Complementarities in Quality Provision within Hospitals: Evidence from a Pay-for-Performance Program

Many healthcare policies aim to increase quality by incentivizing providers to improve health outcomes of patients. These policies often target patients in specific insurance groups. Since providers often treat various patients with different sources of insurance coverage, these insurance group-specific policies result in providers facing a mix of incentives. This paper examines whether asymmetric incentives across insurance groups have effects on targeted as well as non-targeted patients treated by the same provider.

To do so, I study the impacts of the Hospitals Readmissions Reduction Program (HRRP), a pay-for-performance scheme implemented under the Affordable Care Act. The HRRP aims to improve quality by financially penalizing hospitals with excess 30-day risk adjusted readmission rates among Traditional Medicare (TM) patients admitted for heart attack (AMI), heart failure (HF), or pneumonia (PN). The design of this program creates asymmetric incentives across patient groups: the hospital’s incentive to improve quality for a patient depends on their insurer, reason for admission, and date of admission.

I begin by developing a model where hospitals treat patients with different insurance policies and payments for hospital services differ by insurance groups. This model delivers two main predictions about how hospitals will respond to the HRRP: (1) hospitals will respond by improving quality for TM patients if they expect additional readmission to increase penalties and (2) the impact of the program on non-targeted patients depends on the degree of complementarities in quality provision between different insurance groups. Drawing on the variation created by the program, I test the predictions of this model using a matched difference-in-differences strategy. For each TM and non-TM patient admitted for a monitored diagnosis (AMI/HF/PN) in New York State, I match them with an observably similar patient who has the same insurer and is treated at the same hospital, but is admitted for a non-monitored condition.

In line with the first prediction, there is strong evidence that hospitals respond by reducing readmission rates among TM patients. Patients admitted for the two cardiovascular conditions monitored by the program, heart attack and heart failure, see large and statistically significant reductions in readmission probabilities. For example, readmission rates among targeted AMI and HF patients fell by over 2 percentage points (pp) and 1pp respectively. The impacts of the program on TM patients admitted for pneumonia are smaller. The program also had significant effects on non-targeted patients. For example, readmission rates among non-TM AMI patients fell by over 1pp. In light of the second prediction, this suggests that are complementarities in quality provision. Furthermore, there is little evidence that the reductions in readmissions are driven by various forms of hospital gaming, such as increased use of emergency services and delayed readmissions outside the 30-day window. The results therefore suggest that hospitals improved quality of care for both targeted and non-targeted patients.

Finally, to examine the mechanisms driving these results, I estimate several parameters of the model including the degree of cost-complementarities and the cost of quality provision. The parameter estimates also indicate complementarities in quality provision across insurance groups.