INTRODUCTION

Facing pressure to better control health care spending and address inefficiencies in the delivery system, public and private payers have been experimenting with alternative payment models. One model that is gaining traction is episode-based (or bundled) payment (EBP). Under EBP, there is a single spending target for all services delivered during an episode of care, and these services span time, sites of care, and specific providers. For example, an episode might include a surgical procedure and all related services (e.g., pre-operative office visits, post-acute care, readmissions) that the patient receives during a defined time period (e.g., 30 days) before and after the procedure. By aggregating reimbursement across these services and settings, EBP (in theory) incentivizes efforts to improve the efficiency and coordination of care.

The emerging literature finds mixed results regarding the impact of EBP on spending, with the effect varying across the specific procedures and conditions that are bundled. For example, there is evidence1, 2 of savings under Medicare’s voluntary Bundled Payments for Care Improvement (BPCI) initiative for lower extremity joint replacement episodes, but not for several other conditions.3, 4 While initial results5 from Medicare’s mandatory Comprehensive Care for Joint Replacement (CJR) model show no significant change in total spending per episode, more recent work6 finds a decrease in institutional spending per episode. In the commercial sector, an evaluation of EBP for perinatal care in Arkansas estimates that total episode spending decreased during the first year of full implementation.7

Beyond its impact on spending, analysts have raised concerns about the impact of EBP on the number of episodes.1, 4,5,6, 8,9,10,11,12,13,14 Providers can adjust episode volume because they often exercise discretion over whether to perform a procedure and over patient referrals. These decisions can also be influenced by financial incentives. Under EBP, providers can earn bonus payments if they increase the number of patients treated with spending below the payer-specified target, which can be achieved by either reducing per-episode spending (e.g., through efficiencies during care delivery) or selectively treating low-severity patients (who have lower expected costs). Conversely, because bonuses are based on average spending, providers can also benefit if they treat fewer complex patients (who have higher expected costs). The net effect on procedure volume is ambiguous. However, if providers encourage patients to undergo procedures that are not clinically necessary, there could be over-treatment that then offsets any per-episode savings. At the same time, there could be under-treatment if providers avoid treating sicker patients.

There is limited research on the volume effects of EBP, and the existing work focuses on lower extremity joint replacement.5, 6, 14 However, the evidence on how EBP impacts spending varies across episodes, suggesting that the impact on volume may also vary by condition. Thus, it is crucial to expand the set of episodes analyzed. Furthermore, most research to date evaluates Medicare EBP models. It is therefore important to understand how providers might respond to EBP in other patient populations.

In this study, we examine the impact of EBP on procedure volume for different types of episodes. Our empirical setting is Arkansas, where the state’s Medicaid program and two of its largest commercial insurers began adopting EBP for a variety of clinical conditions in 2012. Arkansas’ EBP model uniquely features statewide implementation, multi-payer involvement, mandatory provider participation, and novel episodes.15

EPISODE-BASED PAYMENT IN ARKANSAS

Launched in 2012, the Arkansas Health Care Payment Improvement Initiative is an attempt to move the state’s public and private payment systems toward value-based purchasing. One of its core components is an EBP model, in which Arkansas Medicaid, Arkansas Blue Cross Blue Shield (AR BCBS), and QualChoice (QC) jointly participate. Prior to EBP, these insurers reimbursed providers on a fee-for-service basis.

Provider participation in EBP is mandatory. For each episode, a principal accountable provider (PAP), who is generally the physician that performs the procedure, is identified through claims data and held responsible for total episode spending. Providers are paid fee-for-service during the course of the year. Then, at the end of a performance period (usually 1 year), the payer calculates a PAP’s risk-adjusted16 average spending across all valid episodes and compares it to pre-determined thresholds, which are based on the payer’s historical spending in Arkansas. These thresholds establish “commendable,” “acceptable,” and “unacceptable” spending levels. PAPs with “commendable” spending that also meet specified quality metrics receive 50% of the savings, up to a gainsharing limit. PAPs with “unacceptable” spending must pay back 50% of excess costs above the “acceptable” threshold. Further details are available in the Online Appendix.

Our analysis focuses on four episode types: total joint replacement, cholecystectomy, colonoscopy, and tonsillectomy. We chose these particular episodes because they are primarily elective in nature, and thus there is greater opportunity for volume expansion.8, 17 Also, compared with other episodes covered under Arkansas’ EBP reform, they occur more frequently and therefore provide larger (though still limited) sample sizes for the analysis.18 There are slight differences across episodes in terms of when commercial EBP was implemented and whether both AR BCBS (79% of the large group market) and QC (6% of the large group market) participate in the model.19

METHODS

Data

Our primary data source is the 2011–2016 Truven Health MarketScan Commercial Claims and Encounters database—a convenience sample of enrollees in commercial health insurance plans offered by large employers across the USA. We restrict our sample to beneficiaries who are continuously enrolled for the entire year20 and in health maintenance organization, point-of-service, preferred provider organization, consumer-driven health, and high-deductible health plans, because these plan types have adequate sample size to allow us to control for plan type.21

Identifying Episode Triggers

An episode is triggered by a specific procedure code, and we exclude repeated episodes for the same patient (see Online Appendix). Each payer also publishes criteria regarding which episodes would be excluded from EBP (e.g., episodes for patients with certain high-cost comorbidities). The potential for providers to “game” the exclusions (e.g., by coding intensive patients in a manner such that their episodes are excluded from gainsharing calculations) raises concerns about examining only “valid” episodes. To eliminate this concern, we do not apply the exclusion criteria.

Analyses

We analyze each episode type separately because responses may vary by clinical area, participating payers differ, and the episode types had slightly different start dates (which creates different pre- and post-EBP periods across the conditions). For total joint replacement, the pre-period is 2011–2012 and post-period is 2013–2016. For cholecystectomy, colonoscopy, and tonsillectomy, the pre-period is 2011–2013 and post-period is 2014–2016.

Our control group consists of commercially insured beneficiaries drawn from states that geographically border Arkansas or are also located in the South Central Census Divisions.22 We omit Tennessee as a potential control state, because it launched its own EBP reforms in 2013. We also exclude Kentucky and Oklahoma from the analysis, because of sharp changes in their MarketScan samples over time. Therefore, our baseline set of control states includes Missouri, Alabama, Mississippi, Louisiana, and Texas.23 For each episode type, we test for differential pre-EBP trends in the outcome variables between Arkansas and the control states (see Online Appendix). We only include states in the final control group if their pre-trends are not significantly different from Arkansas’ pre-trend. For all episodes except colonoscopy, all five baseline states pass the pre-trends test.

For each episode type, we plot the annual rate of episode-triggering procedures (number of procedures that occurred in a given year divided by the number of continuously enrolled MarketScan beneficiaries) for Arkansas and for the pooled control states. We utilize a difference-in-differences empirical strategy to formally measure the impact of EBP on the probability of a beneficiary having an episode. We describe our econometric model in the Online Appendix.

We perform a series of robustness checks. First, we determine if our results are sensitive to the choice of control states by repeating the analysis dropping each control state in sequence and pooling the remaining four states. Second, as placebo tests, we repeat the analysis on five common procedures that were not subject to EBP during the study period: lumbar spinal fusion, inguinal hernia repair, appendectomy, upper endoscopy, and cataract surgery. We test the hypothesis that EBP had no significant effect on the volume of these procedures (see Online Appendix). Third, we try adjusting our p values for multiple comparisons.24, 25

All statistical analyses were performed using Stata (version 15.1, StataCorp). We utilize two-tailed hypothesis tests with a statistical significance threshold of α = 0.05.

RESULTS

Study Population

Both before and after EBP implementation, beneficiaries in Arkansas and the control states are close in mean age, though there are slightly more females in the control states (Table 1). There are also some differences in plan type. Nonetheless, changes over time in beneficiary characteristics are similar between Arkansas and the control states.

Table 1 Summary Statistics for Commercially Insured Beneficiaries in Arkansas and Control States, Before and After EBP Implementation
Full size table

Total Joint Replacement

For our analysis of the total joint replacement episode, we restrict the sample to beneficiaries ages 40–64 (Fig. 1). The trends (and levels) for Arkansas and the control group are similar during the pre-EBP period (2011–2012). However, the trends diverge after full EBP implementation is achieved in 2014. Our difference-in-differences estimate of the EBP effect is a 9.9% increase (p = 0.08) in the probability of undergoing total joint replacement in a given quarter, relative to the Arkansas pre-period mean per 1000 beneficiaries of 0.91 (Table 2). While this estimated effect is large in magnitude, it is not statistically significant. When we sequentially drop control states, our estimate of the EBP effect ranges from a 7.4% increase (p = 0.19) when dropping Alabama to a 13.6% increase (p = 0.049) when dropping Texas (Table A-1).

Figure 1
figure 1

Rate of total joint replacement episode–triggering procedures. Blue diamonds indicate annual rates for Arkansas. Green squares indicate annual rates for all controls. The vertical orange line indicates partial commercial EBP implementation (AR BCBS only) in 2013. The vertical maroon line indicates full commercial EBP implementation (AR BCBS and QC) in 2014. All controls refers to Missouri, Alabama, Mississippi, Louisiana, and Texas. Source: Authors’ analysis of Truven Health MarketScan Commercial Claims and Encounters data for 2011–2016.

Full size image
Table 2 Estimated Effect of Episode-Based Payment on Probability of Undergoing Procedure (Per 1000 Beneficiaries)
Full size table

We repeat the analysis for lumbar spinal fusion, a comparable orthopedic surgical procedure that was not reimbursed under EBP (Figure A-1). Our difference-in-differences estimate of the post-EBP effect suggests a 9.0% increase (p = 0.39) in the probability of undergoing lumbar spinal fusion during a given quarter (Table A-2). Although this estimate is positive and has sizeable magnitude, it is not statistically significant; it also reflects an unexplained drop in 2012 (before EBP was introduced), and the increase does not persist.

Cholecystectomy

For our analysis of the cholecystectomy episode, we restrict the sample to beneficiaries ages 18–64 (Fig. 2). Our difference-in-differences estimate of the EBP effect is a 3.6% decrease (p = 0.09) in the probability of undergoing cholecystectomy during a given quarter, relative to the Arkansas pre-period mean per 1000 beneficiaries of 1.7 (Table 2). We repeat the analysis using alternative control groups, and the estimated effect ranges from a 3.1% decrease (p = 0.14) when dropping Alabama to a 4.2% decrease (p = 0.06) when dropping Missouri (Table A-3).

Figure 2
figure 2

Rate of cholecystectomy episode–triggering procedures. Blue diamonds indicate annual rates for Arkansas. Green squares indicate annual rates for all controls. The vertical maroon line indicates commercial EBP implementation (AR BCBS and QC) in 2014. All controls refers to Missouri, Alabama, Mississippi, Louisiana, and Texas. Source: Authors’ analysis of Truven Health MarketScan Commercial Claims and Encounters data for 2011–2016.

Full size image

We repeat the analysis for inguinal hernia repair (Figure A-2) and appendectomy (Figure A-3), two surgical procedures that were not subject to EBP during the study period. Our difference-in-differences estimates of the post-EBP effect for inguinal hernia repair and appendectomy suggest a 3.1% (p = 0.69) and 4.9% decrease (p = 0.57) in probability, respectively (Table A-4, Table A-5).

Colonoscopy

For our analysis of the colonoscopy episode, we restrict the sample to beneficiaries ages 40–64 (Fig. 3).26 We find that only Alabama and Louisiana have pre-trends that are statistically similar to Arkansas, so the control group includes these two states only. Our difference-in-differences results indicate that after EBP was implemented, the probability of undergoing colonoscopy during a given quarter increased by 17.2% (p < 0.001), relative to the Arkansas pre-period mean per 1000 beneficiaries of 15 (Table 2). If we include all five baseline states in the control group, we obtain similar results: an estimated increase in probability of 15.6% (p < 0.001) (Table A-6).

Figure 3
figure 3

Rate of colonoscopy episode–triggering procedures. Blue diamonds indicate annual rates for Arkansas. Green squares indicate annual rates for all controls. Red triangles indicate annual rates for Alabama and Louisiana. The vertical maroon line indicates commercial EBP implementation (AR BCBS) in 2014. All controls refers to Missouri, Alabama, Mississippi, Louisiana, and Texas. While we show the trend for all control states pooled, only Alabama and Louisiana pass the pre-trends test and thereby form the control group for our primary specification. Source: Authors’ analysis of Truven Health MarketScan Commercial Claims and Encounters data for 2011–2016.

Full size image

As a placebo test, we examine how volume changed for upper endoscopy, a related procedure that was not reimbursed under EBP (Figure A-4). We apply the same criteria as in our analysis of colonoscopy, so the control group contains Alabama and Louisiana only. During the post-EBP period, there is no evidence of a sustained increase in the rate of upper endoscopy in Arkansas. Our difference-in-differences results suggest a statistically non-significant 4.8% increase (p = 0.31) in the probability of undergoing upper endoscopy. Even when we include the full set of control states, the estimated effect remains statistically non-significant (Table A-7).

Tonsillectomy

For our analysis of the tonsillectomy episode, we restrict the sample to beneficiaries ages 3–21 (Fig. 4). Our difference-in-differences estimate of the EBP effect is a 2.9% decrease (p = 0.57) in the probability of undergoing tonsillectomy during a given quarter, relative to the Arkansas pre-period mean per 1000 beneficiaries of 1.9 (Table 2). Using alternative control groups, we find that the estimated effect ranges from a 0.7% decrease (p = 0.91) when dropping Texas to a 4.4% decrease (p = 0.38) when dropping Missouri (Table A-8).

Figure 4
figure 4

Rate of tonsillectomy episode–triggering procedures. Blue diamonds indicate annual rates for Arkansas. Green squares indicate annual rates for all controls. The vertical maroon line indicates commercial EBP implementation (AR BCBS) in 2014. All controls refers to Missouri, Alabama, Mississippi, Louisiana, and Texas. Source: Authors’ analysis of Truven Health MarketScan Commercial Claims and Encounters data for 2011–2016.

Full size image

In a placebo test, we analyze volume changes for cataract surgery, a common procedure that was not reimbursed under EBP (Figure A-5). Our difference-in-differences estimate of the post-EBP effect is a 1.2% decrease (p = 0.92) in the probability of undergoing cataract surgery during a given quarter (Table A-9).

DISCUSSION

Similar to the literature on spending per episode, we find that the relationship between EBP and procedure volume varies by procedure. The evidence for colonoscopy suggests a large, significant increase in volume. The effect on total joint replacement volume is also large in magnitude, but it is not statistically significant, and there is a sizeable increase for our related placebo procedure (lumbar spinal fusion). Therefore, we do not conclude that EBP had an impact on total joint replacement volume. There is no discernable impact on cholecystectomy or tonsillectomy volume. Thus, for certain procedures, the potential for volume expansion is meaningful. For other procedures, however, the impact on volume seems minimal.

Various clinical, organizational, and financial factors could be at play that facilitate volume expansion for certain procedures over others. For example, there is variation across procedures in providers’ ability to generate efficiencies, which impacts potential profit. Furthermore, there is variation across procedures in how much clinical flexibility providers have in recommending the procedure to patients, and in how many eligible patients are at the margin of receiving care. Among our selected episodes, the clinical indications for tonsillectomy and cholecystectomy are narrower. There is also limited room to increase efficiency during tonsillectomy and laparoscopic cholecystectomy procedures. These factors may contribute to our finding of no volume changes following EBP implementation. In contrast, there is robust evidence demonstrating that colorectal cancer screening, which is recommended for adults ages 50 to 75, is substantially underutilized among the US population.27 This suggests that there is great potential for appropriate expansion. Accordingly, we find a large, significant increase in colonoscopy volume following EBP implementation.

Ultimately, the perceived consequences of volume expansion will differ depending on the nature of the procedure. While it would be an unfavorable result if volume increases for procedures that are not clinically appropriate, it would be favorable if volume increases for procedures that are clinically indicated. The increase in colonoscopy volume that we observe can potentially be viewed as a desirable shift of resources and may be beneficial to the population, assuming it occurs among clinically indicated patients. Although our data limits our ability to determine whether this increase is clinically indicated, roughly 71% of colonoscopies in our post-EBP Arkansas sample were performed on patients ages 50 and older, versus 67% prior to EBP.

Our study has several limitations. First, we cannot measure underlying need for the procedures at the population level or among those who received treatment. However, this would only bias our results if population-level need, controlling for age and gender, changed differentially across states, which we believe is unlikely. Also, because the only procedure for which we find clear evidence of volume expansion is colonoscopy (which is commonly recommended to older adults), we do not believe that there were significant increases in unindicated care.28 Second, we are unable to observe provider IDs or characteristics in MarketScan. While this limits our ability to identify differential effects across providers or provider types, unobserved provider characteristics would only bias our results if there were differential changes across states. We performed extensive testing to ensure that our control group provides a valid counterfactual, which mitigates this concern. Third, for each episode type, Arkansas Medicaid implemented EBP a few months before commercial payers did.29 To the extent that spillovers exist, our findings would underestimate volume changes because the effect would have begun before our analysis assumes. Additionally, some smaller commercial insurers in Arkansas do not participate in EBP. We are unable to identify insurers in MarketScan, so we may be capturing procedures for patients who are covered by non-participating insurers, which could dilute our results. However, non-participating insurers only comprise about 15% of the large group market,19 so the effects due to non-participating insurers are likely small and would bias us away from finding an effect. An additional limitation is that during the post-intervention period, providers in the control states could have been participating in Medicare EBP demonstrations for joint replacement (the only episode type that overlaps with our analysis). However, only a few providers participated in these demonstrations, and any changes to practice style would only impact the treatment of under-65, commercially insured beneficiaries through spillover effects. Lastly, Arkansas’ payment reforms also include a patient-centered medical home (PCMH) program, which was launched in 2014. To the extent that the PCMH program affects procedure volume, our estimates may be biased. In our placebo tests, we find that the volume of five non-EBP procedures did not change significantly between 2011–2013 and 2014–2016. Still, potential confounding could be of concern for colonoscopy, if the PCMH program (which promotes preventive care) differentially affects volume for colonoscopy versus upper endoscopy (our related placebo procedure).

Despite these limitations, our study provides the first evidence that, in a commercial population, the impact of EBP on procedure volume varies across procedures. Overall, our findings do not indicate significant volume expansion beyond that for colonoscopy, which is generally underutilized. Nevertheless, payers must be vigilant of the potential for volume expansion. Continued refinements to episode payment schemes and constant monitoring will be important for organizations hoping to transition to alternative payment models.