Identification of Computational Phenotypes in Pediatric Status Asthmaticus and Determination of Treatment Effect

Project Summary Pediatric asthma is a growing epidemic associated with significant burden to the healthcare system which lacks individualized treatment approaches. Although standard treatments for pediatric status asthmaticus are adequate for most patients, a subset do not respond quickly. These children are at risk for severe physiologic derangements including hypercarbia, hypoxemia, cardiac dysfunction, and death. Adjunctive asthma therapies can potentially provide benefit to children with status asthmaticus; however, there is limited evidence regarding their efficacy. Not only are these adjunctive therapies potentially ineffective, but many are associated with severe side effects. My central hypothesis is that children presenting to the hospital with status asthmaticus can be effectively subdivided into distinct clinical phenotypes that exhibit heterogeneous treatment effects. I propose to test this hypothesis by (1) identifying distinct clinical phenotypes in children presenting to the hospital with status asthmaticus, (2) validating the clinical phenotypes of status asthmaticus in other pediatric institutions, and (3) comparing effects of the adjunctive asthma treatments magnesium and aminophylline between clinical phenotypes. With the preliminary data generated, new skills obtained, and expertise gained through this K01 award, I plan to construct a robust computational phenotype for pediatric asthma, identify the patients most likely to have improved clinical outcomes with specific asthma therapeutics, and decrease the health burden of children with status asthmaticus in a future R01 study. During the award period, I will continue to conduct research under the primary mentorship of Dr. Eneida Mendonca, who is a pediatric critical care physician and established health informatics researcher, as well as the co-mentors and advisors outlined in the career development plan. I will obtain additional training in artificial intelligence and machine learning methods such as unsupervised clustering methods, health informatics and information systems, and comparative effectiveness research methods through both formal coursework as well as established institutional programs. With the preliminary phenotyping data generated by this study, the experience gained using health informatics and machine learning methods, and a knowledge foundation of comparative effectiveness research methods, I will be well-positioned to transition to research-independence.