Geographic distribution of hospitalization rates, case fatality, and mortality from stroke in the United States.

We analyzed state-specific stroke-hospitalization, case-fatality, and mortality rates for the US Medicare population for 1989, using national data resources of the Health Care Financing Administration (HCFA), the National Center for Health Statistics, and the Bureau of the Census. State-specific hospital admission rates for stroke ranged from 0.66 to 1.26%, compared with the national value of 0.94%. Both hospital-usage rates and deviations of observed rates from predicted values (based on statistical models of the HCFA) showed significant spatial autocorrelation, with high rates clustered in the southeastern United States and low rates clustered in the Mountain census division of the West and also somewhat in the Northeast. Case-fatality rates increased nationally from 14.9% at 15 days after hospital admission to 31.2% at 180 days after hospital admission. State-level case-fatality rates showed relatively little interstate variation and no clear or consistent spatial pattern, although there was statistically significant spatial autocorrelation at several intervals after hospital admission. Admission rates and case-fatality rates were not significantly associated at any interval after admission to 180 days, suggesting that variation in case-fatality rates was not simply a result of differences in severity-of-illness thresholds for hospital admission. State-specific stroke-mortality rates ranged from 294.5 to 523.5 per 100,000 population, compared with the national value of 415.3 per 100,000 population. State-specific mortality rates for stroke showed significant spatial autocorrelation, with high rates clustered in the South and low rates clustered in the Northeast and the Mountain census division of the West. The spatial distribution of stroke-mortality rates strongly resembled the spatial distribution of hospitalization rates but did not resemble the spatial distribution of case-fatality rates at any interval from 15 to 180 days after hospital admission. Indeed, in univariate spatial-regression models fitted to the data using a maximum likelihood procedure and weighted for non-constant variances, the best predictor of state-level stroke-mortality rates was the hospital-utilization rate for stroke; attempts to improve the model by including case fatality at various intervals and interaction terms did not yield a significant improvement. These data suggest that factors determining stroke occurrence and hospital utilization are more important than factors determining case fatality in terms of explaining the long-standing distribution of stroke mortality in the United States. Factors affecting only case fatality but not hospitalization, such as the quality of medical care provided in the hospital, cannot explain the geographic distribution of stroke mortality in the United States.