Association between value-based purchasing score and hospital characteristics

The unit of analysis for our study is a U.S. hospital that is eligible for incentive payment under the VBP program. All Medicare Subsection (d) hospitals are eligible to participate in the VBP program, which includes all acute care hospitals in 50 states and District of Columbia other than rehabilitation hospitals and units; long-term care hospitals (LTCHs); psychiatric hospitals and units; children's hospitals; and cancer hospitals [1]. Furthermore, each of the hospitals must have at least 4 clinical process of care measures and at least 100 Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) surveys to be eligible for the VBP program. Critical Access hospitals are not included. [see Additional file 1: Appendix for further details]. Our data came from three primary sources: (i) Medicare Hospital Compare (HC) Access Database provided hospital specific information for all of the clinical process of care and patient experience measures; (ii) the 2010 American Hospital Association (AHA) Annual Survey database provided data characteristics including nurse staffing level, teaching status, and profit status; (iii) the 2009 Medicare Impact File provided information on the share of low-income patients served by a hospital and patient mix. [See Additional file 1: Appendix for further details] The authors have permission to use the AHA Survey dataset (which Mayo Clinic has purchased), while the other two datasets are publicly available.

A summary of the Medicare VBP program [1] is provided in Additional file 1: Appendix 1 in Supplementary Materials. In short, the VBP score for 2013 is based on a weighted average of either performance or improvement for 12 clinical process of care (CPC) measures and 8 patient satisfaction measures from HCAHPS survey. The HC database was used to extract the individual scores used for VBP score calculation for each hospital, the primary dependent variable in the analysis (Additional file 1: Appendix). The CPC and HCAHPS scores were dependent variables in the secondary analysis. We used performance threshold and benchmark for each measure as reported in the Federal Register [1]. Because HC database provides data only on a yearly basis, the baseline and the performance periods in our study comprised of full-year data as opposed to Medicare’s 3-quarter baseline and performance periods. More specifically, the baseline and performance periods for VBP score calculation in our study were from 4/1/2008 through 3/31/2009, and 4/1/2010 through 3/31/2011, respectively.

Table 1 shows the distribution of the estimated VBP score and its two components (CPC and HCAHPS Scores) overall and by 4 quartiles. The overall mean VBP score was 47, and there was wide variation in each of the scores as reflected by the range.

Hospital characteristics assessed in the study are shown in Table 2, including Medicare-defined disproportionate share index (proportion of low income patients served by the hospital), teaching status, percents of Medicare and Medicaid days to total inpatient days, profit status, geographic region, total number and types of CPC measures reported. Detailed definitions of these variables are provided in Additional file 1: Appendix. Mean and standard deviation (SD) for continuous covariates, and the frequency and the percent for the categorical covariates are provided for the overall sample and for the four quartiles of the estimated VBP score.

The association between the mean VBP score and hospital characteristics was assessed through ordinary least squared (OLS) regression. In selecting variables into the final model, besides including key variables that are expected to be associated with the estimated VBP score (e.g., DHS index, percent of Medicare and Medicaid days to total inpatient days, profit status, bed size, geographic regions), a stepwise forward selection with p-value less than 0.2 was adopted to decide on including other hospital characteristics. The final list of variables included in the model is shown in Table 2. We used conditional quantile regression (QR) to assess the association between various percentiles of the VBP score distribution and hospital characteristics [17, 18]. Since hospitals in the upper tail (e.g., 90^th percentile) of the VBP score distribution are likely to receive the greatest incentive payments, while those in the lower tail (e.g., 10^th percentile or below) are likely to lose money under the VBP program, the QR approach offers insights on the potential determinants of the VBP scores for low-, medium and high-quality hospitals. Additionally, the QR approach helps assess whether a specific hospital characteristic has differential effects on hospitals across different parts of the VBP score distribution. (See Additional file 1: Appendix for further details).

The final sample included 2,491 hospitals with complete observations for all the study variables. Table 2 provides descriptive statistics for hospital characteristics for the overall sample and for the four quartiles of the VBP score distribution.

Reported averages for disproportionate share index, percent of Medicaid inpatient days, and teaching status declined progressively from the first quartile (lower quality) to the fourth quartile (higher quality) of the VBP score distribution (Table 2). The opposite was true of the average nurse staffing level – the average nurse staffing in hospitals in the fourth quartile was 89 registered nurse full time employee (FTE) per 100 bed-days as compared to 74 in hospitals in the first quartile.

The average number of CPC measures reported was 9.7, with lesser mean number of CPC measures reported for 4^th quartile than the 1^st quartile (9.32 vs. 9.76). The distribution of hospitals by profit status across the four quartiles revealed an interesting pattern. The percent of for-profit hospitals increases as one moves from the 1^st to the 4^th quartile of the VBP score distribution; the opposite was true of government hospitals. The distribution of bed categories and geographic regions also differed significantly across the four VBP score quartiles.

Table 3 presents the multivariable results associated with the OLS regression, and the QR estimated at five percentiles of the VBP score, namely, 10^th, 25^th, 50^th, 75^th and 90^th. Figure 1 exhibits the effects of some selected covariates along with their 95% confidence intervals, captured in the OLS and QR frameworks. For the purpose of these graphs, QR was estimated at 10 percentiles (0.1, 0.2, …, 0.9) of the VBP distribution.

Case-mix index, nurse staffing ratio and teaching level had non-significant effects both in OLS and all of QR estimates (not shown in Table 3 to avoid cluttering). Disproportionate share index, percents of Medicare and Medicaid inpatient days were inversely associated with the VBP score. As seen from the corresponding quantile estimates, these covariates had uniform effects across all the 5 percentiles considered.

The total number of CPC measures reported was significantly inversely associated with the mean VBP score, and this effect was increasingly more pronounced going from lower to upper quantiles of the VBP distribution. Compared with non-profit hospitals, for-profit hospitals were likely to have significantly higher mean VBP score. The significant positive association between for-profit status and the VBP score was observed across the entire VBP score distribution. Government hospitals were likely to have lower mean VBP scores and the effects get more pronounced in the uppermost quantile (i.e., 90^th percentile).

Compared with hospitals with 100–199 beds, hospitals with fewer beds had significantly higher VBP score, while those hospitals with 400 or more beds tended to have smaller VBP scores. The analysis also found significant regional differences. Compared to the New England region, hospitals located in Mid Atlantic, West North Central, Mountain and Pacific regions had significantly lower mean VBP scores, while those located in East South Central had significantly higher mean VBP scores. QR estimates revealed heterogeneous effects of geographic location across the VBP score distribution. On average, accreditation by the Joint Commission, hospitals providing obstetric care or wound management services were negatively associated with the VBP score.

Hospitals that reported the following 4 specific CPC measures were likely to have significantly higher mean VBP scores (Table 3): primary PCI within 90 minutes of arrival, prophylactic antibiotic given with 1 hour of incision, cardiac patients with controlled 6 AM postoperative serum glucose, and beta blocker prior to admission and perioperatively. Note however, that these effects got progressively diluted towards the upper quantiles of the VBP score distribution.

The heterogeneous effects of some selected hospital characteristics across different parts of the VBP score distribution are shown in Figure 1, which overlays the quantile estimates with the OLS estimates along with their 95% confidence intervals. As Figure 1 shows, the quantile effects of the number of CPC measures reported was less than the mean effect in the lower tail of the VBP score distribution; however, the effect was substantially higher than the mean effects in the upper tail. The differential effects of profit status and bed size are shown in the other panels of Figure 1. (See Additional file 1: Appendix for geographic regions and percent of Medicare inpatient days)

This sub-analysis addressed the secondary study objective, and was based on the QR framework. The results are shown only for the 90^th percentile of the VBP score (Table 4). The coefficient for the QR effect of case-mix index on the 90^th percentile of the HCAHPS score was 9, while that for CPC score was −8, both statistically significant. These opposing effects resulted in a non-significant net effect of 2.1 on the 90^th percentile of the VBP score. The number of CPC measures reported was negatively associated with both CPC and HCAHPS scores, which translated to a net negative effect on the VBP score.

Profit status had opposite effects on HCAHPS and CPC scores at the 90^th percentile. Compared with non-profit hospitals, for-profit hospitals had negative effects (4 units) on the HCAHPS score but positive (10 units) on the CPC score with a net effect of 7 units on the total VBP score. Government hospitals had no significant effect on HCAHPS score, although it had significantly negative effect on the CPC score that was mediated to the total VBP score.

Our findings allay the concerns that hospitals serving the poor and the elderly are more likely to score lower and hence, likely to be penalized under the VBP program. However, the loss of 1% VBP revenue could significantly influence the future viability of these hospitals that operate under very tight financial constraints. Given that for-profit status explains the most of the variation in the VBP score and potentially the incentive payments, we might witness a gradual change in ownership in the long run. This possibility becomes more real when commercial payers and Medicaid follow suit in adopting value-based purchasing or similar measures. The 2010 ACA has already called for extending value-based purchasing to physician payment [27]. In order to receive highest incentive payments, and remain financially viable, hospitals may have to make structural changes including ownership and types of services offered. As the report to the Congress suggested, it will be of utmost importance to monitor if hospitals deny care to specific groups of patients simply to maintain a high VBP score [14].