Measuring the Impact of Disasters Using Publicly Available Data: Application to Hurricane Sandy (2012)

The unexpected nature of disasters leaves little time or resources for organized health surveillance of the affected population, and even less for those who are unaffected. An ideal epidemiologic study would monitor both groups equally well, but would typically be decided against as infeasible or costly. Exposure and health outcome data at the level of the individual can be difficult to obtain. Despite these challenges, the health effects of a disaster can be approximated. Approaches include 1) the use of publicly available exposure data in geographic detail, 2) health outcomes data-collected before, during, and after the event, and 3) statistical modeling designed to compare the observed frequency of health outcomes with the counterfactual frequency hidden by the disaster itself. We applied these strategies to Hurricane Sandy, which struck the northeastern United States in October 2012. Hospital admissions data from the state of New York with information on primary payer as well as patient demographic characteristics were analyzed. To illustrate the method, we present multivariate logistic regression results for the first 2 months after the hurricane. Inferential implications of admissions data on nearly the entire target population in the wake of a disaster are discussed.