On the explainability of hospitalization prediction on a large COVID-19\n patient dataset

2021/10/28 by Ivan Girardi, Panagiotis Vagenas, Girardi, Ivan +20
Computer Science · Medicine · #Machine Learning in Healthcare #Explainable Artificial Intelligence (XAI) #Sepsis Diagnosis and Treatment

paper · pdf · doi:10.48550/arxiv.2110.15002

Abstract

We develop various AI models to predict hospitalization on a large (over\n110k) cohort of COVID-19 positive-tested US patients, sourced from March 2020\nto February 2021. Models range from Random Forest to Neural Network (NN) and\nTime Convolutional NN, where combination of the data modalities (tabular and\ntime dependent) are performed at different stages (early vs. model fusion).\nDespite high data unbalance, the models reach average precision 0.96-0.98\n(0.75-0.85), recall 0.96-0.98 (0.74-0.85), and F₁-score 0.97-0.98\n(0.79-0.83) on the non-hospitalized (or hospitalized) class. Performances do\nnot significantly drop even when selected lists of features are removed to\nstudy model adaptability to different scenarios. However, a systematic study of\nthe SHAP feature importance values for the developed models in the different\nscenarios shows a large variability across models and use cases. This calls for\neven more complete studies on several explainability methods before their\nadoption in high-stakes scenarios.\n

Citations

• Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD): The TRIPOD Statement
• PROBAST: A Tool to Assess the Risk of Bias and Applicability of Prediction Model Studies

Related

• Deployment of a Robust and Explainable Mortality Prediction Model: The COVID-19 Pandemic and Beyond
• An early prediction of covid-19 associated hospitalization surge using deep learning approach
• An Explainable Artificial Intelligence Approach Using Graph Learning to Predict Intensive Care Unit Length of Stay
• COVID-Net Clinical ICU: Enhanced Prediction of ICU Admission for COVID-19 Patients via Explainability and Trust Quantification
• MedML: Fusing Medical Knowledge and Machine Learning Models for Early Pediatric COVID-19 Hospitalization and Severity Prediction