Abstract
INTRODUCTION: Despite the availability of descriptive clinical and disaster data, the associations between disaster type and clinical morbidity remains unknown. The Federal Emergency Management Agency (FEMA) responds to disasters in the United States by mobilizing federal resources to support state and local emergency response efforts. The Centers for Medicare and Medicaid Services (CMS) provides payment services for covered individuals and activities in the United States. The association of CMS phenotypes of service utilization (i.e. Essential hypertension, Coronary atherosclerosis [Atherosclerotic heart disease], Herpes simplex) with incident types (i.e. Hurricane, Fire, Terrorism) remains under described. METHODS: Historic Emergency Disaster Declarations were retrieved from FEMA API, and the resulting records were transformed to the study data standard. CMS claims were sourced from Chronic Conditions Warehouse and transformed to the study data standard. Records were then merged by county and month to create a Per-Member Per-Month Per-County phenotype model of CMS service utilization with FEMA county-month declaration detail. A generalized linear model was produced using H2o.ai predicting Per Member, Per Month, Per County attack rate within service phenotype, and interactions between phenotype and disaster type produced z-values, standardized coefficients and p-values to assess the association or disassociation of the interaction between phecodes and incident types. RESULTS: Specific disaster declaration types have high coefficient (associated) and low coefficient (disassociated) relationships. These relationships for specific phenotypes show marked consistencies over time across named disaster declarations within disaster type. CONCLUSIONS: CMS claimants seek or avoid specific phenotypes of services statistically attributable to disaster incident types. FEMA disaster exposure is both an accelerant of care seeking and a barrier to seeking kinds of care (especially STD care). Future research should control lag, lead, distance, migrancy, impact and duration effects to improve association detection.