ISeeU2: Visually interpretable mortality prediction inside the ICU using deep learning and free-text medical notes

Accurate mortality prediction allows Intensive Care Units (ICUs) to adequately benchmark clinical practice and identify patients with unexpected outcomes. Traditionally, simple statistical models have been used assess patient death risk, many times sub-optimal performance. On the other hand Deep Learning holds promise positively impact by leveraging medical data assist diagnosis prediction, including prediction. However, as question of whether powerful attend correlations backed sound knowledge when generating predictions remains open, additional interpretability tools are needed foster trust encourage use AI clinicians. In this work we show an interpretable model trained on MIMIC-III predict inside ICU using raw nursing notes, together visual explanations for word importance based Shapley Value. Our reaches a ROC 0.8629 ( ± 0.0058), outperforming traditional SAPS-II score LSTM recurrent neural network baseline while providing enhanced compared similar approaches. Supporting code can be found at https://github.com/williamcaicedo/ISeeU2 . • outperform scores such SAPS-II. Results suggest uses metadata mortality. Performance is competitive state art simpler pre-processing. Values offer without sacrificing predictive