Projecting Local Healthcare Use in the US through 2030

To best model future health workforce needs, an accurate projection of future demand for healthcare is necessary. Previous models have often been hampered by a limited number of distinct types of healthcare use, large geographic areas, and/or choice of parametric model to generate use estimates.  We developed an alternative model.  We use multiple years of the Medical Expenditure Panel Survey and classify visits into three settings (office-based/outpatient; inpatient; emergency room) and up to 19 clinical service areas (modified from the AHRQ’s Clinical Classification System, plus preventive visits) for a total of 55 types of healthcare services. For example, we model the use of “inpatient circulatory”, “outpatient endocrine”, and “emergency room injury” visits.   We then use multiple external contextual data sources to reweight the MEPS data to represent the sociodemographics, socioeconomics, and health behavior (e.g. smoking) of each U.S. county’s population through 2030.   These county-specific population weights are then combined with MEPS visit data to generate the healthcare use projected for each county through 2030.    This non-parametric approach, although computationally intensive, has the benefit of generating estimates independent of statistical assumptions on the data generating process for healthcare use.  We compare these projected use rates with current observed rates and their proxies, examine geographic variation in these projections over time, and simulate the effect of various policy and environmental changes (e.g. health insurance coverage; prevalence of obesity) on healthcare use.  We identify geographic areas with the largest changes in healthcare use resulting from the simulated effects.