THE IMPACT OF GLOBAL BUDGET PAYMENT REFORM ON SYSTEMIC OVERUSE IN MARYLAND

Abstract

Background:

Medical overuse is a leading contributor to the high cost of the US health care system and is a definitive misuse of resources. Elimination of overuse could improve health care efficiency. In 2014, the State of Maryland placed the majority of its hospitals under an all-payer, annual, global budget for inpatient and outpatient hospital services. This program aims to control hospital use and spending.

Objective:

To assess whether the Maryland global budget program was associated with a reduction in the broad overuse of health care services.

Methods:

We conducted a retrospective analysis of deidentified claims for 18 to 64 year old adults from the IBM MarketScan^® Commercial Claims and Encounters Database. We matched 2 Maryland Metropolitan Statistical Areas (MSAs) to 6 out-of-state comparison MSAs. In a difference-in-differences analysis, we compared changes in systemic overuse in Maryland vs the comparison MSAs before (2011–2013) and after implementation (2014–2015) of the global budget program. Systemic overuse was measured using a semiannual Johns Hopkins Overuse Index.

Results:

Global budgets were not associated with a reduction in systemic overuse. Over the first 1.5 years of the program, we estimated a nonsignificant differential change of −0.002 points (95%CI, −0.372 to 0.369; p=0.993) relative to the comparison group. This result was robust to multiple model assumptions and sensitivity analyses.

Conclusions:

We did not find evidence that Maryland hospitals met their revenue targets by reducing systemic overuse. Global budgets alone may be too blunt of an instrument to selectively reduce low-value care.

INTRODUCTION

Health expenditures in the United States are high, with little evidence that this spending leads to better health outcomes [1–5]. Overuse—the provision of care where the potential for harm exceeds the potential for benefit—has been cited as a leading contributor to the high cost of the US health care system and is a definitive misuse of resources [6–8]. The elimination of these services could improve health care efficiency. Over the last decade, diverse stakeholders have adopted alternative payment models in an attempt to control the growth of health care spending and improve health care outcomes but relatively little is known about how these policies impact overuse [9–14].

Preliminary evidence from the Medicare Pioneer Accountable Care Organization (ACO) program and the Massachusetts Alternative Quality Contract (AQC) suggests that global payment initiatives might impact rates of overutilization [15,16]. These models incentivize the delivery of effective and efficient care. Specifically, global budgets provide a fixed amount of funding for a fixed period of time for a specified population, rather than fixed rates for individual cases or services [17]. This budget constraint discourages the provision of services that contribute to spending but not to health. Global budgets are often paired with specific performance incentives and assessments to ensure that providers do not compromise quality or access in their efforts to adhere to their budget.

In 2014, Maryland established its global budget program via a waiver from the Center for Medicare and Medicaid Services (CMS). This program placed the majority of hospitals within the state under an all-payer, annual, global budget for inpatient and outpatient hospital services [18]. In the first three years of the model, all-payer hospital revenue growth stayed below the state-set cap of 3.58 percent, saving Medicare $586 million [19]. To some extent, hospital savings are effectively guaranteed by the design of the program, however, the impact of the model on health care utilization is less clear. Within the Medicare population, different modeling assumptions have led to different conclusions [20–22]. Only one analysis has focused on Maryland residents with commercial plans; it found a significant reduction in hospital admissions, ED visits, and observation stays [21]. It is unknown whether Maryland global budgets have resulted in a reduction in the use of overused services.

We combined 19 claims-based indicators of low-value care to construct an index of regional systemic overuse in commercially insured adults. Using this measure, we conducted a difference-in-differences analysis that compared changes in systemic overuse in Maryland with changes among similar out-of-state regions, before and after the start of Maryland’s global budget program.

METHODS

Maryland’s Hospital Global Budget Program

On January 1, 2014, under a CMS waiver, Maryland initiated an all-payer global budget program for the 36 nonfederal acute-care hospitals in the state [23]. As part of the agreement, Maryland pledged to achieve substantial cost savings (i.e., limit annual growth of per capital hospital costs for all payers to 3.58%; generate $330 million in savings to Medicare by 2019) and quality improvements (i.e., reduce its 30-day readmission rate to the unadjusted national Medicare average over 5 years) [21,24]. The Maryland Health Services Cost Review Commission (HSCRC) is responsible for establishing the annual global budget, or allowed revenues, for each hospital. This budget is set based on historical utilization, regulated prices, operating costs, and market share.

By July 1, 2014, 36 general acute-care hospitals were operating under an all-payer global budget for inpatient, emergency department, and hospital outpatient services [24]. The negotiated, finalized budgets were retroactively applied to the first of the year. Hospital revenues were expected to conform closely to the global budget, and penalties were applied to the portion of overages and underages that exceeded 0.5% of the hospital budget. To meet this constraint, hospitals were permitted to iteratively adjust their prices within a corridor of ±5% throughout the year; hospitals with excess utilization must lower their prices, while hospitals with less utilization could increase their prices [25]. This provides an opportunity for hospitals to maintain (or increase) their profit margin via a reduction in the delivery of services that do not improve patient outcomes.

Data

We used the IBM MarketScan^® Research Databases to generate a measure of systemic overuse. The Commercial Claims and Encounters Database includes inpatient, outpatient, and prescription drug claims that are submitted by private health plans and employers. These are claims from more than 47 million US residents aged 0 to 64 years old with employer-sponsored private health insurance from all 50 states and the District of Columbia. The databases capture medical claims and spending for employees and their insured dependents. Data from the Area Health Resource File (AHRF), American Community Survey (ACS), and American Hospital Association (AHA) Annual Survey were used for the matching process.

Study Sample

In a serial cross-sectional design, we used the inpatient and outpatient claims in semiannual, 6-month intervals from January 2011 to June 2015. Beneficiaries were included if they were 18 to 64 years old and had continuous health insurance coverage during the specified time period. Individuals were assigned to a Metropolitan Statistical Area (MSA) based on the geographic location of their residence.

Our analyses focused on the 2 largest Maryland Metropolitan Statistical Areas (MSAs): “Baltimore-Columbia-Towson” and “Silver Spring-Frederick-Rockville.” Based on preliminary analyses of the ACS and the AHA Annual Survey, these two regions are home to close to 70% of the entire Maryland population and include 27 of the 36 acute-care hospitals affected by the global budget program (Figure 1; Appendix Table A1). These 27 hospitals constitute 87% of the Maryland hospital beds that were exposed to the global budget program. Three Maryland metropolitan areas were excluded because they contain counties that are in other states. For example, the “Washington-Arlington-Alexandria” MSA includes regions in the District of Columbia, Virginia, Maryland, and West Virginia. The “California-Lexington Park, MD” metropolitan area, which serves predominantly rural and seasonal communities in Western Maryland and the Eastern Shore, was excluded because of limited data availability.

Figure 1. Metropolitan Statistical Areas and Hospitals in Maryland.

This map depicts the hospitals that entered Maryland’s global budget program in 2014 and the Maryland MSAs that are included in our analyses.

Systemic Overuse

We calculated the Johns Hopkins Overuse Index (JHOI) for each MSA (N=375) in the U.S. in each of the 9 time periods. As detailed in prior studies, the index uses the occurrence of 19 clinically diverse services to measure the latent tendency of a region to overuse health care resources (Appendix Table A2) [26–30]. Consistent with the definition of overuse, a higher JHOI is associated with increased regional per capita spending without clinical benefit [28,29]. These 19 services were chosen because they can be operationalized using commercial claims data and each one is overused with sufficient frequency and variance to contribute meaningfully to an index. Most relevant to this work, each low-value service has the potential to be delivered in regulated, hospital inpatient and outpatient settings. Each overuse event is identified using patient demographic information, clinical characteristics, and a combination of International Classification of Diseases, Ninth Revision (ICD-9) and Current Procedural Terminology (CPT) codes. The index is generated using multilevel linear regression to model the probability that an eligible beneficiary experienced an overuse event as a function of beneficiary age and sex, a fixed effect for each of the 19 procedures, and a MSA fixed effect. The coefficient associated with the MSA fixed effect describes the latent tendency of a MSA to overuse diverse health care resources. This estimate is then standardized to create the JHOI, which has a national average of 0 and a standard deviation of 1.

Matching

To form a comparison group for the difference-in-differences analyses, we used coarsened exact matching (CEM) to match each of the 2 Maryland MSAs to out-of-state control MSAs that were not involved in a global budgeting program. CEM is a monotonic imbalance reducing matching method—it finds exact matches on temporarily “coarsened” variables [31–33]. We matched on health system characteristics and demographic characteristics that are predictive of hospital utilization: number of hospital beds, physicians per capita, poverty rates, median income, proportion of white civilians, and proportion of uninsured residents [20,24]. We coarsened the variables to reflect the distribution of each variable across all MSAs. We required an exact match on all coarsened variables and checked that this procedure provided improved balance on several other characteristics that were not used for matching.

Statistical Analysis

In accordance with a difference-in-differences design, we quantified changes in the semiannual JHOI in Maryland that differed from concurrent changes in the comparison group from the pre-policy period (2011–2013) to the post-policy period (2014–2015). This analysis was conducted on an intention-to-treat basis, considering all beneficiaries in the 2 intervention MSAs to be exposed to the program regardless of where they received care. We fit the following linear regression model:

$$Y_{m,t}=\alpha +{\delta }_{t}Marylan{d}_{m,t}\times Pos{t}_t+\beta X_{m,t}+{\mu }_{m}MS{A}_m+{\gamma }_{t}Tim{e}_t+{\epsilon }_{m,t}$$

where m indexes metropolitan areas and t semiannual time periods. The variable X is a matrix of MSA, time-specific covariates, μ_m is a vector of MSA fixed effects, and γ_t are time fixed effects. The variable Maryland equals 1 if a given MSA is in Maryland in time period t, and set to zero if otherwise. The variable Post is set to 1 beginning in 2014. Thus, the regression coefficient, δ_t, represents the adjusted annual differential change in systemic overuse from the full pre-policy period to the full post-policy period, between comparable MSAs inside and outside of Maryland. Models were estimated with robust standard errors clustered at the MSA.

As sensitivity analyses, we first assessed changes in systemic overuse from the pooled baseline period to each of the 3 semiannual, post-intervention periods separately. We also conducted the analysis without the 6-month period immediately preceding or following the introduction of global budgets to account for possible anticipatory changes or a delayed effect of the policy. Third, to assess whether our estimates were sensitive to the specification of the comparison group or to policy changes that may have coincided with Maryland’s reform, we used two alternate comparison groups and removed MSAs that were differentially exposed to Medicaid expansion [34]. Fourth, in an attempt to account for heterogeneity between regions, we conducted a separate analysis for each Maryland MSA. Fifth, to examine the effect of uncertainty in the JHOI estimate, we defined the JHOI in the pre-policy period as the upper (lower) bound of each MSA’s fixed effect coefficient’s 95% confidence interval for the treated (control) group, and as the lower (upper) bound of the confidence interval for the treated (control) group in the post-policy period. Finally, we used two alternative measures of systemic overuse: the proportion of beneficiaries that received at least one overused service and the number of overused services per 100 beneficiaries.

A standard assumption of this analysis is that the difference in rates of systemic overuse between the Maryland MSAs and the comparison MSAs in the pre-intervention period would have remained constant in the post-intervention period in the absence of the Maryland global budget program. We tested this assumption with inclusion of an interaction term between the Maryland indicator and a linear time trend in the above model [20,21,24,25]. This approach estimates changes in systemic overuse in Maryland versus the comparison group, less any changes that would be expected from a continuation of the pre-intervention trends. The parallel trend assumption is satisfied if the estimates from these models remain consistent.

Analyses were performed using SAS version 9.4 and Stata 14 software [35,36]. Our study was determined to be exempt from review by REDACTED Institutional Review Board (REDACTED).

RESULTS

The study sample included 1,525,310 beneficiary-level observations in the 2 Maryland MSAs. Compared to all other MSAs, the Maryland MSAs had higher rates of educational achievement among their residents, more favorable economic indicators, a smaller proportion of uninsured individuals, and more health care providers (Table 1). The CEM procedure yielded 6 MSA matches; two in California and one each in New Jersey, Pennsylvania, Virginia, and Washington (Appendix Table A3). The comparison group included 10,350,653 beneficiary-level observations. Demographic and health system characteristics were more similar between Maryland and the matched comparison group, and differential changes over time were nonsignificant (Table 1).

Table 1.

Area-Level Characteristics Before and After Start of the Global Budget Program

	2011–2013			2014–2015a			Differential Changeb
Characteristic	Maryland (N=2)	Control (N=6)	Unmatched (N=359)	Maryland (N=2)	Control (N=6)	Unmatched (N=359)	Maryland vs Control	P Value
Johns Hopkins Overuse Index	0.6	0.6	−0.0	0.6	0.7	−0.0	−0.1	.74
Age, mean	41.7	42.1	42.2	41.8	41.8	42.1	−0.4	.28
Proportion male	49.3	48.3	48.2	48.8	48.0	48.1	0.2	.75
Household income, medianc,d	$88,906	$75,210	$53,519	$86,229	$77,260	$52,662	4727.0	.42
Proportion unemployed	6.8	8.5	8.5	5.9	6.2	6.4	−1.3	.10
Proportion below FPLc,d	8.2	11.7	16.2	8.5	10.9	15.9	−1.1	.31
Education
Less than High School	8.1	9.6	11.3	8.3	9.2	11.0	−0.6	.35
With College Degree	46.7	42.5	28.5	48.1	44.2	29.3	0.3	.92
Proportion uninsuredc,d	13.6	16.4	20.2	10.0	11.4	15.1	−1.3	.33
Proportion whitec,d	61.5	62.1	79.3	60.1	60.6	78.8	−0.2	.94
Population densityd,e	961	1033	285	985	1057	290	1.0	.99
Total MD physiciansc,d,f	5.1	3.9	2.5	5.2	3.9	2.6	−0.1	.92
Primary care providersf	1.1	0.9	0.7	1.1	0.9	0.7	0.0	.91
Specialistsf	4.1	3.0	1.9	4.2	3.0	1.9	−0.1	.89
Hospital bedsc,d,f	2.4	2.7	3.3	2.4	2.6	3.2	−0.1	.81
Hospitalsd,g	10.4	12.5	20.8	10.2	12.2	20.5	−0.1	.97

Notes: The unit of analysis for this table is the MSA, which is not weighted for population size in the statistics presented in this table.

Abbreviations: FPL, Federal Poverty Line.

^aData through June 2015

^bAfter matching, there were no differential changes in observed regional characteristics from the preintervention to the postintervention period in Maryland vs. the comparison group.

^cVariable used in Coarsened Exact Matching

^dVariable used as covariate in regression models

^ePopulation density is calculated as (Persons/square miles)

^fSupply per 1,000 residents

^gHospitals per 1,000 residents * 1,000

During the pre-intervention period, the national JHOI had a mean of 0 and standard deviation of 1 with a minimum of −4.00 in Bismarck, ND and a maximum of 3.51 in Pittsburgh, PA. Before implementation of the global budget program, the average semiannual JHOI in Maryland (JHOI 0.62; SE 0.02) and the comparison group (JHOI 0.59; SE 0.09) did not differ from one another (p=0.42) (Figure 2; Appendix Figure A1). Both groups overused healthcare more than the national average, and pre-intervention trends were similar (p=0.39).

Figure 2. Unadjusted Trends in Systemic Overuse Among Beneficiaries in Maryland and Comparison Metropolitan Statistical Areas, 2011–2015.

This figure shows the unadjusted semiannual Johns Hopkins Overuse Index among commercially insured adults residing in the 2 Maryland MSAs that began global budgets in 2014 versus the matched comparison group of 6 non-Maryland MSAs. The error bars represent the standard error for the point estimate in the given time period and were calculated using a test of differences between groups.

The Maryland global budget program was not associated with a reduction in systemic overuse. Over the first year and a half of the program, there was a nonsignificant differential change in the JHOI [−0.002 points (95% CI, −0.372 to 0.369; p=0.993)] relative to the comparison group. We did not detect heterogeneity in the effect of the policy over different periods of time; in the first 6 months after implementation there was a nonsignificant differential decrease in the JHOI of −0.039 (95% CI, −0.437 to 0.359; p=0.822) and 12–18 months after implementation there was a nonsignificant differential increase in the JHOI of 0.138 (95% CI, −0.435 to 0.710; p=0.587). The estimate was also robust to alternative specifications of the comparison group, treatment group, and dependent variable (Figure 3) (Appendix Table A4). When we used the extremes of the confidence intervals surrounding the JHOI estimates for Maryland and the comparison regions, we estimated a nonsignificant differential change in the JHOI that is consistent with our main analysis [0.049 points (95% CI, −0.326 to 0.426; p=0.765]. Using alternative outcome measures, there was a nonsignificant differential reduction of −0.213 (95% CI, −0.819 to 0.394; p=0.435) in the percent of beneficiaries that experienced an overuse event and a nonsignificant differential reduction of −0.227 (95% CI, −0.857 to 0.402; p=0.421) overuse events per 100 beneficiaries. Estimates were not meaningfully affected by adjusting for pre-intervention trends (Appendix Table A5).

Figure 3. Adjusted Differential Changes in Systemic Overuse in Maryland vs Comparison Group, Main Effect and Sensitivity Analyses.

This figure presents adjusted differential changes in the semiannual JHOI for the 2 Maryland MSAs vs the comparison group from the pre-intervention period (2011–2013) to the post-intervention period (June 2015). We present the difference-in-differences coefficient for our main model specification as well as our sensitivity analyses. The square indicates the differential reduction estimated by the main model, circles by the alternate control groups, diamonds by different temporal specifications, and triangles by different regional specifications; limit lines, 95% CI. Details on the 3 alternate control groups can be found in Table A4. 2014a includes January through June of 2014 and 2013b includes July through December of 2013.

DISCUSSION

In this work, we examine changes in systemic overuse associated with the introduction of global budgets in Maryland hospitals. We conducted a difference-in-differences analysis that compared changes in an index of overused procedures in Maryland to a matched out-of-state control group. We did not find any evidence that Maryland hospitals met their revenue targets via a reduction in systemic overuse. This result is robust to multiple specifications of our model and outcome.

As the nation experiments with new payment strategies, an understanding of the mechanisms by which spending is reduced is as important as the knowledge that it can be reduced at all. To date, the design and evaluation of existing alternative payment models have overlooked changes in low-value service use as a mechanism by which to reduce spending and improve care quality. Reflective of this, neither the Pioneer ACO program, Massachusetts AQC, nor Maryland global budget program had financial or quality reporting requirements that were specific to low-value care. Despite this, the Pioneer ACO program and Massachusetts AQC achieved small, but noteworthy reductions in utilization or spending on low-value care [15,16]. A systematic review of interventions to reduce low-value care concluded that multicomponent interventions that incorporate clinical decision support and performance feedback are promising strategies to combat overuse but that further research is needed on the effectiveness of pay-for-performance, insurer restrictions, and risk-sharing contracts [37]. Because no single payment model or intervention is likely to suffice on its own, purposeful experimentation that incorporates overuse as a standard component of quality measurement is the only way to determine the extent to which different strategies can be combined to increase high-value care and decrease low-value care [38].

There are several possible explanations for why we did not find a relationship between global budgets and systemic overuse in Maryland. First, the global budget program did not directly incentivize reductions in overuse. Even though minimizing unnecessary utilization would control overall spending, hospitals were explicitly incentivized to meet other performance benchmarks, such as reductions in readmissions and potentially preventable complications. It is possible that hospitals did not have sufficient resources to meet the specified requirements and target overuse. Arguably, meeting these quality benchmarks should increase health care value, but these improvements, in isolation, are unlikely to reduce the use of wasteful and harmful services. Second, it is plausible that use of some individual overused services declined while others increased, leading to a null result. Third, Maryland hospitals were essentially mandated to participate in the global budget program. While other alternative payment models have been voluntary, or even specifically designed for high performing organizations, the Maryland model provides an unbiased glimpse in to the average effect of global budgets. Important qualitative work suggests that some Maryland hospitals had difficulty adjusting to the payment model and implementing new care management programs [25]. Finally, the Maryland model did not explicitly include or directly incentivize (e.g., reward or penalize) physicians. With little personal financial risk in the success of the program, physicians may not have done much to change their established patterns of care delivery.

As documented by other evaluations of the Maryland global budget program, identifying an appropriate comparison group for the state of Maryland is challenging because it has had different hospital regulatory and payment policies than the rest of the country for several decades [20,21,24,25]. We opted to match the selected Maryland MSAs with out-of-state comparison MSAs based on baseline demographic and health system characteristics using a CEM procedure. While the 6 matched comparison MSAs more closely resembled the Maryland MSAs than the unmatched sample, there were still a number of regional characteristics that remained unbalanced. Despite this, pre-intervention trends in systemic overuse did not differ between groups and sensitivity analyses that adjusted for assumed differences produced consistently null results.

This study is not without limitations. Our analyses only examined changes in systemic overuse in the privately insured population and relied on data from a large convenience sample of claims, which may be more representative of market dynamics in some MSAs than others. Under the CMS waiver, there are several requirements that are specific to Medicare beneficiaries. Even though overuse is known to exist across insurance types and populations, the impact of global budgets on overuse may not generalize across groups. Additionally, we only examined the first year and a half of the global budget program. The program could become more effective over time, however our analysis of each individual postintervention time period did not suggest that this is likely. Moreover, evaluations of other global budget models have detected changes in spending and utilization within the first year of implementation [15,16].

Our measure of systemic overuse is defined at a regional level and is constructed from a combination of inpatient and outpatient claims data. As a result, our analysis may mask meaningful heterogeneity between hospitals. In addition, the JHOI may include services that were delivered in unregulated, nonhospital spaces. This may have led us to underestimate the effect of hospital global budgeting on overuse, however existing evaluations suggest that hospitals did not systematically shift their patients to nonregulated or out-of-state providers in order to meet their budgets [21,24]. The regional dependent variable also limits our sample size. Finally, we considered all beneficiaries in the 2 intervention MSAs to be exposed to the program regardless of where they received the indexed care.

Multiple factors contribute to overuse and the slow pace of its elimination. Overuse is motivated by the health care system, the practice environment, the culture of professional medicine, the culture of health care consumption, and individual patient and clinician characteristics [39]. Examining the relative strength of these domains would inform the development of novel intervention strategies. Specifically, more research is needed to understand the complex array of behavioral and cultural factors that influence overuse [40, 41]. Future research should leverage primary data collection to examine the association between physician beliefs and attitudes, as well as overall practice culture, on low-value care. Looking forward, health system leaders, policy makers, payers, and consumer advocates will likely need to use multiple synergistic levers to reduce low-value care [37].

As lessons from global budget payment systems amass, an understanding of the mechanisms by which spending is reduced is imperative. Previous evaluations of the Maryland global budget program have raised questions about the extent to which the observed reductions in health care spending reflect meaningful changes in utilization. We found that the Maryland global budget program was not associated with a change in systemic overuse. We suggest that use of hospital global budgets, in isolation, may be too blunt of an instrument to selectively reduce low-value care. This study demonstrates the feasibility and importance of incorporating measures of overuse into the evaluation of alternative payment models.

HIGHLIGHTS.

• Diverse stakeholders have adopted alternative payment models to improve health care value but this work has overlooked low-value service use as a primary target. We examined whether Maryland’s hospital global budget program was associated with reductions in the broad overuse of health care services.

• Comparing Maryland to an out-of-state comparison group, we did not find evidence that Maryland hospitals met their revenue targets by reducing systemic overuse.

• Global budgets, in isolation, may be too blunt of an instrument to selectively reduce low-value care. Our findings should encourage policy makers and health system leaders to incorporate low-value service use into the design and evaluation of new payment models.