Association for Public Policy Analysis & Management

Objective: Under the Oncology Care Model (OCM), oncology practices volunteered to be accountable for financial and quality outcomes for cancer patients’ treatment episodes that are triggered by chemotherapy administration. A valid counterfactual is needed to understand and evaluate the model’s impacts. We discuss the use of propensity score matching (PSM) to create the comparison group for OCM and evaluate the representativeness of the practices that elected to participate.

Design: We constructed a comparison group of oncology practices using nearest neighbor PSM methods, accounting for practice, market, and patient characteristics, and evaluated covariate balance and parallel trends. Using Medicare claims, we identified six-month cancer episodes initiated by chemotherapy and attributed episodes to unique practices using program rules. Data from SK&A were used to identify practice ownership and affiliation. Descriptive and multivariate analyses quantified patterns in the characteristics, utilization, and Medicare spending for episodes attributed to OCM practices relative to comparisons and to a national sample of oncology practices.

Population: The population studied consisted of Medicare beneficiaries undergoing cancer chemotherapy as part of two million episodes attributed to 2,148 oncology practices nationwide in the baseline period 1/2012-6/2015. We compared 190 OCM practices (691,885 episodes) and 319 matched comparisons (477,390 episodes), along with an additional 1,639 oncology practices in neither group (830,725 episodes).

Findings: OCM attracted the largest oncology practices in the nation, and standard PSM approaches failed to fully balance size-related covariates. Data transformations improved balance. With the exception of practice size (provider and episode count), OCM and comparison practices were statistically similar across market and practice characteristics (e.g., supply of providers and facilities, demographics). Nearly all OCM and comparison practices are located in urban markets, as are oncology practices nationally. OCM practices are more likely to operate in multiple sites and are more often affiliated with academic medical centers than comparison practices. Parts of the U.S. are underrepresented in both OCM and comparison groups.

OCM episodes were similar to comparison and national episodes in terms of cancer mix and patient race, ethnicity, age, and gender during the baseline period. In the majority of episodes, beneficiaries had Part D coverage, and fewer than 15% of the episodes were for dual-eligible beneficiaries. Utilization and Medicare spending during the baseline were higher for OCM than comparison episodes. For example, average total Medicare spending per episode for Parts A, B, and D was $27,386 for OCM practices and $26,234 for comparisons. Trends in key measures were similar between the two groups during the baseline period.

Conclusion: The participating practices and episodes are similar to the matched comparison group across numerous demographic and market characteristics. The differences that do exist highlight the need to account for practice characteristics, such as size, in future evaluation analyses.

Implications: OCM practices receive incentives based on their ability to achieve Medicare savings while maintaining (or improving) quality, relative to baseline benchmarks. It is important to understand key characteristics of OCM practices to assess the generalizability of final evaluation results.

Funding: Section 3021,Affordable Care Act