The Effects of Global Budgets on Hospital Care Quality in Maryland
Methods: I used data on all discharge abstracts in Maryland hospitals from the Healthcare Cost and Utilization Project (HCUP) State Inpatient Databases (SID) for years 2007-2012. I compared the treatment hospitals with two different control groups. The main control group consists of three rural hospitals that were eligible for the program but declined participation. The secondary control group consists of all general acute care hospitals in the state except for the state's teaching hospitals. Using AHRQ methodology, I calculated observed and risk-adjusted rates and confidence intervals for the Prevention Quality Indicators (PQI) for the treatment and control groups at the provider and area levels. I compared changes in rates pre- and post-intervention using a difference-in-differences (DID) approach with hospital-level fixed effects and controlling for secular trends. I also performed non-parametric tests to test for statistically significant differences. Finally, I performed sensitivity analyses including placebo tests assigning "fake" TPR implementation dates and randomization tests to ensure other factors not related to the intervention were not responsible for the observed effects.
Results: TPR hospitals were slightly smaller and had lower volume than the main control group the year before the intervention (151 vs. 196 staffed beds, 10,994 vs 13,369 total discharges on average). Their population also had a higher proportion of white and older patients on average (58.4 vs. 55.8 years average age in 2007). The risk-ajdusted area-level PQI90 composite indicator decreased in the treatment group from 21.72 (CI = [20.83-22.61]) in 2007 to 14.92 (CI = [14.05-15.77]) admissions per 1,000 population. It also decreased in the main control group from a lower base of 17.50 (CI = [16.84-18.15]) to 13.83 (CI = [13.23-14.44]). These trends were driven by both the acute composite indicator (PQI91) and the chronic composite indicator (PQI92). Although the patient-level DID analysis found that the differences were statistically insignificant at the 0.05 level, comparing the changes in risk-adjusted area-level rates (as recommended by AHRQ documentation) using non-parametric tests finds that the results are statistically significant. Overall, the TPR program associated with a higher decrease in admissions for ambulatory care sensitive conditions in rural Maryland counties compared to controls. Sensitivity analyses provided additional evidence that the TPR program was responsible for these effects.
Conclusion/Implications: Hospital global budgets are a promising policy for decreasing preventable hospitalizations in rural hospitals. However, they may not be applicable in all states due to Maryland's unique all-payer rate-setting system. Further research is needed to evaluate the program's effects on total costs and utilization of inpatient services.