Hospital Payment Incentives and Opioid Prescribing

Project Outcomes Statement

There are two major outcomes of this research project. One is accomplishing the three specific aims in terms of providing some of the first analyses to answer the questions posed. The second major outcome of the research project is the knowledge created (in the form of replication code that is shareable) in assembling a rich and unique database, combining five data sources, which involved identifying and innovating algorithms for measuring concepts related to the specific aims. The replication codes created are a valuable product for ensuring further innovations that build on the outcomes of our exploratory R21. We detail each of these outcomes below.

1) This research project assesses whether, and to what extent, incorporating patient experience into Medicare’s hospital reimbursement, as part of the Hospital Value Based Purchasing (HVBP) Program, impacted opioid use in the shorter and longer-terms post-hospital discharge. Our findings indicate that the shift towards pain scales being incorporated in payment formulas do not appear to have increased prescriptions for opioids. This conclusion results from our analysis of the analytic database, in which we detailed the prevalence of opioid use measures among opioid-naïve beneficiaries who had an inpatient visit, separating surgical and non-surgical admissions. We estimated difference-in-differences models, comparing trends in opioid prescribing measures before and after the implementation of the HVBP program, across hospitals participating in HVBP (treatment group) and those which are exempt (critical access hospitals and/or hospitals in MD). We then assessed heterogeneity in these effects across various margins related to hospital characteristics, hospitals’ position along the total performance score (TPS) distribution, and patient and provider characteristics. These findings represent the first empirical analysis of how incorporating patient experience of care metrics as part of HVBP is impacting post-discharge opioid prescribing. As we find no evidence thus far of an increase in opioid prescribing post-discharge for opioid-naïve patients in the average treated hospital, we have considered possible explanations. One possibility is that the average response may be very small. We have conducted additional analysis to zero in on a small subset of hospitals for whom incentives to improve scores may be largest. Another possibility is that the time frame of our data (ending in 2017), five years post-HVBP adoption, may not be long enough for effects to yet materialize. Early incentives were small (they were 1% in 2013, 1.25% in 2014, 1.5% in 2015, 1.75% in 2016). It could be that corresponding incentives in HVBP to improve clinical outcomes and readmission rates are also reducing opioid prescribing, which in turn may explain why there are no discernible increases in opioid prescribing and related adverse health events following the adoption of the HVBP program.

2) Our outcome of creating a unique and rich database to enable further research on hospital-based regulatory factors that might influence opioid prescription is detailed below. We constructed through several painstaking processes a unique database that we share and disseminate (in the form of replication code with our analysis). The main data base is the Medicare claims data hosted at the NBER. We supplemented these data with the American Hospital Association (AHA) data, which are also available through the NBER. The Medicare data (20% sample of Medicare beneficiaries) includes Part A inpatient claims (MedPAR), Part B outpatient (carrier) claims, Part D (prescription drugs) claims, and Medicare Beneficiary Summary File (MBSF; for patient demographics). Newly created and identified variables for our analysis included categorizing the current procedural terminology (CPT) codes (allowing us to identify surgical vs. nonsurgical episodes in the MedPAR claims), hospitalization codes (to define inpatient and ED admissions in the MedPAR), diagnoses codes to identify substance use disorders in the carrier claims post hospital discharge, and ICD-10 codes to identify relevant causes of death in the National Death Segment of the Medicare Beneficiary Summary File. Products of this project include analytic program files to code/clean/merge the records across the Part A, B, D and MBSF datasets, along with AHA data on hospital characteristics related to ownership, provider composition, and patient composition. We developed an algorithm for identifying hospitals that are fully integrated with physician organizations from those that are only partially or not at all integrated. Additional data were processed from CMS’ Hospital Compare dataset, wherein we downloaded and put together a dataset of hospital performance and the total performance score (TPS) for each hospital. These data were matched, by hospital / provider ID, to the Medicare files. Fourth, we obtained data on providers from the National Plan and Provider Enumeration System (NPPES). We coded up these data for provider characteristics, and merged with the Medicare files by provider ID. Fifth, we updated our prior datasets of county-level health care resources (population and demographics, health market characteristics such as hospital beds, physicians per capita by specialty, health professional shortage areas) with newer data from various sources including the American Community Surveys and the Area Health Resource File. We also updated our prior datasets of state policies related to opioids from various sources to include state websites, PDAPS, and primary legislative search. Finally, we assembled the requisite information for forming our key outcome measures related to Rx opioid use, inappropriate prescribing, and adverse events. Specifically, we identified through the therapeutic class code (TCC) and the national drug code (NDC) opioid products, and classified further sub-types (shortacting, long-acting, schedule II, schedule III). We gathered the requisite information for classifying opioid overdose events through diagnoses, procedure, and death codes. Our replication files for this process provide a valuable resource to the research community to study opioid prescribing associated with hospital inpatient admissions.