac Arrest.

BACKGROUND: Shared decision-making after pediatric cardiac arrest (CA) requires early, accurate prognostication. We analyzed whether adding brain injury biomarkers and neuroimaging findings to clinical models increased prognostic accuracy after pediatric CA. METHODS: This planned analysis of the POCCA study included children aged 48 hours to 17 years with ICU admission after CA. Clinical predictors included demographic, CA, and resuscitation variables. Blood neurofilament light, ubiquitin carboxyl-terminal esterase L1, glial fibrillary acidic protein, and tau were measured on day 1 post-CA. MRI Injury Score was assessed within 2 weeks post-CA. Primary outcome was unfavorable outcome (death or Vineland Adaptive Behavior Scale <70) at 1 year. Likelihood ratio tests (LRT) compared model fit with and without biomarkers. Four alternative models were analyzed in an Outcomes Cohort using bootstrap-generated AUROCs, and we assessed whether MRI Injury Score provided incremental value. RESULTS: Of 163 enrolled children, 118 with day 1 biomarker data and clinical predictors formed the Outcomes Cohort. Biomarkers improved model fit over clinical variables alone in the Outcomes Cohort (LRT p=0.002), but not in the full, imputed POCCA Cohort (LRT p=0.07). The penalized regression model showed the highest discrimination among four approaches (optimism-corrected AUROC, 0.868 [95% CI, 0.82-0.94]; calibration intercept, 0.050 [0.00-0.26]; calibration slope, 1.067 [1.00-1.47]). The addition of MRI Injury Score improved predictive performance (aOR=1.13; 95% CI, 1.03-1.24, p=0.016). CONCLUSIONS: Brain injury biomarkers and MRI Injury Score improved prognostic accuracy over clinical variables alone for 1-year outcome in the Outcomes Cohort. External validation is needed before these findings can inform clinical decision-making.