Use of Machine learning to predict asthma exacerbations

Abstract

Background: Asthma exacerbations negatively impact disease progression and can lead to hospitalizations and death. Ability to predict exacerbations may allow intervention for prevention and improve outcomes. We aimed to develop models using machine learning to predict risk of exacerbations, using Swedish patient level data. Methods: Data for 33,538 asthma patients were collected from electronic medical records and national registries covering healthcare contacts, diagnoses, prescriptions, lab tests, hospitalizations and socioeconomic factors, between 2000 and 2013. Machine-learning classifiers and logistic regression were used to create models to predict exacerbations within the next 15 days for two groups of adult asthma patients, either including or excluding Chronic Obstructive Pulmonary Disease (COPD). Model performance was assessed by mean cross validation score of area under precision-recall curve (AUPRC) and Area under receiver operating curve (AUROC) was used to compare performance with previous studies. Results: The predictors of exacerbation were comorbidity burden, time since first exacerbation, number of previous exacerbations and number of healthcare contacts due to asthma within the last year. Model validation on test data yielded an AUROC of 0.9 and AUPRC of 0.010, when COPD was included and AUROC=0.90 and AUPRC=0.007 when COPD was excluded. Conclusion: Our work suggests that clinically available information on patient history collected via EMRs and national registries might not suffice to form the basis for tools to predict future risk of asthma exacerbation. Supplementation with other kinds of data might be necessary to improve performance of the predictive model to develop a more clinically useful tool.