Developing a short-term prediction model for asthma exacerbations from Swedish primary care patients' data using machine learning - Based on the ARCTIC study. (August 2021)

• Record Type: Journal Article
• Authors: Lisspers, Karin; Ställberg, Björn; Larsson, Kjell; Janson, Christer; Müller, Mario; Łuczko, Mateusz; Bjerregaard, Bine Kjøller; Bacher, Gerald; Holzhauer, Björn; Goyal, Pankaj; Johansson, Gunnar
• Is Part Of: Respiratory medicine. Volume 185(2021)
• Publication Date: 2021-08
• Subjects: Asthma -- Machine learning -- Exacerbation -- Hospitalization; Chest -- Diseases -- Periodicals; Chest -- Diseases -- Great Britain -- Periodicals; Respiratory organs -- Diseases -- Periodicals; Respiratory Tract Diseases -- Periodicals; Appareil respiratoire -- Maladies -- Périodiques; Thorax -- Maladies -- Périodiques; Appareil respiratoire -- Maladies -- Traitement -- Périodiques; Electronic journals; 616.2
• DOI: 10.1016/j.rmed.2021.106483
• Languages: English
• ISSNs: 0954-6111

Abstract

Abstract: Objective: The ability to predict impending asthma exacerbations may allow better utilization of healthcare resources, prevention of hospitalization and improve patient outcomes. We aimed to develop models using machine learning to predict risk of exacerbations. Methods: Data from 29, 396 asthma patients was collected from electronic medical records and national registers covering clinical and epidemiological factors (e.g. comorbidities, health care contacts), between 2000 and 2013. Machine-learning classifiers were used to create models to predict exacerbations within the next 15 days. Model selection was done using the mean cross validation score of area under precision-recall curve (AUPRC). Results: The most important predictors of exacerbation were comorbidity burden and previous exacerbations. Model validation on test data yielded an AUPRC = 0.007 (95% CI: ± 0.0002), indicating that historic clinical information alone may not be sufficient to predict a near future risk of asthma exacerbation. Conclusions: Supplementation with additional data on environmental triggers, (e.g. weather, pollen count, air quality) and from wearables, might be necessary to improve performance of the short-term predictive model to develop a more clinically useful tool. Highlights: The results highlight the importance of comorbidities for the prognosis of asthma. Models based on retrospective clinically available information may only help risk stratify patients. Additional data on environmental triggers is necessary to improve performance of the predictive model to develop a more clinically useful tool.