From what areas does a hospital draw on to fill its beds? There have been many attempts to define a hospital’s catchment area. The Dartmouth Atlas Group uses hospital referral regions (HRRs) and hospital service areas (HSAs). One method is to determine a minimum admission rate for a given geographic unit (e.g., county, census tract, zip code). For instance, a given zip code would be placed in a hospital’s catchment area if that zip code made up at least 0.5% of hospital admissions. Conversely, one could include all areas where at least a certain percent of resident admissions were to the hospital in question.
A paper by Gilmour (2010) examines how to create a hospital catchment area using K-means clustering. The goal of this process is to assign local authority districts to hospitals based on the how likely the individuals are to visit a certain hospital. K-means clustering is used to partition n observations into k clusters in which each observation belongs to the cluster with the nearest mean. The author applies the standard K-means clustering algorithm as follows:
- Two cluster centers are chosen arbitrarily,
- Each observation is assigned into the cluster whose center it lies closest to,
- The center of the cluster formed by this assignment is recalculated, and
- The process is repeated until the cluster assignments cease to change.
Gilmour uses a multivariate approach to estimate “closeness.” He uses principal components analysis to incorporate additional information such as the size and distribution of the hospital’s activity.
Although the K-means clustering captures a larger share of the hospital’s admissions, the catchment areas are generally much larger than is the case using the marginal methods.
- Stuart John Gilmour (2010), Identification of Hospital Catchment Areas Using Clustering: An Example from the NHS. Health Services Research, 45: 497–513. doi: 10.1111/j.1475-6773.2009.01069.x