Objective. To analyze geographic variability in the utilization and co
st of post-stroke medical care using multiple linear regression. Data
Sources/Study Setting. A 20 percent random sample of Medicare benefici
aries with an admission to an acute care hospital for stroke during th
e first six months of 1991, supplemented by data from their Medicare c
laims and beneficiary records, the Medicare Cost Reports for hospitals
and nursing homes, and the Area Resource File. Study Design. Weighted
least squares regression is used to analyze variations in poststroke
practice patterns across 151 MSAs (Metropolitan Statistical Areas). Av
erage post-stroke costs, utilization rates, and facility lengths of st
ay are regressed on patient and market characteristics. Data Collectio
n/Extraction Methods. For a six-month post-stroke interval, beneficiar
y-level post-stroke costs and service utilization are averaged by MSA.
Variables describing market conditions are then added to these MSA-le
vel records. Principal Findings. Patient variables rarely explain more
than a third of practice variation, and often they explain substantia
lly less than that. Market variables (with some exception) tend to be
relatively less important. Finally, one-half to two-thirds of the prac
tice variation across MSAs is unexplained by the patient and market fa
ctors measured in our data Conclusions. A substantial portion of inter
-MSA variability in utilization and intensity of post-stroke rehabilit
ation services cannot be explained by differences in patient character
istics. Given the large practice differences observed across MSAs, it
seems unlikely that unmeasured patient differences can account for muc
h more of the practice differences.