With mid-pandemic zip code, demographic data, and ICD-10 diagnosis codes to identify COVID-19 patients, a machine learning prediction approach will be used to build a model that can be applied to an early pandemic sample of patient data. The model will classify which respiratory admissions were actually COVID-19 related. Hospital specific quality will then be estimated using a two-stage model. In the first stage, patients with COVID-19 symptoms choose a hospital for treatment. In the second stage, patients receive treatment, and a mortality outcome is observed. The model allows for both the mortality outcome and patient choice to be a function of patient illness severity which is unobserved to the econometrician. Combined data and model estimates will measure the mortality consequences of hospital selection and hospital quality. We will then investigate how hospital quality levels for COVID-19 improved over time and evaluate the extent of learning within a hospital system, from other local hospitals, and across geographic areas in the U.S.
Hospital Preparedness and Health Outcomes
Pillar 2
Latent effects on industries and economy
Need
Variation in reported mortality measures during the COVID-19 pandemic suggests that some U.S. hospitals were better prepared to deal with the influx of patients and medical uncertainty than others. However, observational comparisons do not account for the unobserved severity of patients seen at a given hospital, which is a function of hospital quality and patient’s choice of hospital taken together.
Proposed Solution
Use inpatient claims data to analyze hospital quality responses and the subsequent effect on health outcomes to better understand the types of hospitals that performed well when treating COVID-19 patients, and to ascertain the path of learning by which hospitals improve outcomes.