Predicting residential exposure to allergens in an electronic health record cohort of children with asthma in Massachusetts, USA
Year: 2021
Type: conference-abstract
Abstract: BACKGROUND AND AIM: Children with asthma spend much of their time indoors where they can be exposed to numerous allergens that exacerbate symptoms. Generating systematic knowledge in a clinical setting of whether residential triggers are present can be resource intense (e.g., home visits, surveys). In addition, people’s surrounding environments are complex mixtures, making identifying key elements affecting health-related outcomes a challenge. Recent advances in electronic health record (EHR) data use and novel statistical methods provide opportunities to predict allergen presence indoors. METHODS: We developed prediction models for cockroach, rodent (mice or rats), and mold presence using data from an asthma intake form among n=1,767 children with asthma who visited Boston Medical Center from 2004-2015. Multilevel predictors (p=72) included individual, housing, and neighborhood factors that were linked to geocoded patient addresses from EHRs and modeled using an ensemble machine learning approach. To estimate effects, we reduced dimensionality by regularized regression and then simulated our resultant ensemble model across exposure levels via G-computation, a causal inference method. RESULTS:Prediction models had reasonable discrimination in classifying allergen presence by area under the receiver operating curve (AUC: 0.64). For cockroaches, estimated naïve average causal effects (NACE) of tax-exempt parcels consistent with large public or affordable housing were negative (NACE: -0.033; 95%CI: -0.061, -0.004). For rodents, effects were only suggestive of statistically significant relationships. The protective relationship with tax exempt parcels likely captures proxy effects, such as integrated pest management plans commonly implemented at large public and affordable housing complexes. CONCLUSIONS:Integrating novel clinical and geographical databases with cutting-edge statistical methods, we predicted population-wide allergen presence and identified potentially salient components of surrounding environmental mixtures. This approach can provide clinicians and epidemiologists information on individual environmental exposures for informing potential interventions and epidemiologic EHR-based cohort creation. KEYWORDS: asthma, allergies, exposure assessment, mixtures
Source: ISEE Conference Abstracts
Institution: Boston University
Language: English
Subfield: Physiology
Field: Medicine
Domain: Health Sciences
Open Access status: diamond
OpenAlex ID: W3195614177