Application of Machine Learning and Causal Inference Approaches to Bronchiolitis and Asthma Research
MetadataShow full item record
CitationRaita, Yoshihiko. 2020. Application of Machine Learning and Causal Inference Approaches to Bronchiolitis and Asthma Research. Master's thesis, Harvard Medical School.
AbstractOVERVIEW: During the curriculum of Master of Medical Sciences Program at Harvard Medical School, I have been focusing on two inferential goals through respiratory disease research—1) prediction and 2) causal inference—which are central pillars of research questions as clinical investigators. Bronchiolitis in infants and asthma in adults are two of the major respiratory diseases in industrialized nations. Indeed, bronchiolitis is the leading cause of infant hospitalization in the US, accounting for 107,000 infant hospitalizations each year. Besides, previous literature suggests that bronchiolitis during infancy is one of the major risk factors for subsequent asthma. Likewise, approximately 26 million Americans have asthma. Asthma exacerbations account for a substantial proportion of the personal and societal burden, leading to approximately 340,000 hospitalizations annually.
Paper 1: The literature has demonstrated a high variability in bronchiolitis management across the nation, which is, at least partially, attributable to the variable clinical course in this population. Despite the development of prediction scoring models (e.g., logistic regression models), identifying the subgroup of infants with bronchiolitis who require higher acuity care (e.g., positive pressure ventilation, intensive care unit [ICU] admission) remains an important challenge. The difficulty and uncertainty of predicting acute severity—and, consequently, the appropriate level of care for infants with bronchiolitis—are reflected by the well-documented variability in inpatient management across the nation. Recently, machine learning models have gained increasing attention because of their advantages, such as the ability to incorporate high-order, nonlinear interactions between predictors and to yield more accurate and stable predictions. While machine learning approaches may enhance the predictive ability, little is known about their utility to predict acute severity outcomes in infants with bronchiolitis. We developed advanced machine learning models using the data from a well-characterized prospective cohort study.
Paper 2: Previous research has demonstrated that patients with chronic asthma have a high incidence of cardiovascular events. Besides, the emerging evidence suggests that acute inflammatory processes (e.g., bacteremia, pneumonia, influenza virus infection, Streptococcal infection, and acute exacerbation of COPD) have been linked to acute cardiovascular outcomes. However, despite the clinical and research importance of asthma exacerbations, little is known about its acute effect on cardiovascular outcomes. Herein, using large population-based data of four diverse states in the US, I conducted a self-controlled case series study to investigate the acute causal effect of asthma exacerbation on cardiovascular events.
Citable link to this pagehttps://nrs.harvard.edu/URN-3:HUL.INSTREPOS:37365248