This study provides insights into relative efficiency in the provision of dialysis services and about the potential factors that affect efficiency, for free-standing dialysis centers in the United States. Although about one-quarter (26.6 percent) of these facilities may be considered efficient (score = 1.0), most facilities (73.4 percent) were relatively inefficient. These findings are largely consistent with the results of Ozgen and Ozcan (21.1 percent efficient) that applied the DEA technique to analyze efficiency in dialysis centers more than a decade ago (Ozgen and Ozcan 2002). In both studies, the percentage of facilities on the efficiency frontier was low compared with that found in other DEA studies of health care providers. For example, a study of U.S. urban hospitals found that 45 percent of hospitals were efficient (Ozcan and Luke 1993). Another study used DEA to analyze the efficiency of acute care nursing units in U.S. hospitals and found 40.4 percent were on the efficiency frontier (Mark et al. 2009). Efficiency of nursing homes was analyzed using DEA and an average DEA score of .869 was reported (DeLellis and Ozcan 2013), compared with a mean of .783 found for dialysis units in this study. This comparative analysis suggests that there is a greater opportunity for improvement in the technical efficiency of dialysis facilities than in other health care institutions.
This study found that organization size/chain affiliation had a significant (negative) association with efficiency; that is, the larger the organization, the less likely that the units in the organization will be efficient. This was also a finding in the dialysis study by Ozgen and Ozcan (2002). In contrast, size and chain affiliation was positively related to efficiency in a study of urban hospitals (Ozcan and Luke 1993). Ozcan and Luke also reported that having a higher percentage of patients with Medicare was strongly related to lower efficiency. As dialysis facilities treat predominantly Medicare patients, this factor may help explain why the dialysis industry appears less efficient than the urban hospitals. In a near single-payer system, dialysis facilities do not compete on price, which may result in little incentive to operate more efficiently.
Other major factors found to be significantly associated with the efficiency of dialysis centers were geographic region of the country and whether facilities were in urban or rural locations. Note that, while organization size may be somewhat within the power of facility owners to change, geographic location and urban versus rural location are factors that, for the most part, cannot be changed because of the localized nature of dialysis treatment. Environmental factors such as local market competition or the percentage of facilities in a local market that were for-profit were not found to be significant indicators of relative efficiency. Being nonprofit was positively associated with efficiency, but it was not significant. This is consistent with studies by both Ozcan and Luke (urban hospitals) and DeLellis and Ozcan (nursing homes), which found that not-for-profits were more likely to be efficient than for-profits. In contrast, Ozgen and Ozcan (2002) found that nonprofit ownership in dialysis centers was significantly negatively associated with efficiency.
In the study reported herein, the effects of a change to a less frequent ESA dosing regimen on dialysis facility efficiency were analyzed by estimating the administration time and supply cost savings that would be associated with once-monthly compared with thrice-weekly dosing, and altering the inputs to the DEA model accordingly. The reduced costs of this change in anemia management did not alter the distribution of efficiency scores in any appreciable way. This is understandable as anemia management staff time was less than 3 percent of total staff time and anemia treatment-related supply costs were less than 1 percent of total nonlabor costs. To alter efficiency statistics appreciably, it would be necessary to affect somewhat larger categories of cost. Furthermore, it is unlikely that real FTE savings could be achieved, as most facilities do not employ sufficient nursing staff to reduce by even a half FTE as a result of the anemia management time savings that might be realized. There may be other inefficiencies in the dialysis center that should be addressed before considering a change in anemia management to become more cost-efficient. For example, the time spent on managing patient laboratory testing, which consumes a substantial amount of staff resources (Vizethann 2012), or implementing protocols to reduce access site infections may be areas where operational and technical improvements could result in substantial reductions in cost and improvements in efficiency.
Although the findings of this study provide informative insights into the technical efficiency of U.S. dialysis centers, the results should be interpreted with caution. Of the various statistical techniques available for frontier analysis in health care, the selection of DEA for this study was made to directly compare with the results of Ozgen and Ozcan (2002), and to reduce specification errors given the perceived questionable quality and non-normal distributional characteristics of the Medicare cost report data used. DEA is a robust and widely regarded methodology, but it has some drawbacks. First, the findings from a DEA study reflect relative efficiency, not absolute efficiency. Although the dialysis facilities in the sample that were operating at the highest level of efficiency compared with their peers were designated as efficient, DEA does not allow determination of operating efficiency according to any external standard. Secondly, the DEA technique has a weakness in that it does not impose an error term in the efficiency model and observed inefficiency is often attributed to poor management (Bryce, Engberg, and Wholey 2000). However, there may be other measurement errors due to factors such as unobserved variables, or output measure(s) may be underidentified. For example, in the definition of the output measure used in this study, it was not possible to account for variations in the quality of care. To the extent that quality varies in the production of dialysis services, this may be a missing contributory component in the input–output equation that determines efficiency. A facility may have been deemed efficient while providing a lower quality of care, or conversely, may use a higher number of inputs to produce higher quality outputs. There may also be differences in output units between dialysis provided in the facility versus self-administered dialysis performed at home, and some centers may appear more or less efficient because of a mix of in-center and home dialysis services. However, in our sample of facilities that provided in-center hemodialysis in 2010, only 9 percent of 4-hour session equivalents were provided in the home setting. As a result, we did not believe that creating a separate output measure was warranted, but this remains a limitation of this study.
A criticism of DEA is that, as a nonstochastic method, it is particularly sensitive to the problems of mismeasurement; one mismeasured firm may define the frontier for all firms based on erroneous data. Although our primary data source was government-audited cost reports, which should provide some assurances of data quality, extensive missing data and outlier edits were applied to the study dataset (as described in Table 1) to minimize the potential for misreporting errors. The efficiency frontier in this study was defined by over one-quarter of all observations scoring 1.0, which provides further reassurance of the accuracy of the frontier definition.
Regarding the regression analysis, multinomial logistic regression assumes independence among the dependent variable choices. However, DEA efficiency scores cannot be independent of each other because the scores are relative, that is, the calculation of the efficiency score for each observation involves all the other observations (Xue and Harker 1999). Violation of this assumption can lead to meaningless predictions; for example, efficiency scores outside of the range of 0 to 1. Although this study did not encounter problems of this type, alternative statistical techniques that can be used to avoid such problems include fractional regression (Ramalho, Ramalho, and Henriques 2010) and bootstrapping (Simar and Wilson 1998).
Additional limitations were the absence of measures to adjust for variability in patient case mix across facilities, and the possibility of selection bias due to the fact that some facilities had to be excluded because of missing data. For the regression model, although efforts were made to specify a model that would reduce the potential for collinearity among the explanatory variables, our results may be biased due to endogeneity in the model (Orme and Smith 1996). Finally, estimated savings related to less frequent ESA dosing were based on theoretical estimates from two time and motion studies, which may not reflect actual savings. As this was a U.S.-specific study, in the free-standing dialysis facility setting only, findings about efficiencies that might be gained through switching to less frequent ESA dosing cannot be generalized to other countries, to other disease states such as oncology, or to the hospital-based dialysis setting. Extended ESA dose frequency may have benefits in other countries, treatment settings, or in the treatment of anemia in other diseases.
The incentives for cutting costs while maintaining quality continue to evolve in the Medicare ESRD program. Efficiency measures are under consideration for discussions relating to the ESRD QIP. In the new era of pay for performance, it will be important to continue to monitor dialysis center efficiency, and to concurrently investigate the use of efficiency models that incorporate quality of care dimensions as inputs, outputs, or both. The DEA approach is a methodology to consider in this discussion, along with a number of enhancements and possible improvements that might be pursued relative to the current model presented. Measures such as mortality rates, hospitalization rates, transfusion rates, and dialysis dose adequacy can be considered for inclusion if differences in quality of care in the production of dialysis services are to be understood and controlled for. Going forward, it will be important to examine the effects of the new ESRD prospective payment system on dialysis center efficiency by extending this work via a longitudinal study. Alternative modeling techniques such as stochastic frontier models or a Malmquist approach (Ozgen and Ozcan 2004) may be useful to consider. Lastly, to the extent that dialysis facilities modify practices in response to the availability of technology innovations such as longer acting ESAs, real-world data should be used in lieu of simulation to understand the effects of these changes.