Can vertically integrated health systems provide greater value: The case of hospitals under the comprehensive care for joint replacement model?

Abstract

Objective

We aim to assess whether system providers perform better than nonsystem providers under an alternative payment model that incentivizes high‐quality, cost‐efficient care. We posit that the payment environment and the incentives it provides can affect the relative performance of vertically integrated health systems. To examine this potential influence, we compare system and nonsystem hospitals participating in Medicare's Comprehensive Care for Joint Replacement (CJR) model.

Data Sources

We used hospital cost and quality data from the Centers for Medicare & Medicaid Services linked to data from the Agency for Healthcare Research and Quality's Compendium of US Health Systems and hospital characteristics from secondary sources. The data include 706 hospitals in 67 metropolitan areas.

Study Design

We estimated regressions that compared system and nonsystem hospitals' 2017 cost and quality performance providing lower joint replacements among hospitals required to participate in CJR.

Principal Findings

Among CJR hospitals, system hospitals that provided comprehensive services in their local market had 5.8 percent ($1612) lower episode costs (P = .01) than nonsystem hospitals. System hospitals that did not provide such services had 3.5 percent ($967) lower episode costs (P = .14). Quality differences between system hospitals and nonsystem hospitals were mostly small and statistically insignificant.

Conclusions

When operating under alternative payment model incentives, vertical integration may enable hospitals to lower costs with similar quality scores.

What This Study Adds.

• To date, the empirical literature suggests that care provided by health systems is more expensive than care provided outside of such systems, without commensurate improvements in quality. However, these studies generally estimate associations across broad populations and consequently reflect environments where fee for service remains the predominant reimbursement method and system and nonsystem providers alike may lack sufficient motivation to improve quality or reduce costs. The objective of this paper was to assess whether system providers perform better than nonsystem providers under a mandatory episode payment model that incentivizes high‐quality and cost‐efficient care.

• This study's findings suggest that vertically integrated system hospitals, particularly those that provide comprehensive services in their local market, were more successful at reducing episode costs than nonsystem hospitals or system‐affiliated hospitals that are not substantially vertically integrated within their local market. This suggests that systems may be capable of providing greater value when payment models incentivize them to do so.

1. INTRODUCTION

Vertically integrated health systems have grown rapidly in recent years. ¹ This study examines whether—among hospitals required by Medicare to participate in the Comprehensive Care for Joint Replacement (CJR) bundled payment model—being part of a vertically integrated health system was associated with greater success in lowering costs and improving quality of lower extremity (hip or knee) joint replacements (LEJRs).

One of the promises of vertically integrated health systems—used to counteract concerns about the potential anticompetitive effects from greater consolidation—has been their greater ability to succeed under alternative payment models designed to reward value. Advocates argue that common ownership or joint management of hospitals and physician practices enables shared infrastructure, economies of scale, and alignment of economic incentives that can facilitate clinical integration and adherence to evidence‐based practices. ² , ³ , ⁴ Hence, vertical integration could mediate the effect of value‐based incentives on provider performance.

However, evidence suggests few economies of scale in physician care. Instead, systems might be using their greater market power to bargain for higher reimbursements, making the care provided by health systems more expensive (without commensurate improvements in quality), or to avoid or preempt pressures to enter into alternative payment arrangements. ⁵ , ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹²

These findings on the cost of care delivered by systems might reflect the fact that fee‐for‐service (FFS) reimbursement remains predominant, and consequently, health systems may be able to exercise their greater pricing power without sufficiently strong incentives from alternative payment models to lower costs and improve quality. Therefore, existing research likely reflects health system performance under existing incentives rather than the capacity or potential for health systems to improve quality or reduce costs under value‐oriented incentives.

1.1. The comprehensive care for joint replacement model

CJR makes hospitals accountable for the cost and quality of LEJR episodes. Episodes span the inpatient admission (MS‐DRGs 469 or 470) and extend 90 days postdischarge. Episode costs include all Medicare Part A and Part B (inpatient, outpatient, and postacute) expenditures, with some exclusions for services unrelated to the joint replacement. Providers are paid on a FFS basis with retroactive reconciliation with hospitals of completed episode costs at the end of each performance year (PY), based on a hospital's cost and quality performance. Hospitals receive shared savings if average episode costs fall below their target price, provided they achieve a minimum quality score. Otherwise, hospitals repay Medicare a share of costs in costs exceeding their target price. In the program's first PY, 2016, hospitals were not liable for repayments to CMS; this feature of the model began in PY2. ¹³

Evaluation research comparing the performance of CJR hospitals to a group of matched comparison hospitals suggests that the model was associated with lower costs in the first two years of the model. Specifically, that Medicare Part A costs decreased significantly by about three percent for CJR hospitals, driven largely by reductions in postacute care or the substitution of less expensive home care for institutional postacute care with no detectable changes in complications, mortality, volume of episodes, or patient characteristics. ¹⁴ , ¹⁵ , ¹⁶

CJR participation was mandatory during the first two years of the model for nearly all hospitals in 67 metropolitan statistical areas (MSAs), randomly selected from all MSAs with sufficient numbers of eligible LEJR episodes. This feature of the model enabled us to compare system and nonsystem performance without the risk of selection bias from hospitals choosing to participate in the model.

2. METHODS

We compare 2017 (PY2) episode cost and quality performance of CJR hospitals that are and are not part of vertically integrated health systems. We use PY2 data because CJR’s first performance year only lasted nine months, hospitals did not face downside financial risk until PY2, and PY2 provided hospitals greater time to adapt to CJR incentives. We do not evaluate CJR's success in improving performance because we lack a comparison group of hospitals not participating in CJR.

Because health systems include hospitals in markets where they employ or manage few or no physicians, we distinguish between locally and nonlocally integrated system hospitals, based on the hospitals’ ability to provide comprehensive services through system‐affiliated physicians located in the same market. To align our analytic approach with features of the CJR model, we defined markets as MSAs.

2.1. Data sources

As part of its Comparative Health System Performance Initiative (CHSP), the Agency for Healthcare Research and Quality (AHRQ) developed the 2016 Compendium of US Health Systems, a database of health systems nationally, as well as linkages to hospitals and physician practices (including physicians directly employed by hospitals) associated with systems. We used CJR hospital cost and quality data from CMS linked to the Compendium of US Health Systems and associated hospital and group practice linkage files. CMS provided program data for CJR hospitals that allowed us to calculate average hospital episode costs during the performance year and during the 3‐year historic period prior to CJR’s implementation (2012‐2014). Quality measures are publicly available on CMS’s Hospital Compare website. Other hospital characteristics came from the 2015 American Hospital Association Annual Survey, 2016 hospital cost reports (Healthcare Cost Report Information System), and the 2016 Provider of Services File.

As part of CHSP, AHRQ also convened a national set of experts to develop a consensus definition of a vertically integrated health system, which we used. Systems must include at least one hospital and a set of physicians providing comprehensive services defined as 50 or more total physicians including 10 or more primary care physicians. Our data allowed us to determine whether a hospital was owned or managed by a health system and to identify the physician groups owned or managed by the same systems in the same markets as CJR hospitals.

2.2. Outcome variables

We calculated the average cost of LEJR episodes during 2017 for each hospital. This measure truncated extremely high‐cost episodes and use “standardized costs” which eliminate geographic adjustments in Medicare payments. It is not equivalent to CJR target prices. We used two key quality measures that CJR uses in its composite quality score: (1) the total hip arthroplasty and/or total knee arthroplasty (THA/TKA) risk‐standardized complication measure (National Quality Forum [NQF] #1550) and (2) the Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) survey patient experience measure (NQF #0166). The former is calculated over a three‐year measurement period and is specific to Medicare FFS beneficiaries aged 65 and older. Quality measures based on small patient sample sizes are not reported in Hospital Compare, and we excluded hospitals with missing quality values from relevant analyses.

2.3. Key independent variables

Our key independent variables describe whether a hospital operates as one of two types of system hospitals. Health systems are often complex organizations that comprise multiple subsystems operating in different geographic areas under a parent system. Some systems include hospitals that employ or manage few or no physicians in their local market. Therefore, we distinguished between system hospitals providing comprehensive services (as specified in the CHSP consensus definition) in the same MSA from those system hospitals that do not. Figure 1 illustrates the distinction between locally integrated system hospitals, nonlocally integrated system hospitals, and nonsystem hospitals in two hypothetical markets (MSAs 1 and 2). The technical Appendix S1 details how health systems were defined and classified.

FIGURE 1.

Illustration of locally and nonlocally integrated system hospitals and nonsystem hospitals [Color figure can be viewed at wileyonlinelibrary.com]

Note: A comprehensive set of physicians is defined as 50 or more total physicians including 10 or more primary care physicians. In MSA 1, hospitals A and B are part of a vertically (and horizontally) integrated system in the context of their local market and thus are locally integrated system hospitals. Hospital C in MSA 1 is associated with a parent system that also owns hospital D in MSA 2. Hospital D has a comprehensive set of physicians within its market and thus is classified as a locally integrated system hospital, whereas hospital C does not have a comprehensive set of physicians in its market and thus is classified as a nonlocally integrated system hospital. Finally, Hospital E in MSA 2 is a nonsystem hospital as it fails to meet system criteria within its local market and is not part of a parent system

2.4. Other covariates

Hospitals with high LEJR episode costs before CJR might find it easier to trim excess costs and achieve cost savings than historically more efficient hospitals. ¹⁷ Moreover, pre‐CJR cost levels might be associated with system affiliation, either as a reflection of how systems influence cost performance or of past decisions to join or form health systems. Therefore, in cost equations, we controlled for hospitals’ average historical costs during 2012‐2014, adjusted to eliminate Medicare geographic payment adjustments. These historical costs were trended forward to account for changes in Medicare payment to 2017 and adjusted to match the mix of LEJR episodes during the PY. We also include an indicator variable for low‐volume hospitals, those with fewer than 20 episodes from 2012 through 2014.

We accounted for case‐mix differences across hospitals in a manner approximating CJR’s methodology. We controlled for the proportion of each hospital's LEJR episodes with or without complicating conditions (MS‐DRGs 469 and 470, respectively) and with or without a hip fracture. We lacked patient‐specific data that would allow us to control for other patient characteristics potentially affecting cost and quality. ¹⁸ As described below, evidence suggests this limitation is unlikely to markedly bias our results.

We included other hospital controls including bed size, profit status, the number of LEJR episodes, and whether the hospital is an orthopedic hospital, is integrated with postacute providers, has a major teaching function, and serves a large share of uninsured/low‐income patients. The Appendix S1 provides a complete list of our independent variables and their data sources. Because local market characteristics are complex and often unavailable, we controlled for market differences by including MSA fixed effects.

2.5. Estimation

We regressed cost and quality outcomes on categorical variables indicating the CJR hospital was part of a locally integrated system hospital or part of a nonlocally integrated system (using nonsystem hospitals as the reference group), the set of hospital covariates, and market fixed effects. OLS regressions used robust standard errors to account for heteroscedasticity. A semi‐log specification enabled us to express results in terms of a percentage change and to account for moderate skewness of average episode costs. The technical Appendix S1 contains the results of several sensitivity tests, which generally found results to be robust.

3. RESULTS

3.1. Sample characteristics

Table 1 contains descriptive information for the 706 CJR hospitals with LEJR episodes in PY2. The vast majority (87 percent) were part of health systems, among which three in four were part of locally integrated system hospitals. Locally integrated system hospitals had the lowest average episode cost ($26 801), while nonlocally integrated system hospitals had the highest ($29 815). Differences in quality measures were small across hospital groups. System hospitals were larger than nonsystem hospitals and less likely to serve large shares of low‐income patients. Among system hospitals, locally integrated system hospitals were more likely to be teaching facilities and less likely to be for‐profit than both nonsystem and nonlocally integrated system hospitals.

TABLE 1.

Characteristics of system and nonsystem CJR hospitals

	System hospitals		Nonsystem hospitals
	Locally integrated	Not locally integrated
Outcomes—PY2
Per‐episode cost, mean	$26 800.8	$29 814.9	$28 854.3
HCAHPS patient experience survey measure, mean	85.5	84.2	85.5
THA/TKA complication measure, mean	2.6	2.6	2.5
Hospital characteristics
Hospital historical costs, mean	$29 762.6	$32 342.6	$30 440.0
Proportion of episodes
470 no fracture	76.4%	71.9%	78.2%
470 fracture	17.2%	18.9%	17.1%
469 no fracture	2.2%	2.5%	1.5%
469 fracture	4.2%	6.6%	3.2%
Number of PY2 CJR episodes, mean	150.6	126.3	123.2
Low‐volume hospital	5.9%	4.4%	12.8%
Number of beds, mean	296.6	215.9	131.7
Major teaching hospital	22.1%	8.0%	11.7%
Total margins	5.1%	6.3%	5.9%
For‐profit ownership	12.4%	51.1%	43.6%
Orthopedic hospital	0.6%	0.7%	8.5%
High DSH patient percentage hospital	22.7%	24.1%	37.2%
Owns or manages skilled nursing facility	8.2%	5.1%	5.3%
Provides outpatient rehabilitation services	61.7%	58.4%	46.8%
Owns or manages home health agency	9.5%	3.6%	7.4%
Number	475	137	94

The hospital characteristics provided in this table comprise the list of control variables included in our main regressions.

Abbreviations: CJR, Comprehensive Care for Joint Replacement; DSH, disproportionate share hospital; HCAHPS, Hospital Consumer Assessment of Healthcare Providers and Systems; MSA, metropolitan statistical area; PY, performance year; THA/TKA, total hip arthroplasty/total knee arthroplasty.

3.2. Regression results

Compared to nonsystem hospitals, locally integrated system hospitals had 5.8 percent ($1612) lower per‐episode costs (P = .01) in PY2 of the CJR model (Table 2). Nonlocally integrated system hospitals were associated with 3.5 percent ($967) lower per‐episode costs, though that result was not statistically significant at the 0.05 level. Differences in locally and nonlocally integrated system hospital coefficients also did not reach conventional levels of statistical significance (P = .11). There were no statistically significant differences between locally integrated system hospitals and nonsystem hospitals for the two quality measures (or between locally and nonlocally integrated health systems). We found nonlocally integrated system hospitals had slightly lower, statistically significant, HCAHPS patient experience scores relative to nonsystem hospitals (−1.4 percent, P = .01). Locally integrated system hospitals had slightly higher HCAHPS patient experience scores relative to nonlocally integrated system hospitals (1.4 percent, P < .001).

TABLE 2.

Estimated relationships between hospital system affiliation and lower joint replacement episode costs and quality

Outcome	N	Locally integrated system hospitals		Nonlocally integrated system hospitals
		Percentage change	P‐value	Percentage change	P‐value
Average episode cost	706	−5.8	0.01	−3.5	0.14
HCAHPS survey measure	692	0.0	0.96	−1.4	0.01
THA/TKA complication measure	573	−0.3	0.91	3.1	0.36

Values in the table represent the percentage change in the outcome for locally integrated system hospitals and nonlocally integrated system hospitals compared to nonsystem hospitals. The Appendix S1 provides full regression results, including coefficients for all hospital covariates.

Abbreviations: CI, confidence interval; HCAHPS, Hospital Consumer Assessment of Healthcare Providers and Systems; N, number; THA/TKA, total hip arthroplasty/total knee arthroplasty.

4. DISCUSSION

Our findings are consistent with the hypothesis that vertically integrated health system has the capacity to provide greater value in health care, provided payment incentives encourage this outcome. System affiliation was associated with lower CJR episode costs. Although not quite statistically significant at conventional levels, locally integrated system hospitals appear more successful at reducing costs than nonlocally integrated system hospitals, suggesting benefits from vertical integration with physicians in a local market.

These findings differ from past studies that associated health systems with higher costs. ⁵ , ⁶ , ⁷ , ¹⁰ , ¹¹ In part, this is likely explained by our focus on Medicare beneficiaries for whom prices are set administratively, not negotiated, and by our use of a representative sample of hospitals, all of which faced equivalent value‐based payment incentives. Other studies’ results reflect system versus nonsystem performance in predominantly FFS environments that fail to consistently incentivize lower costs or improved quality.

This study has several limitations. First, its cross‐sectional, observational design and lack of random selection of hospitals into systems preclude causal inferences. Systems may preferentially integrate with more efficient hospitals or physicians. Second, CJR quality measures are missing for some hospitals and might not adequately incentivize hospitals to improve quality or be sensitive enough to detect impacts. The patient experience measure is not specific to LEJR patients, and the complication measure is a three‐year average. Third, we lacked patient‐level data such that unmeasured patient case mix could confound results. However, this may not be a serious concern given that results were robust to the inclusion of hospital characteristics associated with patient‐level case mix (eg, teaching hospital and low‐income patient population indicators). Finally, the study cannot address the appropriateness of LEJR procedures, patient selection, or other possible unintended consequences of the CJR model. ¹⁹

Relative performance of system versus nonsystem hospitals under bundled payments for LEJR procedures might not generalize to performance under other APMs. We also cannot generalize beyond the provision of LEJRs. Cost savings under CJR and similar bundled payment models have come largely from more efficiently managed postacute care without jeopardizing revenue from inpatient care. ¹⁵ , ¹⁶ Findings could differ under an APM that provides stronger incentives to reduce the number of hospitalizations. Indeed, research shows greater success of ACOs run by physician groups as compared to those run by hospitals. ²⁰

There are several mechanisms by which policy makers might leverage the potential benefits of hospital and physician integration while avoiding associated pitfalls. For example, making it more difficult for systems that acquire physician offices to generate greater revenue by billing office‐based services using a hospital outpatient department site of service would likely reduce costs associated with health systems. CMS has taken steps to do this in the 2020 physician fee schedule rule changes. Also, policies should encourage value‐based payment model participation. Beyond current Medicare Quality Payment Program initiatives, these might include broader adoption of mandatory Medicare APMs and of federal and state initiatives to encourage multipayer collaboration in payment models.

Supporting information

Supplementary Material

Appendix S1

Machta RM, Reschovsky J, Jones DJ, Furukawa MF, Rich EC. Can vertically integrated health systems provide greater value: The case of hospitals under the comprehensive care for joint replacement model?. Health Serv Res. 2020;55:541–547. 10.1111/1475-6773.13313