Asthma impacts more than 25 million adults and children in the U.S. with high associated morbidity and socioeconomic disparities in outcomes. Because effective medications are available to treat and prevent exacerbations of asthma and evidence-based interventions exist to mitigate the impact of harmful socioeconomic factors, early identification of those at highest risk is crucial. However, efforts to predict future exacerbations of asthma have yielded modest results with infrequent inclusion of comprehensive information on social hardships, such as food insecurity and housing instability, or financial hardships, such as difficulty affording the costs of controller medications which is particularly relevant for those with private health insurance. Identifying social and financial hardships requires broad-based screenings which are resource intensive, difficult to implement in clinical settings and often incomplete or limited to care seeking populations. Further, few asthma risk prediction modalities incorporate time-variable (temporal) data on important social, clinical, and environmental factors. Machine learning, an advanced computational approach to risk prediction, has great potential to improve upon conventional approaches to risk prediction of asthma exacerbations through indirect estimation of social hardships and inclusion of temporal risk factors. Implementation of enhanced asthma risk-prediction models in a health plan setting offers distinct advantages due to existing investments in asthma care management and access to timely claims data across the full care continuum. Accordingly, the aims of the SPACER study (Sociomarkers to Predict Asthma Control and Emergency Room visits) are 1) To describe social and financial hardships in privately insured adults and children with asthma, and association with medication adherence and exacerbations, 2) To indirectly estimate self-reported social and financial hardships using routinely collected health plan and spatial data, and 3) To develop and validate a machine learning network model, incorporating temporal sociomarker, clinical, and environmental data, to predict asthma exacerbations in a health plan setting. The research leverages the unique research environment of the Department of Population Medicine, an academic research department of Harvard Medical School, situated in a regional non-profit health plan, Harvard Pilgrim Health Care. The mentored career development award will support Dr. Alon Peltz, a physician and health services researcher, in developing expertise in machine learning modeling and use of social data to improve prediction of adverse clinical outcomes.
The objective of this proposal is to develop new methods for asthma risk prediction in a health plan setting by incorporating social and financial hardship data, temporal features, and machine learning. Applications of this research will directly enable U.S. health plans to better identify adults and children for enrollment in asthma care management programs in order to moderate their combined clinical and social risks for poor asthma quality. This career development award will support Dr. Alon Peltz, a physician health services researcher, in developing expertise in machine learning risk prediction incorporating expanded social data.