Facility Practice Variation to Help Understand the Effects of Public Policy: Insights from the Dialysis Outcomes and Practice Patterns Study (DOPPS)

Abstract

Recent Centers for Medicare & Medicaid Services policies have used dialysis facility practice variation to develop public ratings and adjust payments. In the Dialysis Facility Compare star rating system (DFC SRS), facility-relative rates of performance-based clinical measures varied nearly two-fold for mortality (standardized mortality ratio; 10th/90th percentiles: 0.71, 1.34) and hospitalization (standardized hospitalization ratio; 10th/90th percentiles: 0.64, 1.37), and nearly four-fold for transfusion (standardized transfusion ratio; 10th/90th percentiles: 0.43, 1.65). Medicare claims data (from July of 2014) demonstrate that facility variation for the proportions of patients on hemodialysis hospitalized (10th/90th percentiles: 27%, 50%) and transfused (10th/90th percentiles: 3%, 17%) within 6 months that far exceeds relatively modest recent overall longitudinal trends. DFC SRS–rated facility variation is also substantial for fistula (10th/90th percentiles: 50%, 78%) and catheter use >90 days (10th/90th percentiles: 3%, 19%). By contrast, DFC SRS–rated facility distributions for adult hemodialysis Kt/V>1.2 (10th/90th percentiles: 84%, 97%) and total serum calcium >10.2 mg/dl (median, 1%; 75th/90th percentiles: 3%, 5%) are quite narrow and may be of questionable value. Likewise, variation in the US Dialysis Outcomes and Practice Patterns Study is over two-fold for facility median serum parathyroid hormone (10th/90th percentiles: 290 pg/ml, 629 pg/ml) and ferritin (10th/90th percentiles: 469 ng/ml, 1143 ng/ml) levels, and facility mean treatment time varies by 30 minutes (10th/90th percentiles: 204 minutes, 234 minutes). Rising serum parathyroid hormone and ferritin levels, and generally short dialysis treatment time, represent areas unchecked by existing policy; both overall trends and facility variation in these values may reflect unintended consequences of policy or reimbursement pressures and therefore raise concern. Additionally, outcomes in the transition period from advanced CKD to dialysis remain poor, and policy initiatives and performance accountability in this area remain insufficient. Innovative models of comprehensive care in advanced CKD and the early dialysis period which are more amenable to policy oversight are needed. In summary, facility variation is typically larger than prevailing longitudinal trends, and should not be overlooked. The combination of nationally representative observational databases (e.g., the Dialysis Outcomes and Practice Patterns Study) and ESRD registries can provide policy makers with additional tools to evaluate facility variation, develop policies, and monitor unintended effects.

Introduction

In 2008, the United States Congress passed the Medicare Improvements for Patients and Providers Act (MIPPA), which mandated the creation of a new bundled prospective payment system (PPS, or “bundle”) for dialysis reimbursement by Medicare (1). The PPS added an array of services, including most dialysis-related medications and laboratory services that were previously separately billable, to the fixed dialysis treatment payment (2). Implemented over 4 years beginning in January of 2011, the bundle was expected to reduce costs to Medicare, and there has been progress in that regard (3). However, many stakeholders anticipated that the proposed reductions in reimbursement could substantially affect patient care, particularly with regard to anemia management (4–6).

In part to defend against the possibility of worsening care, the Medicare Improvements for Patients and Providers Act additionally mandated the development of a quality incentive program (QIP) (7). First implemented in 2012 and revised annually thereafter, the QIP defines a set of payment-linked measures for which penalties of up to 2% may be levied against facilities failing to meet national or within-facility improvement targets. In 2015, amid widespread controversy (8–10), Medicare introduced a separate star rating system (SRS) to its Dialysis Facility Compare (DFC) tool. The SRS aims to provide patients on dialysis with a summary measure of provider quality by ranking facilities according to achievement of nine facility-level performance metrics (Table 1). Although the SRS is not directly linked to facility payments, several SRS measures overlap with the payment-linked QIP (Table 2).

Table 1.

Summary of performance measures in the Dialysis Facility Compare star rating system, 2016 (44)

Outcome	Performance Measure
Standardized outcomes	Standardized transfusion ratio
	Standardized mortality ratio
	Standardized hospitalization ratio
Other outcomes 1	Percentage of adult patients receiving treatment through an arteriovenous fistulaa
	Percentage of adult patients who had a catheter >90 da
Other outcomes 2	Percentage of adult patients on hemodialysis who had Kt/V≥1.2a
	Percentage of pediatric patients on hemodialysis who had Kt/V≥1.2a
	Percentage of adult patients on peritoneal dialysis who had Kt/V≥1.7a
	Percentage of adult patients who had 3-month average calcium >10.2 mg/dla

^aMeasure is also used in the payment year 2017 Quality Incentive Program (Table 2).

Table 2.

Summary of payment-linked measures in the payment year 2017 Quality Incentive Program (45)

Payment-Linked Measure
Percentage of patient months on hemodialysis using an arteriovenous fistulaa
Percentage of patient months on hemodialysis with a catheter ≥90 da
Percentage of patient months on hemodialysis with spKt/V≥1.2a
Percentage of patient months on peritoneal dialysis with Kt/V≥1.7a
Percentage of pediatric patient months on hemodialysis with spKt/V≥1.2a
Percentage of patient months with average of serum calcium >10.2 mg/dla
Standardized (bloodstream) infection ratio
Standardized hospital readmission ratio
Administration of the in-center hemodialysis CAHPS survey
No. of months for which facility reports serum phosphorus values
No. of months for which facility reports ESA dosage and hemoglobin/hematocrit

spKt/V, single-pool Kt/V; CAHPS, Consumer Assessment of Healthcare Providers and Systems; ESA, erythropoiesis-stimulating agent.

^aMeasure is also used in the Dialysis Facility Compare star rating system (Table 1).

Additional factors influencing the dialysis business environment since 2011 include the promotion of ESRD Seamless Care Organizations (ESCOs), which as part of the Affordable Care Act, aim to combine primary care with ESRD services under a shared-savings model (11); “bundle rebasing,” which aims to lock in savings from reductions in dialysis medication usage (12); and the introduction to the United States market of new products that are anticipated to increase competition and drive down costs. Examples in the anemia management domain include Mircera (methoxyl polyethylene glycol-epoetin β; Roche, Basel, Switzerland), first marketed for patients on dialysis in the United States in 2014 and now the first erythropoiesis-stimulating agent (ESA) other than Amgen’s Epogen (epoetin alfa) and Aranesp (darbepoetin alfa) to be widely used in United States dialysis units, and ESA biosimilar products with US Food and Drug Administration approval are anticipated as early as 2017 (13,14).

Soon after the new PPS was implemented in January of 2011, the prescribing information for ESAs was updated in June of 2011 in response to new clinical trial data to reflect a more permissive stance for lower hemoglobin levels, and the QIP soon followed suit in July of the same year (7,15). In response to suddenly aligned regulatory, reimbursement (PPS), and payment measure (QIP) policies, United States dialysis providers made rapid adjustments, with dramatic reductions in hemoglobin levels among patients treated with ESA (16) (from median of 11.5 g/dl in January of 2010 to 10.6 g/dl in April of 2012) and mean ESA dosage (17) (from 17,036 units per week in January of 2010 to 10,529 units per week in December of 2013).

Other significant practice changes in United States dialysis practice since August of 2010 have been reported by the US Dialysis Outcomes and Practice Patterns Study (DOPPS) Practice Monitor (DPM), including sustained increases in ferritin (18,19) and parathyroid hormone (PTH) levels (20). These changes had not been overtly predicted before 2011, as had declines in ESA dosage and hemoglobin levels. However, they are likely, at least in part, because of these dialysis-related policy changes (i.e., unintended consequences), and their effects on patient outcomes are uncertain. In October of 2014, the DPM began reporting unadjusted facility-level distributions of hospitalization and transfusion event frequency on the basis of Medicare claims data (21).

Although it is often obscured in public discourse by aggregated trends across dialysis centers, facility practice variation typically is substantially greater than prevailing overall trends, and recent Centers for Medicare & Medicaid Services (CMS) policy efforts (e.g., DFC and QIP) use practice variation to publicly rate performance and adjust payments to dialysis facilities. Believing facility practice variation to be an important source for understanding the effects of public policy changes on dialysis practice (and vice versa), herein we provide an overview of facility variation for current DFC SRS performance measures for dialysis centers using Medicare data and US DOPPS data. We also use US DOPPS data to highlight marked facility variation for practice areas unchecked by current policy that may reflect unintended consequences of existing dialysis policy.

Materials and Methods

For measures on the basis of the DFC SRS (Table 1), we report percentiles (10th, 25th, 50th, 75th, and 90th) and mean values among facilities listed in the publicly available DFC database from January of 2016 (22). Adapted from results shown in the DPM database from August of 2015 (www.dopps.org/DPM), with permission from Arbor Research Collaborative for Health, we additionally present Medicare data since 2010 (100% inpatient and outpatient claims; data use agreement no. 23721) showing unadjusted facility distributions for hospitalization or transfusion within 6 months. Transfusion events each month were identified by the presence of (1) Healthcare Common Procedure Coding System codes P9010–P9011, P9016, P9021–P9022, P9038–P9040, P9051, P9054, and P9056–P9058; (2) claim value code 37; (3) International Classification of Diseases, 9th edition procedure codes 99.03 and 99.04; or (4) Current Procedural Terminology code 36430 (Table 3). Hospitalization events each month were identified by the presence of an inpatient admission claim for any reason. Medicare beneficiaries with ≥1 outpatient ESRD service claim (billing type ‘72×’) were assigned to the facility providing the most outpatient ESRD service claims to the beneficiary that month (minimum of eight claims).

Table 3.

Description of codes used to identify transfusions in Medicare claims

Code Type	Code	Code Description
CPT	36430	Transfusion, blood or blood components
HCPCS	P9010	Blood (whole), for transfusion, per unit
	P9011	Blood, split unit
	P9016	Red blood cells, leukocytes reduced, each unit
	P9021	Red blood cells, each unit
	P9022	Red blood cells, washed, each unit
	P9038	Red blood cells, irradiated, each unit
	P9039	Red blood cells, deglycerolized, each unit
	P9040	Red blood cells, leukocytes reduced, irradiated, each unit
	P9051	Whole blood or red blood cells, leukocytes reduced, CMV-negative, each unit
	P9054	Whole blood or red blood cells, leukocytes reduced, frozen, deglycerol, washed, each unit
	P9056	Whole blood, leukocytes reduced, irradiated, each unit
	P9057	Red blood cells, frozen/deglycerolized/washed, leukocytes reduced, irradiated, each unit
	P9058	Red blood cells, leukocytes reduced, CMV-negative, irradiated, each unit
ICD-9	99.03	Other transfusion of whole blood; transfusion: blood NOS, hemodilution, NOS
	99.04	Transfusion of packed cells
Value	37	Pints of blood furnished

CPT, Current Procedural Terminology; HCPCS, Healthcare Common Procedure Coding System; CMV, cytomegalovirus; ICD-9, International Classification of Diseases 9th edition; NOS, not otherwise specified.

The DOPPS study design and methods have been published (23,24). Briefly, cohorts of 20–40 prevalent patients are randomly selected within nationally representative samples of dialysis facilities in each participating DOPPS country. Patient replenishment occurs every 4 months in DOPPS 4 (2008–2011) and DOPPS 5 (2012–2015), which provide the data used in this review. Monthly or 4-monthly cross-sections of patients in the United States are constructed using published methods, using poststratification weights within each facility to correct for nonconsent because of age, years with ESRD, sex, black race, and diabetes as primary cause of ESRD (25). We present weighted facility distributions for serum PTH and ferritin levels, and dialysis session treatment time. Selected facility distributions are shown graphically using box-whisker plots, in which the mean and median are represented with a diamond and horizontal line, respectively. Error bars denote the 10th and 25th facility percentiles (lower bar) and the 75th and 90th facility percentiles (upper bar).

Results

Distributions of the DFC SRS measures from January of 2016 are in Table 4.

Table 4.

Facility distributions for components of the Dialysis Facility Compare star rating system, January of 2016

Performance Measure		DFC Facility Percentiles
	N	10th	25th	50th	75th	90th
Clinical events (standardized rate ratio)
Mortality (SMR)	5896	0.71	0.84	1.00	1.16	1.34
Hospitalization (SHR)	5992	0.64	0.79	0.97	1.16	1.37
Transfusion (STrR)	5594	0.43	0.64	0.92	1.26	1.65
Other outcomes 1 (% patients in facility)
Fistula use	5944	50	57	65	72	78
Catheter use >90 d	5944	3	6	9	13	19
Other outcomes 2 (% patients in facility)
HD Adequacy (Kt/V≥1.2, adults)	5862	84	89	93	95	97
HD Adequacy (Kt/V≥1.2, pediatrics)	10	22	57	71	90	94
PD Adequacy (Kt/V≥1.7)	1328	68	81	89	93	96
Hypercalcemia (serum Ca ≥10.2 mg/dl)	6095	0	0	1	3	5

DFC, Dialysis Facility Compare; SMR, standardized mortality ratio; SHR, standardized hospitalization ratio; STrR, standardized transfusion ratio; HD, hemodialysis; PD, peritoneal dialysis; Ca, calcium. Results modified from reference 22, with permission.

DFC SRS Standardized Outcomes

Performance measures for clinical outcomes are expressed as relative rates comparing a facility’s observed performance to expected performance given its case-mix, and variation by facility in these measures is high. Relative rate differences of about 30% higher or lower compared with the median facility are observed for mortality (standardized mortality ratio; 10th/90th percentiles: 0.71, 1.34; median, 1.0). Comparisons of percentiles for hospitalization (standardized hospitalization ratio; 10th/90th percentiles: 0.64, 1.37; median, 0.97) and transfusion events (standardized transfusion ratio; 10th/90th percentiles: 0.43, 1.65; median, 0.92) reveal even greater variability.

US DOPPS estimates of the absolute proportions of facility patients with hospitalization and transfusion claims within 6 months are shown in Figures 1 and 2, respectively. Data from July of 2014 indicate that 38% of patients were hospitalized within 6 months in the median facility, and the 10th and 90th facility percentiles were 27% and 50% (nearly two-fold variation), respectively. For transfusions, 9% of patients had a transfusion within 6 months in the median facility, and the 10th and 90th facility percentiles were 3% and 17% (over five-fold variation), respectively. A general time trend toward fewer hospitalizations was observed at all facility percentiles, and a slight increase in transfusions was observed from 2010 into 2011, before a retracement began in 2012–2014. Although these unadjusted, absolute estimates cannot be directly compared with the standardized, relative estimates in the DFC database, the degree of facility variation demonstrated using both methods was generally similar.

Figure 1.

Facility percentiles, percentage of patients in facility with hospitalization claim within 6 months. Values for each month reflect the distribution of the facility percent of Medicare ESRD beneficiaries with a hospitalization claim, among facilities with ≥20 patients. For example, in July of 2014, 10% of facilities have ≤27% patients with a hospitalization claim within 6 months (10th percentile), and 10% of facilities have ≥50% patients with a hospitalization claim within 6 months (90th percentile). ESA, erythropoiesis-stimulating agent; QIP, quality incentive program. Source: Medicare claims, 2010–2014; results modified from reference 27, with permission.

Figure 2.

Facility percentiles, percentage of patients in facility with transfusion claim within 6 months. Values for each month reflect the distribution of the facility percent of Medicare ESRD beneficiaries with a red blood cell transfusion claim, among facilities with ≥20 patients. For example, in July of 2014 10% of facilities have ≤3% patients with a transfusion claim within 6 months (10th percentile), and 10% of facilities have ≥17% patients with a transfusion claim within 6 months (90th percentile). The maximum number of procedures per inpatient claim in this Medicare dataset increased from 6 to 25, starting in January of 2011. ESA, erythropoiesis-stimulating agent; QIP, quality incentive program. Source: Medicare claims, 2010–2014; results modified from reference 27, with permission.

DFC SRS Other Outcomes 1

By widespread consensus in the United States dialysis community, high use of surgical arteriovenous (AV) access remains a top priority. In this context, DFC facility variation in the use of arteriovenous fistula (AVF) and central venous catheter (>90 days) are also substantial. For AVF, the 10th and 90th facility percentiles are 50% and 78%, respectively, and the median facility has 65% of patients dialyzing with a fistula. As AVF prevalence has increased, long-term catheter use (>90 days) has decreased, but facility variation remains high given the importance of this metric; the 10th and 90th facility percentiles are 3% and 19%, respectively, with 9% of patients using a catheter >90 days in the median facility.

DFC SRS Other Outcomes 2

This grouping includes dialysis adequacy (Kt/V≥1.2) and hypercalcemia (serum calcium >10.2 mg/dl), the two DFC SRS measures on the basis of laboratory variables for adult patients on hemodialysis. For serum calcium >10.2 mg/dl, the facility distribution is exceptionally narrow, with median, 75th, and 90th percentiles of 1%, 3%, and 5%, respectively. The facility distribution for dialysis adequacy (Kt/V≥1.2) among adult patients on hemodialysis is somewhat greater, with 10th and 90th percentiles at 84% and 97%, respectively (higher is better; median, 93%).

Selected US DOPPS Variables

In contrast to the DFC SRS hypercalcemia measure, US DOPPS data indicate much greater facility variation for two other laboratory variables measured and managed routinely at all dialysis centers, namely PTH and ferritin levels (Table 5). The facility median PTH level ranges more than two-fold, from 290 pg/ml at the 10th percentile to 629 pg/ml at the 90th percentile, with a median PTH level of 461 pg/ml among patients in the median facility (Table 5). The facility proportion of patients with PTH>600 pg/ml (approximately the revised upper limit now recommended by Kidney Disease Improving Global Outcomes [26]) ranges from 5% at the 10th percentile to 37% at the 90th percentile, with 21% of patients having PTH>600 pg/ml in the median facility (Figure 3). The facility median ferritin level ranges from 469 ng/ml at the 10th percentile to 1143 ng/ml at the 90th percentile, with a median ferritin level of 756 ng/ml among patients in the median facility. The facility proportion of patients with ferritin >800 pg/ml ranges from 12% at the 10th percentile to 77% at the 90th percentile, with 44% of patients having ferritin >800 ng/ml in the median facility (Figure 4).

Table 5.

Facility median (PTH, ferritin) and mean (dialysis treatment time) values in the US DOPPS sample at three time points (2010, 2012, and 2015)

Variable		US DOPPS Facility Percentiles
	N	10th	25th	50th	75th	90th
Median PTH in facility, pg/ml
August 2010	90	245	272	317	381	448
December 2012	189	285	317	390	451	547
April 2015	144	290	381	451	532	629
Median ferritin in facility, ng/ml
August 2010	90	418	527	638	727	928
December 2012	189	528	654	765	905	1081
April 2015	144	469	621	756	952	1143
Mean treatment time in facility, min
August 2010	90	194	205	215	227	234
December 2012	194	198	206	217	227	237
April 2015	146	204	212	219	227	234

PTH, parathyroid hormone; DOPPS, Dialysis Outcomes and Practice Patterns Study. Results modified from reference 27, with permission.

Figure 3.

Facility percentiles, percentage of patients in facility with PTH>600 pg/ml. Values for each month reflect the distribution of the facility percent of patients with PTH>600 pg/ml. For example, in April of 2015 10% of facilities have ≤5% patients with PTH>600 pg/ml (10th percentile), and 90% of facilities have ≥37% patients with PTH>600 ng/ml (90th percentile). ESA, erythropoiesis-stimulating agent; PTH, parathyroid hormone; QIP, quality incentive program. Results modified from reference 27, with permission.

Figure 4.

Facility percentiles, percentage of patients in facility with serum ferritin >800 ng/ml. Values for each month reflect the distribution of the facility percent of patients with ferritin >800 ng/ml. For example, in April of 2015 10% of facilities have ≤12% patients with ferritin >800 ng/ml (10th percentile), and 90% of facilities have ≥77% patients with ferritin >800 ng/ml (90th percentile). ESA, erythropoiesis-stimulating agent; QIP, quality incentive program. Results modified from reference 27, with permission.

Variation in facility mean treatment time is much greater than for the DFC SRS Kt/V≥1.2 measure. Facility distribution in the US DOPPS national sample shows mean treatment times at the 10th and 90th percentiles of 204 and 234 minutes, respectively, with mean treatment time of 219 minutes among patients in the median facility. The facility proportion of patients with treatment time <210 minutes ranges from 42% at the 10th percentile to 87% at the 90th percentile, with 75% of patients having treatment time <210 minutes in the median facility (Figure 5).

Figure 5.

Facility percentiles, percentage of patients in facility with treatment time <210 minutes. Values for each month reflect the distribution of the facility percent of patients with treatment time <210 minutes. For example, in April of 2015 10% of facilities have ≤42% patients with treatment time <210 minutes (10th percentile), and 90% of facilities have ≥87% patients with treatment time <210 minutes (90th percentile). ESA, erythropoiesis-stimulating agent; QIP, quality incentive program. Results modified from reference 27, with permission.

Discussion

Recent experience in the United States under the combined pressures of the new PPS (2011), the updated ESA prescribing guidelines (2011), and evolving payment-linked (QIP) and performance (e.g., SRS) measures suggests that some facility practices change quickly when incentivized. Biochemical and dialysis treatment parameters appear generally responsive to policy interventions applied to individual dialysis facilities (e.g., declines in ESA dosage and hemoglobin levels in 2011, increase over time to a very high proportion with Kt/V≥1.2), perhaps accelerated because of the high proportion of facilities owned by dialysis chains that can rapidly disseminate updated clinical pathways and protocols that individual dialysis centers can incorporate at their discretion.

In the case of policies intended to affect practices easily controlled by dialysis providers, national data sources may not fully support timely monitoring for unplanned consequences, such as greater direct effect of a new policy than anticipated, compensatory changes in practices indirectly related to the new policy, or changes in dialysis facility admission practices for new patients. Observational databases, such as the DPM, offer the dialysis community the opportunity to track their occurrence. For example, the DPM reported an aggregate 50% rise in median PTH levels (from 247 pg/ml in August of 2010 to 371 pg/ml in June of 2015), and a doubling of the number of patients with PTH>600 pg/ml (from 11% to 22%) (27). Similar increases in median serum ferritin levels (23%; from 598 ng/ml to 736 ng/ml) and the proportion of patients with ferritin >800 ng/ml (from 31% to 44%) were also reported. Evidence suggests that the increasing levels overall, and huge facility variation, in PTH and ferritin levels are influenced by the 2011 overhaul of CMS dialysis payment and quality monitoring policies (rise in ferritin due to lower hemoglobin levels and ESA dosage [19]) and, at least in part, the 2009 Kidney Disease Improving Global Outcomes-Mineral Bone Disorder guidelines (rise in PTH due to more permissive PTH upper targets [20,26]). No performance- or payment-linked measures to disincentivize very high PTH or ferritin levels exist in the United States at present because evidence linking these to adverse outcomes is mixed (27–33). At the same time, the variable exposure across dialysis centers to high PTH and ferritin levels gives some in the community pause as the implications for patients are uncertain.

Although many United States dialysis units readily exceed the Kt/V threshold for most patients by reliance on high blood flow rates, many authorities believe that dialysis treatment time is an important measure separate from, or in addition to, Kt/V. In fact, Saran et al. have shown that, adjusted for Kt/V, longer treatment times are associated with better clinical outcomes (34). Although the recent increase that we report in treatment time reflects a slight rebound from the previously declining trend in the United States, prevailing opinion holds that dialysis treatment times in the United States are probably shorter than optimal. Indeed, our international collaborators routinely express disapproval at how short treatment times are in the United States, as well as how high ferritin levels are. Both are outlier practice patterns compared with other DOPPS countries. Clinical trials of dialysis treatment time among patients on incident hemodialysis (Time to Reduce Mortality in End-stage Renal Disease [TiME]; Clinicaltrials.gov identifier: NCT02019225) and intravenous iron dosing on the basis of different ferritin thresholds (Proactive IV Iron Therapy in Haemodialysis patients [PIVOTAL]; UK Clinical Trials Gateway identifier: CPMS15250) are ongoing and will be of value, even if not entirely decisive.

Payment-linked QIP measures (Table 2) are reviewed annually, and analyses are performed to evaluate whether continued improvements in facility measures can be reasonably expected. For example, facility distribution of mean hemoglobin >12 g/dl, one of the original QIP measures, achieved “topped-out” status on the basis of clinical data collected during 2014, and was removed from the QIP for reimbursements beginning in payment year 2016. Current DFC data indicate that the hypercalcemia and hemodialysis Kt/V measures have very little variation and appear to be of questionable value as performance measures used to rank facilities. However, both of these measures still retain value as an ongoing check against possible adoption of practices inconsistent with clinical guidelines, and thus will remain in the QIP at least until 2019 (7,35).

Policy makers should exercise caution in targeting aspects of care that may be beyond the ability of dialysis providers or performance-based measures alone to modify. Measures on the basis of clinical outcomes (e.g., hospitalization and transfusion) cannot be attributed solely to practices of the dialysis clinic and therefore see slower change. Coordination-of-care approaches with shared savings incentives represent another path forward, and disease management demonstration projects have shown promising improvements in delivering care at lower costs (36,37). ESRD ESCOs represent the next evolution of this idea (11). Promoted by the Affordable Care Act, ESCOs combine ESRD services with primary medical care coordinated under a single authority. Thirteen such projects are currently in demonstration; if successful, these new models have the potential to become a reliable source of data from which performance measures for clinical outcomes could be developed, and to provide a care structure that is amenable to policy oversight. Attention to the patient experience, and including patient care/advocacy groups in the policy-making process is also important.

Despite the anticipated success of ESCOs and other managed-care models, the CKD-dialysis transition period remains an area where thoughtful and coordinated policy interventions are lacking or absent, despite having quite poor outcomes. Robinson et al. showed that United States mortality is highest in the first 120 days after dialysis start, at 33 per 100 patient years compared with 22 per 100 patient years in the subsequent 8 months (38). Although this pattern of elevated mortality soon after dialysis initiation was observed in each of the DOPPS countries surveyed, the United States mortality rates were among the highest in the study. Pisoni et al. showed that catheter use among patients on incident hemodialysis (<60 days) in the US DOPPS was 67% (data from the US Renal Data System 2013: 80% at hemodialysis initiation, 68% after 90 days [17]), lower than only Canada, Belgium, and the six countries comprising the Gulf Cooperation Council, despite only 27% of patients having 1 month or less of predialysis nephrology care (39). Lastly, preliminary patient survey data from United States patients in the new Chronic Kidney Disease Outcomes and Practice Patterns Study (CKDopps) (40) found 57%–60% of patients with GFR<30 ml/min per 1.73 m² were unsure what treatment they would choose if their kidneys failed completely in the next month (41). Some dialysis provider-based initiatives targeting incident dialysis patients have reported encouraging improvements in early dialysis survival and higher use of preferred (surgical) vascular access methods during the first year of dialysis (42,43). However, careful investigations of predialysis CKD practice and transition to dialysis care (e.g., CKDopps) are needed to identify optimal practice patterns that can facilitate the design of meaningful interventions to improve the experiences and outcomes of these patients in communities across the country.

Clinical outcomes measures (e.g., standardized hospitalization and mortality ratios) reported in the DFC SRS are standardized to the actual case-mix within each facility, and consequently cannot be used to assess longitudinal trends directly. By contrast, the DPM reports longitudinal trends of the proportion of patients overall and by facility that have a hospitalization or transfusion claim for Medicare reimbursement during subsequent 1- or 6-month periods. Although the DPM estimates are unadjusted for case-mix, they can be interpreted on an absolute basis and indicate, for example, that in the median facility the proportion hospitalized during the following 6 months has commendably declined from 42% in January of 2010 to 38% in July of 2014 (27). This information is not available in the DFC database but, in our view, is of relevance to patients, providers, payers, and policy personnel. Importantly, variations in clinical outcomes between facilities (whether relative or absolute) remain substantial and are typically much larger than the general longitudinal trends.

Limitations of our study include our use of unadjusted estimates to illustrate facility variation in US DOPPS data. Facility variation may be augmented or diminished by a number of factors unrelated to modifiable facility preferences. Examples include patient case-mix, environmental factors (e.g., geography), local economic conditions, and statistical artifacts. Although we feel it is more likely that the residual variation in the DFC measures (which are risk-adjusted) reflects unwarranted variation in practice rather than external random variation from broader contextual effects or unmeasured patient variables alone, careful assessment of the sources of facility variation and their magnitudes should be made when policy makers consider new incentives and evaluate their effects.

In summary, recent CMS policy efforts demonstrate an interest in dialysis facility variation to identify outlier practices for public rating (i.e., DFC SRS) and payment adjustment (i.e., QIP). Facility variation is typically larger than prevailing longitudinal trends and therefore should not be overlooked as an important tool to guide policy development in dialysis care. Nationally representative (but sample-based) databases with comprehensive observational data collection, such as the DOPPS and the new CKDopps, in conjunction with national data for fewer variables collected by ESRD registries, can provide policy makers with necessary information to understand whether and what policy incentives may be needed to manage unwarranted variation and to rapidly evaluate the results, and unintended effects, of such policies.

Disclosures

B.M.R. has received speaker fees for Kyowa Hakko Kirin. D.S.F. has no relevant financial interests to disclose.