Who dies from COVID-19? Post-hoc explanations of mortality prediction models using coalitional game theory, surrogate trees, and partial dependence plots
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (ScreenIT)
Abstract
As of early June, 2020, approximately 7 million COVID-19 cases and 400,000 deaths have been reported. This paper examines four demographic and clinical factors (age, time to hospital, presence of chronic disease, and sex) and utilizes Shapley values from coalitional game theory and machine learning to evaluate their relative importance in predicting COVID-19 mortality. The analyses suggest that out of the 4 factors studied, age is the most important in predicting COVID-19 mortality, followed by time to hospital. Sex and presence of chronic disease were both found to be relatively unimportant, and the two global interpretation techniques differed in ranking them. Additionally, this paper creates partial dependence plots to determine and visualize the marginal effect of each factor on COVID-19 mortality and demonstrates how local interpretation of COVID-19 mortality prediction can be applicable in a clinical setting. Lastly, this paper derives clinically applicable decision rules about mortality probabilities through a parsimonious 3-split surrogate tree, demonstrating that high-accuracy COVID-19 mortality prediction can be achieved with simple, interpretable models.
Article activity feed
-
SciScore for 10.1101/2020.06.07.20124933: (What is this?)
Please note, not all rigor criteria are appropriate for all manuscripts.
Table 1: Rigor
NIH rigor criteria are not applicable to paper type.Table 2: Resources
Software and Algorithms Sentences Resources The model was trained with default parameters using the Python xgboost package. Pythonsuggested: (IPython, RRID:SCR_001658)Results from OddPub: Thank you for sharing your code and data.
Results from LimitationRecognizer: An explicit section about the limitations of the techniques employed in this study was not found. We encourage authors to address study limitations.Results from TrialIdentifier: No clinical trial numbers were referenced.
Results from Barzooka: We found bar graphs of continuous data. We recommend replacing bar graphs with more informative graphics, as many different …
SciScore for 10.1101/2020.06.07.20124933: (What is this?)
Please note, not all rigor criteria are appropriate for all manuscripts.
Table 1: Rigor
NIH rigor criteria are not applicable to paper type.Table 2: Resources
Software and Algorithms Sentences Resources The model was trained with default parameters using the Python xgboost package. Pythonsuggested: (IPython, RRID:SCR_001658)Results from OddPub: Thank you for sharing your code and data.
Results from LimitationRecognizer: An explicit section about the limitations of the techniques employed in this study was not found. We encourage authors to address study limitations.Results from TrialIdentifier: No clinical trial numbers were referenced.
Results from Barzooka: We found bar graphs of continuous data. We recommend replacing bar graphs with more informative graphics, as many different datasets can lead to the same bar graph. The actual data may suggest different conclusions from the summary statistics. For more information, please see Weissgerber et al (2015).
Results from JetFighter: We did not find any issues relating to colormaps.
Results from rtransparent:- Thank you for including a conflict of interest statement. Authors are encouraged to include this statement when submitting to a journal.
- Thank you for including a funding statement. Authors are encouraged to include this statement when submitting to a journal.
- No protocol registration statement was detected.
-
