Multimodal deep learning for COVID-19 prognosis prediction in the emergency department: a bi-centric study. (preprint)

Clinical deterioration of COVID-19 patients is still a challenging event to predict in the emergency department (ED). The present study developed an artificial neural network using textual and tabular data from ED electronic medical reports. Predicted outcomes were 30-day mortality and ICU admission. Consecutive patients between February 20 and May 5, 2020, from Humanitas Research Hospital and San Raffaele Hospital, in the Milan area, were included. COVID-19 patients were 1296. Textual predictors were patient history, physical exam, and radiological reports. Tabular predictors were age, creatinine, C-reactive protein, hemoglobin, and platelet count. Tabular-textual model performance indices were compared to a model implementing only tabular data. For 30-day mortality, the combined model yielded slightly better performances than the tabular model, with AUC 0.84 ± 0.02, F-1 score 0.56 ± 0.04 and an MCC 0.44 ± 0.04. Tabular model performance was: AUC 0.84 ± 0.02, F-1 score 0.55 ± 0.03 and MCC 0.43 ± 0.04. As for ICU admission, the combined model was not superior to the tabular one.  The present data points to the effectiveness of a textual and tabular model for COVID-19 prognosis prediction. Also, it may support the ED physician in their decision-making process.

On the explainability of hospitalization prediction on a large COVID-19 patient dataset (preprint)

We develop various AI models to predict hospitalization on a large (over 110$k$) cohort of COVID-19 positive-tested US patients, sourced from March 2020 to February 2021. Models range from Random Forest to Neural Network (NN) and Time Convolutional NN, where combination of the data modalities (tabular and time dependent) are performed at different stages (early vs. model fusion). Despite high data unbalance, the models reach average precision 0.96-0.98 (0.75-0.85), recall 0.96-0.98 (0.74-0.85), and $F_1$-score 0.97-0.98 (0.79-0.83) on the non-hospitalized (or hospitalized) class. Performances do not significantly drop even when selected lists of features are removed to study model adaptability to different scenarios. However, a systematic study of the SHAP feature importance values for the developed models in the different scenarios shows a large variability across models and use cases. This calls for even more complete studies on several explainability methods before their adoption in high-stakes scenarios.