JMIR Publications

ABSTRACT

Background:

Asthma is a major chronic disease posing a heavy burden on healthcare. To facilitate preventive care and improve outcomes for high-risk asthmatic patients via care management, we recently built a machine learning model to predict asthma hospital visits in the subsequent year in asthmatic patients. This model is more accurate than the prior models. However, like most machine learning models, our model offers no explanation of its prediction results. This creates a hurdle to use it in care management, where interpretability is desired.

Objective:

To address this limitation, this study aims to develop a method to automatically explain the model’s prediction results and to recommend tailored interventions without lowering any of the performance measures of the model.

Methods:

Our data are imbalanced, with only a small portion of data instances linking to future asthma hospital visits. To handle imbalanced data, we extended our prior method for automatically offering rule-formed explanations for any machine learning model’s prediction results on tabular data without lowering any of the performance measures of the model. In a secondary analysis of the 334,564 data instances from Intermountain Healthcare during 2005-2018 used to form our model, we employed the extended method to automatically explain our model’s prediction results and to recommend tailored interventions. The patient cohort consisted of all asthmatic patients who obtained care at Intermountain Healthcare during 2005 to 2018 and resided in Utah or Idaho as recorded at the visit.

Results:

Our method explained the prediction results for 89.68% (391/436) of the asthmatic patients whom our model correctly predicted to incur asthma hospital visits in the subsequent year.

Conclusions:

This study is the first to demonstrate the feasibility of automatically offering rule-formed explanations for any machine learning model’s prediction results on imbalanced tabular data without lowering any of the performance measures of the model. After further improvement, our asthma outcome prediction model coupled with the automatic explanation function could be used for decision support to guide allocation of limited asthma care management resources.