Validation of a visual-based analytics tool to predict outcomes of polytrauma patients: The IBM WATSON trauma pathway explorer

Abstract

Abstract Objective Big data-based artificial intelligence (AI) is on the way to develop into a part of daily clinical life and its reasonable application could help to improve disease or injury outcomes. A visual polytrauma analytics tool based on IBM WATSON was developed and described in a previous publication. The present article relates to the validation of the IBM WATSON Trauma Pathway Explorer. Methods A retrospective prediction model validation in a level I trauma center including 107 patients with an Injury Severity Score (ISS) ≥16 and age ≥16 was performed. Age, ISS, temperature and the presence of head injury were the predictors used to validate the following three outcomes: SIRS and sepsis within 21 days since admission of the patient, as well as early death within 72 hours since admission. The area under the receiver operating characteristic (ROC) curve was used to determine predictive quality. Calibration plots showed the graphical goodness of fit. The Brier score assessed the overall performance of the two models. Results The area under the curve (AUC) is 0.77 (95% CI: 0.679-0.851) for SIRS, 0.71 (95% CI: 0.578-0.831) for sepsis and 0.90 (95% CI: 0.786-987) for early death. The Brier scores are as follows: early death 0.06, sepsis 0.12 and SIRS 0.15. Conclusion The validation has shown that the predictive performance of WATSON for SIRS and early death corresponds to the clinical outcome in nearly 80% of cases and 90% of cases, respectively. The concordance for sepsis was modest with over 70% of cases. This visual analytics tool for polytrauma patients can be used to obtain valid predictions for SIRS, sepsis and early death. Here, we can present a possible working variant of AI in trauma surgery.