Exploratory research on asthma exacerbation risk factors using the Japanese claims database and machine learning: a retrospective cohort study.

OBJECTIVE

Analytical studies of risk factor assessment using machine learning have recently been reported. We performed an exploratory detection study of asthma exacerbation-related factors using health insurance claims data and machine learning to explore risk factors that have high generalizability and can be easily obtained in daily practice.

METHODS

A dataset of asthma patients during May 2014-April 2019 from the Japanese insurance claims database, MediScope® (DB) was used. Patient characteristics and disease information were extracted, and association with occurrence of asthma exacerbation was evaluated to comprehensively search for exacerbation risk factors. Asthma exacerbations were defined as the co-occurrence of emergency medical procedures, such as emergency transport and intravenous steroid injections, with asthma claims, which were recorded in the database.

RESULTS

In total, 5,844 (13.7%) subjects had exacerbations in 42,685 eligible cases from the DB. Information on approximately 3,300 diseases was subjected to a machine learning, and 25 variables were extracted as variable importance and targeted for risk assessment. As a result, sex, days without exacerbation from cohort entry date at look-back period, Charlson Comorbidity Index, allergic rhinitis, chronic sinusitis, acute airway disease (upper airway), acute airway disease (lower airways), Chronic obstructive pulmonary disease/chronic bronchitis, gastroesophageal reflux disease, and hypertension were significantly associated with exacerbation. Dyslipidemia and periodontitis were detected as associated factors of reduced exacerbation risk.

CONCLUSIONS

A comprehensive analysis of claims data using machine learning showed asthma exacerbation risk factors mostly consistent with those in previous studies. Further examination in other fields is warranted.Supplemental data for this article is available online at https://doi.org/10.1080/02770903.2021.1923740 .