JMIR Publications

ABSTRACT

Background:

Asthma affects a large proportion of the population and leads to a lot of hospital encounters covering both hospitalizations and emergency department visits every year. To lower the number of such encounters, many healthcare systems and health plans deploy predictive models to prospectively find patients at high risk and offer them care management services for preventive care. Yet, the previous models do not have enough accuracy to serve this purpose well. Embracing the modeling strategy of examining many candidate features, we newly built a machine learning model to forecast asthmatic patients’ future asthma hospital encounters at Intermountain Healthcare, a non-academic healthcare system. This model is more accurate than the previous published models. But, it is unclear how well our modeling strategy generalizes to academic healthcare systems, whose patient composition is different from Intermountain Healthcare’s.

Objective:

This study evaluates our modeling strategy’s generalizability to University of Washington Medicine (UWM), an academic healthcare system.

Methods:

All of the adult asthmatic patients who visited UWM facilities between 2011 and 2018 served as the patient cohort. We considered 234 candidate features. Through secondary analysis of 82,888 UWM data instances from 2011 to 2018, we built a machine learning model to forecast asthmatic patients’ asthma hospital encounters in the subsequent 12 months.

Results:

Our UWM model yielded an area under the receiver operating characteristic curve (AUC) of 0.902. When placing the cutoff point for making binary classification at the top 10.00% (1,464/14,644) of asthmatic patients with the biggest forecasted risk, our UWM model yielded an accuracy of 90.60% (13,268/14,644), a sensitivity of 70.18% (153/218), and a specificity of 90.91% (13,115/14,426).

Conclusions:

Our modeling strategy showed excellent generalizability to UWM, leading to a model with an AUC that is higher than all of the AUCs previously reported in the literature for forecasting asthma hospital encounters. After further optimization, our model could be employed to facilitate efficient and effective allocation of asthma care management resources to improve outcomes.