Asthma affects 12 million Americans, but the contributions towards asthma exacerbations do not equally affect all populations. African Americans (AA) are four times more likely to be hospitalized and five times more likely to die from asthma than European Americans. In addition, after being diagnosed with asthma, AAs have poor control leading towards more exacerbations. We developed a series of Machine Learning (ML) models including Logistic Regression (LR), Decision Tree (DT), eXtreme Gradient boosting (XGB) and Random Forest (RF), to determine asthma disparities between AA and EA patients in the Cincinnati Metropolitan area (Figure 1).
We leveraged data from University of Cincinnati (UC) electronic healthcare records (EHR) from the Emergency Department, pollution data from the Environmental Protection Agency (EPA) and environmental data from the Ohio Air Quality Agency (OAQA) between November 2012 and December 2017. We linked EHR clinical data together via unique deidentified identification numbers for a cohort of 42,374 patients (Table 1). We used the recorded entry date to merge the pollution, mold, and EHR data sets. This dataset was cleaned and fed to multiple machine models to predict both asthma severity and emergency visit disparities. FEV1% Values classified asthma severity, FEV1% ≥ 80, moderate (M), FEV1% 60-80, and severe (S) FEV1% < 60. Emergency Department (ED) visit frequency was counted for each person’s encounter in the ED in the time period. ED visitations below and at the third quartile for total visits by the cohort were classified as intermittent (I) visitors and visit frequencies above the third quartile as frequent (F) visitors.
AAs samples were characterized by severe asthma and higher ED visits than their EA counterparts. AAs had a higher risk for environmental exposure with more elevated spores from mold and pollen than EAs. XGB produced the best predictive model for both AA and EA ED visit and severity (Figure 2). While pollution affected both groups, a higher predictive model was obtained for AAs than EAs. Environmental factors may contribute towards the disparities between races in asthma-related ED visits, and attention should be given to address these disparities.
Learning Objective: At the conclusion of this activity, participants will be able to:
Establish a machine learning pipeline for asthma severity and emergency department visits through environmental and electronic healthcare record data.
Distinguish asthma contributing environmental factors between African Americans and European Americans.
Adeboye Adejare (Presenter)
University of Cincinnati
Yadu Gautam, Cincinnati Children's Hospital Medical Center
Mekibib Altaye, Cincinnati Children's Hospital Medical Center
Tesfaye Mersha, Cincinnati Children's Hospital Medical Center