Organizational Change, Payment Reform, And The Value Of Hospital Care

Thumbnail Image
Degree type
Doctor of Philosophy (PhD)
Graduate group
Managerial Science and Applied Economics
Health and Medical Administration
Grant number
Copyright date
Related resources
Martinez, Joseph

This dissertation applies economic, program evaluation, and machine learning tools to study productivity in the hospital sector. The market for hospital care has been transformed by two broad developments over the previous decade. First, traditionally independent physicians and hospitals have increasingly opted to form integrated organizations, consolidating control over both in a single firm. Second, provider payments have increasingly become linked to quality or value, generating new incentives to reduce spending and improve "value." The first half of this dissertation describes new data on physician-hospital vertical integration (VI), changes in quality and spending due to VI, and whether VI promotes savings under value-based payment. The second half of this dissertation documents causal effects, selection bias, and treatment effect heterogeneity under voluntary and mandatory payment programs. The following four chapters contribute to an important literature on hospital productivity and its relationship to organizational form and the payment system. (i) Vertical Mergers & Firm Performance. I estimate the effects of VI on quality of care and spending during and after acute hospitalization. To ameliorate concerns about patient sorting, I use an instrumental variables strategy leveraging persistent referral patterns. On average, I find no evidence that financial integration changed quality of care or spending for hospitalized patients. However, I find that average effects of VI mask substantial heterogeneity by patient illness severity. For patients in the top decile of severity, VI reduced mortality by 2 percentage points without increases in spending. However, spending increased by 4% for patients in the bottom 40% of the severity distribution without mortality benefit. (ii) Vertical Integration & Savings in Bundled Payments. This study examines the association between VI and treatment effects of bundled payments estimated from a randomized control trial of mandatory bundled payments using a difference-in-differences framework. Institutional spending per joint replacement fell more for treated hospitals on average. These changes in spending were not associated with VI. The lack of association suggests that any advantage integrated hospitals have in coordinating care may not be necessary for success in bundled payments. (iii) Selection & Causal Effects in Voluntary Programs. Most payment reform programs allow participants to opt in; this can improve productivity due to selection on treatment gains but may limit validity of estimated effects due to confounding. We study both in the setting of a large voluntary bundled payment reform in Medicare. We use variation in costs from idiosyncratic program rules to estimate marginal treatment effects. Our results suggest there was minimal sorting on unobserved treatment gains but meaningful selection bias due to confounding. Observable factors cannot explain more than 30% of this bias, illustrating the challenge of assessing voluntary programs. (iv) Personalized Payment Reform? I analyze the extent to which observable characteristics predict the returns to participation in value-based payment. I leverage a field experiment of bundled payments and machine learning tools to model variation in effects across participants. Spending reductions due to bundles vary significantly – cases predicted to have the largest reductions saved $2,700 more than those that saved least. While I find no effect on average quality, complications were reduced by 2 percentage points for the group most affected by the program. Evidence suggests that vulnerable patients receive the most benefit, a finding which holds for both spending and complications.

Atul Gupta
Date of degree
Date Range for Data Collection (Start Date)
Date Range for Data Collection (End Date)
Digital Object Identifier
Series name and number
Volume number
Issue number
Publisher DOI
Journal Issue
Recommended citation