Key Points
Question
What is the association between Medicare’s Comprehensive End-Stage Renal Disease Care model with Medicare payments and quality of care for beneficiaries with end-stage renal disease?
Findings
This economic evaluation including 133 558 Medicare patients with end-stage renal disease found that, in its first 2 performance years, the Comprehensive End-Stage Renal Disease Care model was associated with lowered Medicare payments to providers and improved performance health care use and quality-of-care measures. Lower payments were primarily associated with reduced numbers of hospitalizations and readmissions; however, Medicare experienced net losses when shared savings payments were taken into account.
Meaning
Improved care coordination for Medicare beneficiaries with end-stage renal disease and better adherence to dialysis may reduce cost and improve quality.
Abstract
Importance
Medicare beneficiaries with end-stage renal disease (ESRD) are a medically complex group accounting for less than 1% of the Medicare population but more than 7% of Medicare fee-for-service payments.
Objective
To evaluate the association of the Comprehensive End-Stage Renal Disease Care (CEC) model with Medicare payments, health care use, and quality of care.
Design, Setting, and Participants
In this economic evaluation, a difference-in-differences design estimated the change in outcomes for 73 094 Medicare fee-for-service beneficiaries aligned to CEC dialysis facilities between the baseline (from January 2014 to March 2015) and intervention periods (from October 2015 to December 2017) relative to 60 464 beneficiaries at matched dialysis facilities. In the CEC model, dialysis facilities, nephrologists, and other providers partner to form ESRD Seamless Care Organizations (ESCOs), specialty-oriented accountable care organizations that coordinate care for beneficiaries with ESRD. ESCOs with expenditures below a benchmark set by the Centers for Medicare & Medicaid Services are eligible to share in savings if they meet quality thresholds. A total of 685 dialysis facilities affiliated with 37 ESCOs participated in the CEC model as of January 2017. Thirteen ESCOs joined the CEC model on October 1, 2015 (wave 1), and 24 ESCOs joined on January 1, 2017 (wave 2). Patients with ESRD who were aligned with CEC dialysis facilities were compared with patients at matched dialysis facilities.
Main Outcomes and Measures
Medicare total and service-specific payments per beneficiary per month; hospitalizations, readmissions, and emergency department visits; and select quality measures.
Results
Relative to the comparison group (n = 60 464; 55% men; mean [SD] age, 63.5 [14.4] years), total Medicare payments for CEC beneficiaries (n = 73 094; 56% men; mean [SD] age, 63.0 [14.4] years) decreased by $114 in payments per beneficiary per month (95% CI, −$202 to −$26; P = .01), associated primarily with decreases in payments for hospitalizations and readmissions. Payment reductions were offset by shared savings payments to ESCOs, resulting in net losses of $78 in payments per beneficiary per month (95% CI, −$8 to $164; P = .07). Relative to the comparison group, CEC beneficiaries had 5.01 fewer hospitalizations per 1000 beneficiaries per month (95% CI, −8.45 to −1.56; P = .004), as well as fewer catheter placements (CEC beneficiaries with catheter as vascular access for periods longer than 90 days decreased by 0.78 percentage points [95% CI, −1.36 to −0.19; P = .01]) and fewer hospitalizations for ESRD complications (CEC beneficiaries were 0.11 percentage points less likely [95% CI, −0.20 to −0.02; P = .01] to be hospitalized in a given month). Total dialysis sessions and payments increased, suggesting improved adherence to dialysis treatments.
Conclusions and Relevance
Early findings from the CEC model demonstrate that a specialty accountable care organization model focused on a particular population was associated with reduced payments and improved quality of care. Future research can assess the longer-term outcomes of the CEC model and its applicability to populations with other complex chronic conditions.
This economic evaluation assesses the association of the Comprehensive End-Stage Renal Disease Care model with Medicare payments, health care use, and quality of care.
Introduction
End-stage renal disease (ESRD) is a debilitating condition characterized by a high mortality rate, reduced quality of life, and extensive health care use and spending.1,2,3,4,5 Although a kidney transplant is generally the ideal mode of treatment for patients with ESRD, because of a shortage in the United States of organs from deceased donors and the unsuitability of some patients for transplant, most patients with ESRD are dependent on kidney replacement therapy.6 In 2016, 63% of all patients with ESRD received hemodialysis (98% in a center and 2% at home), 7% received peritoneal dialysis, and 30% received a functioning kidney transplant.5
Medicare is the primary insurer for people with ESRD in the United States. In 2016, there were 618 818 beneficiaries with ESRD with Medicare as a primary or secondary payer (81% of all patients in the United States with ESRD).5 Although they comprised less than 1% of the Medicare population, beneficiaries with ESRD accounted for more than 7% of Medicare fee-for-service (FFS) payments.5
Because of the high morbidity rate and increased payments associated with ESRD, the Centers for Medicare & Medicaid Services (CMS) launched various initiatives during the past decade to reduce Medicare expenditures while preserving or enhancing quality. One such initiative is the Comprehensive ESRD Care (CEC) model, administered through the CMS Center for Medicare and Medicaid Innovation (ie, the Innovation Center). Based on the accountable care organization (ACO) concept, dialysis facilities, nephrologists, and other providers and suppliers form ESRD Seamless Care Organizations (ESCOs) to coordinate care for beneficiaries with ESRD and are accountable for quality and financial outcomes for their aligned Medicare beneficiaries. For the pre-2016 period, the exclusion regarding enrollment in a designated CMS shared savings model encompassed alignment with the Independence at Home program, Pioneer ACO, and the Medicare-Medicaid Coordination Office Financial Alignment Initiative. For 2016 and later, this exclusion encompassed alignment with the Independence at Home program, Pioneer ACO, Medicare Shared Savings Program when the beneficiary was categorized as Track 3, Financial Alignment Initiative program, and the Next Generation ACO program. ESCOs are eligible for shared savings if they lower Medicare Part A and Part B payments and meet quality standards. Dialysis organizations with more than 200 facilities (large dialysis organizations) must share any financial losses, whereas nonlarge dialysis organizations can choose 1-sided or 2-sided risk. ESCOs are unique in that they are specialty-based ACOs and services that are organized around the dialysis facility and nephrologist as opposed to a primary care clinician.
To be eligible to participate in the CEC model, ESCOs had to be able to serve 350 beneficiaries with Medicare FFS as the primary payer and to recruit nephrologists and dialysis facilities as owners. For large dialysis organizations, they had to be willing to accept the downside risk. In the first performance year (PY1), which began on October 1, 2015, 13 ESCOs (wave 1) applied for participation and met the model’s requirements. Another 24 ESCOs applied and met participation criteria in the second PY (PY2), which started on January 1, 2017 (wave 2).6 A total of 71 677 Medicare FFS beneficiaries, or 14% of the population with ESRD, were aligned with ESCOs between September 1, 2015, and December 31, 2017.
Methods
Data Sources
Medicare Part A and Part B enrollment and claims data from January 1, 2014, to December 31, 2017, were used to determine beneficiary demographic and enrollment characteristics, chronic condition indicators, and service use. These data were linked to the ESRD Medicare Patient Registration (CMS Form 2728), which contains the dialysis start date, cause of ESRD, and pre-ESRD nephrology care. Dialysis facility organizational characteristics and quality metrics were derived from the Dialysis Facility Compare annual files. Area Health Resource Files from 2014 to 2017 provided market socioeconomic characteristics and health care supply indicators. The lists of ESCO facilities and aligned beneficiaries were obtained from the CMS. Data on the costs to providers of implementing changes under the CEC were not observed. This study was exempt from institutional review board review on the basis of the federal common rule (section 45 CFR 46.101[b][5]), because the purpose of the study was to evaluate a public benefit program.
CEC Population
The CEC population included Medicare FFS beneficiaries who were aligned to a dialysis facility that joined an ESCO (referred to as CEC facilities) at any time during the study period (2014-2017). To be aligned, beneficiaries had to reside in the United States, be 18 years of age or older prior to the first day of the month, not have received a kidney transplant within the previous 12 months, not be enrolled in a designated CMS shared savings model, and have Medicare as the primary payer. For the pre-2016 period, the exclusion regarding enrollment in a designated CMS shared savings model encompassed alignment with the Independence at Home program, Pioneer ACO, and the Medicare-Medicaid Coordination Office Financial Alignment Initiative. For 2016 and later, this exclusion encompassed alignment with the Independence at Home program, Pioneer ACO, Medicare Shared Savings Program when the beneficiary was categorized as Track 3, Financial Alignment Initiative program, and the Next Generation ACO program.
Comprehensive ESRD Care beneficiaries were first aligned to an ESCO when the earliest dialysis service during the study period was performed. They could be de-aligned at the end of every calendar year for the next PY if they no longer met the inclusion criteria, died, or no longer received treatment at any of the facilities included in their ESCO during that year.7 Becoming aligned to another CMS shared savings model or program or no longer receiving treatment at any of the facilities included in their ESCO during that year were de-alignment rules new to PY2; however, we applied all de-alignment rules to both PYs. Although de-alignment occurred at the end of the year, these patients did not remain in the analytic sample if they died or received a transplant. Monthly eligibility rules were also applied in the analysis that excluded observations of eventually de-aligned patients at the point that an alignment or eligibility rule was violated. For example, a patient who received a kidney transplant in June 2014 would be de-aligned December 31, 2014; however, all monthly observations after and including the transplant month would be excluded from analysis via eligibility rules. In addition, patients who were de-aligned from ESCOs were never part of the comparison group. In those cases, member months in the year they were aligned to an ESCO counted toward the CEC population. Beneficiaries who were de-aligned could be realigned to an ESCO in a later year if they resumed dialysis service at any CEC facility and met the eligibility criteria, but they were permanently excluded from the comparison group.
Comparison Population
The comparison population included Medicare FFS beneficiaries who were aligned to a matched comparison facility at any time during the study period. Eligibility, alignment, and de-alignment were assessed using the same methods as the CEC population. To determine the pool of comparison facilities used in the matching model, a series of eligibility criteria were applied (eFigure 1 in the Supplement). Matched comparison dialysis facilities were identified using propensity score matching without replacement, in which each CEC facility was matched to a non-CEC facility with the lowest absolute difference in log odds propensity score. The distributions of propensity scores for the CEC and comparison groups were more closely aligned after the match (eFigure 2 in the Supplement). The propensity score was based on market and facility characteristics and historical (2012-2014) quality and payment outcomes (Table 1). Comprehensive ESRD Care and non-CEC facilities with incomplete claims, facility organization information, or quality metrics during the study period were excluded from the propensity score model, which excluded 53 of the 685 CEC facilities from the analysis. There were no meaningful differences in the mean values of the market-level and facility-level characteristics between the 53 unmatched CEC facilities and the facilities included in the analysis. Additional restrictions were imposed on non-CEC facilities, as described in the eAppendix in the Supplement.
Table 1. CEC Facilities and Non-CEC Facilities Before and After Matching, Key Characteristics, January 2012 Through December 2014a.
| Characteristic | Prematch | Postmatch | ||||
|---|---|---|---|---|---|---|
| All CEC facilities (n = 685) | Non-CEC facilities (n = 4019) | Standardized difference | CEC facilities (n = 632) | Comparison facilities (n = 632) | Standardized difference | |
| Market characteristics (2014) | ||||||
| ESRD beneficiary population in market >350 individuals, mean (SD), % | 96 (20) | 79 (40) | 0.52 | 96 (20) | 94 (24) | 0.09 |
| Beneficiaries ≥65, mean (SD), % | 13 (2) | 13 (3) | −0.25 | 13 (2) | 13 (3) | −0.21 |
| Race/ethnicity, mean (SD), % | ||||||
| White | 59 (15) | 63 (19) | −0.20 | 59 (14) | 61 (17) | −0.09 |
| Black | 17 (10) | 14 (11) | 0.28 | 17 (10) | 16 (11) | 0.17 |
| No high school diploma, mean (SD), % | 14 (4) | 15 (5) | −0.19 | 14 (4) | 14 (4) | 0.04 |
| Single-parent households with children, mean (SD), % | 33 (5) | 34 (6) | −0.09 | 33 (5) | 34 (6) | −0.05 |
| Beneficiaries with ESRD, mean (SD), % | 0.13 (0.04) | 0.13 (0.04) | −0.12 | 0.13 (0.03) | 0.13 (0.04) | 0.03 |
| Beneficiaries dually eligible for Medicaid and Medicare, mean (SD), % | 3 (1) | 3 (1) | −0.31 | 3 (1) | 3 (1) | −0.08 |
| Beneficiaries with ESRD dually eligible for Medicaid and Medicare, mean (SD), % | 51 (8) | 52 (10) | −0.14 | 50 (8) | 51 (10) | −0.05 |
| Median household income, mean (SD), $ | 56 147 (9309) | 52 283 (10 538) | 0.39 | 56 008 (9330) | 54 967 (12 147) | 0.10 |
| Medicare Advantage penetration, mean (SD), % | 28 (14) | 27 (13) | 0.05 | 28 (14) | 29 (13) | −0.09 |
| Primary care providers per 10 000, mean (SD), No. | 7.8 (1.5) | 7.6 (1.7) | 0.09 | 7.8 (1.5) | 7.7 (1.44) | 0.03 |
| Skilled nursing facility beds per 10 000, mean (SD), No. | 48.3 (18.0) | 51.2 (20.6) | −0.15 | 48.4 (18.1) | 49.5 (19.6) | −0.06 |
| Specialists per 10 000, mean (SD), No. | 11.4 (4.9) | 10.1 (4.6) | 0.27 | 11.4 (4.8) | 10.8 (4.3) | 0.12 |
| Hospitals with kidney transplant services per 10 000, mean (SD), No. | 0.01 (0.01) | 0.01 (0.01) | 0.05 | 0.01 (0.01) | 0.01 (0.01) | 0.00 |
| Rural, mean (SD), % | 8 (28) | 16 (36) | −0.22 | 8 (28) | 11 (31) | −0.10 |
| Extra-rural, mean (SD), % | 0 (7) | 6 (23) | −0.31 | 0 (6) | 0 (7) | −0.03 |
| Facility characteristics (2012-2014) | ||||||
| No. of dialysis stations, mean (SD) | 19.7 (8.4) | 18.3 (7.7) | 0.16 | 20.4 (8.0) | 20.1 (7.9) | 0.04 |
| Late shift (facility is open after 5 pm), mean (SD), % | 22 (41) | 18 (38) | 0.09 | 23 (42) | 23 (42) | 0.00 |
| Peritoneal service offered, mean (SD), % | 47 (50) | 60 (49) | −0.26 | 46 (50) | 53 (50) | −0.14 |
| Patients receiving hemodialysis, mean (SD), % | 94 (14) | 94 (9) | −0.01 | 96 (9) | 95 (8) | 0.09 |
| Patients receiving peritoneal dialysis, mean (SD), % | 8 (17) | 9 (12) | −0.02 | 6 (12) | 8 (10) | −0.10 |
| Patients with vascular catheter, mean (SD), % | 10 (5) | 11 (7) | −0.20 | 10 (5) | 11 (6) | −0.16 |
| Patients with arteriovenous fistula, mean (SD), % | 62 (10) | 63 (11) | −0.14 | 62 (10) | 63 (11) | −0.11 |
| Standardized ratio, mean (SD) | ||||||
| Hospitalization | 0.99 (0.25) | 0.99 (0.27) | 0.00 | 0.99 (0.25) | 1.01 (0.28) | −0.06 |
| Readmission | 0.96 (0.29) | 0.97 (0.30) | −0.03 | 0.96 (0.28) | 0.97 (0.29) | −0.04 |
| Mortality | 0.95 (0.22) | 1.01 (0.28) | −0.27 | 0.95 (0.22) | 0.96 (0.23) | −0.06 |
| Ownership, mean (SD), % | ||||||
| DaVita | 16 (36) | 44 (50) | −0.65 | 15 (36) | 22 (41) | −0.18 |
| DCI | 9 (29) | 3 (16) | 0.28 | 9 (29) | 9 (28) | 0.02 |
| Fresenius | 72 (45) | 21 (41) | 1.21 | 73 (45) | 67 (47) | 0.12 |
| Total Medicare payments PBPM, mean (SD), $ | 6557 (1861) | 6499 (1174) | 0.04 | 6445 (828) | 6427 (994) | 0.02 |
| Patients with no prior nephrology care, mean (SD), % | 45 (13) | 46 (15) | −0.08 | 45 (12) | 45 (14) | 0.00 |
| Patients new to dialysis, mean (SD), % | 11 (10) | 12 (9) | −0.15 | 10 (6) | 10 (6) | −0.08 |
| Facility payments relative to market Medicare Part A and Part B PBPM ratio, mean (SD)b | 1.02 (0.28) | 1.02 (0.15) | −0.02 | 1.00 (0.10) | 1.00 (0.12) | 0.01 |
Abbreviations: CEC, Comprehensive ESRD Care; ESRD, end-stage renal disease; PBPM, per beneficiary per month.
Unadjusted proportions and means are shown. Characteristics were based on 2012-2014 data, prior to announcement of the start of the CEC model.
Mean Medicare payment PBPM across Medicare beneficiaries with ESRD receiving dialysis services at the facility, divided by the mean Medicare payments PBPM across all Medicare beneficiaries with ERSD receiving service in the same core-based statistical area.
The final sample consisted of 73 094 CEC beneficiaries and 60 464 comparison beneficiaries. Matched comparison facilities were not statistically different from CEC-participating facilities with respect to market-level and facility-level characteristics (Table 1).
Outcomes
Monthly total Medicare Part A and Part B and service-level payments (eg, hospitalizations, readmissions, and outpatient dialysis) were calculated for CEC and comparison beneficiaries during the months they were aligned to a CEC or comparison facility and were CEC eligible. Payments were standardized to remove the effects of Medicare’s geographic wage, teaching, and other payment adjustments and are presented as per beneficiary per month (PBPM).
Utilization measures of hospitalizations, emergency department visits, readmissions within 30 days of a hospitalization, and outpatient dialysis sessions were assessed as the number of events in a month and summarized as mean rates per 1000 beneficiaries per month. Thirty-day readmissions were calculated for beneficiaries who had at least 1 hospitalization during a given month.
Quality of care was assessed monthly through the following adverse events: catheter use for hemodialysis for more than 90 days, hospitalizations for complications associated with vascular access or ESRD care, and use of emergency dialysis. Emergency dialysis is an unscheduled or emergency dialysis treatment in a hospital outpatient department that is not certified as an ESRD facility. An ESRD complication was based on International Classification of Diseases, Ninth Revision (ICD-9) diagnosis codes 27650, 27651, 27652, 2767, 27669, 40403, 40413, 40493, 5184, 514, 4281, and 428x (ie, the first 3 digits are 428) and International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) diagnosis codes E860, E861, E869, E875, E8770, E8779, I132, J810, J811, and I50x (ie, first 3 digits are I50). A vascular access complication was based on ICD-9 diagnosis codes 9961, 99656, and 99673 and ICD-10 diagnosis codes T82318A, T82319A, T82328A, T82329A, T82338A, T82339A, T82398A, T82399A, T8241XA, T8242XA, T8243XA, T8249XA, T82510A, T82511A, T82518A, T82520A, T82521A, T82528A, T82529A, T82530A, T82531A, T82538A, T82590A, T82591A, T82598A, T85611A, T85621A, T85631A, T85691A, T82818A, T82828A, T82838A, T82848A, T82858A, T82868A, and T82898A. Use of a catheter, which is associated with higher infection and complication rates (this outcome is restricted to only hemodialysis beneficiaries with at least 3 months of hemodialysis),8 is the least preferred permanent form of vascular access. Vascular access complications included malfunctioning of the vascular device, leakage, and hemorrhage of vascular grafts. End-stage renal disease care complications included volume depletion, hyperpotassemia, fluid overload, heart failure, and pulmonary edema. Use of emergency dialysis included unscheduled or emergency dialysis treatments in a hospital outpatient department that was not certified as an ESRD facility. Quality results are presented as the percentage of aligned beneficiaries who experienced at least 1 event in a given month.
Statistical Analysis
The analysis uses a difference-in-differences design to assess the CEC model by comparing changes in outcomes for the CEC population before and after model implementation with changes for the comparison population. This approach controls for beneficiary-level, market-level, and facility-level differences between the CEC and comparison populations, minimizes biases from time-invariant differences between the CEC and comparison populations, and controls for secular trends. The analysis period was divided into baseline, transition, and intervention for each CEC start date. The baseline period for facilities that joined the CEC model in October 2015 was from January 2014 through March 2015, followed by a 6-month transition period from April 2015 through September 2015 to account for the delayed start of the model. The baseline period for facilities that joined the CEC model in January 2017 was from January 2014 through June 2016, followed by a 6-month transition period from July 2016 through December 2016. Wave 1 ESCOs had facilities joining the CEC model in both periods, while all wave 2 ESCO facilities joined the CEC model in January 2017. Performance year 1 was from October 2015 to December 2016 and included wave 1, while PY2 was from January 2017 to December 2017 and included waves 1 and 2.
Multivariate regression models were estimated with specifications suited for each outcome as described in the eAppendix in the Supplement. Covariates included beneficiary demographics (eg, age, sex, disability, and Medicaid status) and clinical risk factors (eg, reason for ESRD, number of months receiving dialysis, comorbidities, and body mass index at dialysis incidence), facility characteristics (eg, size, quality metrics, and organization type), market characteristics (eg, ACO penetration, population size, mean Medicare payments, percentage eligible for Medicaid, and medical provider supply indicators), and quarterly fixed effects (eTable 1 in the Supplement). Standard errors for nonlinear models were calculated using the delta method, allowing for 2-way clusters at the beneficiary and dialysis facility levels (2-way clusters account for intracluster correlation among beneficiaries receiving services from the same facility and correlation across observations from the same beneficiary across time).9 Models were specified to estimate the overall outcomes as well as differential outcomes by ESCO wave and by PY. A core assumption of the difference-in-differences design is that the intervention and comparison populations have parallel trends for a given outcome during the baseline period. Parallel trend tests were conducted for all outcomes at the 5% level. All outcomes, except catheter use for wave 1 facilities, passed parallel trend tests (eTable 2 in the Supplement). Although statistical trend tests of wave 1 catheter use did not pass, visual inspection of the relative trends to the comparison group appeared parallel. In addition, the coefficient of the difference in trends at baseline, although significant, equalled 0.00046. Overall, parallel trends for all ESCO groups were achieved. All P values were from 2-sided tests and results were deemed statistically significant at P < .05.
Sensitivity analysis was conducted for total payments (eFigure 3 in the Supplement). These analyses included adding wave fix effects and extending the transition period for 1 quarter. We also conducted a secondary analysis stratifying the sample by number of months receiving dialysis.
Results
Comprehensive ESRD Care–participating facilities differed from nonparticipating facilities along several market-level and facility-level characteristics (Table 1). Compared with nonparticipating facilities (n = 4019) prior to the start of the model, CEC facilities (n = 685) were less rural (mean [SD] percentage rural, 8% [28%] vs 16% [36%]) and were located in larger markets that had a lower mean (SD) proportion of white beneficiaries with ESRD (59% [15%] vs 63% [19%]) and had higher median incomes ($56 147 vs $52 283) and more specialists per 10 000 population (mean [SD], 11.4 [4.9] vs 10.1 [4.6]). Comprehensive ESRD Care facilities had more dialysis stations (mean [SD], 19.7 [8.4] vs 18.3 [7.7]), lower standardized risk-adjusted mortality rates (mean [SD], 0.95 [0.22] vs 1.01 [0.28]), were more likely to be Fresenius facilities (mean [SD], 72% [45%] vs 21% [41%]), and were less likely to provide peritoneal dialysis (mean [SD], 8% [17%] vs 9% [12%]). Wave 1 facilities had higher Medicare payments, standardized risk-adjusted hospitalization rates, and readmission rates than nonparticipating facilities (eTable 3 in the Supplement). In contrast, wave 2 facilities had lower payments, standardized hospitalization rates, and readmission rates than nonparticipating facilities. Beneficiary characteristics were similar between wave 1 and wave 2 facilities except that there was a higher proportion of CEC beneficiaries with full Medicaid benefits in wave 1 (eTable 4 in the Supplement). Beneficiaries at CEC facilities were also generally similar to those at comparison facilities, except that there was a slightly lower percentage of black beneficiaries at comparison facilities. Beneficiaries who were de-aligned from CEC and comparison facilities had similar patterns in reasons for de-alignment (eTable 5 in the Supplement) and were similar to those who were de-aligned from comparison facilities (eTable 6 in the Supplement).
Mean total Medicare payments to providers decreased from $6315 PBPM in the baseline (pre-CEC) period to $6199 PBPM in the intervention period for CEC beneficiaries. The comparison group mean total Medicare payments decreased from $6317 PBPM in the pre-CEC period to $6315 PBPM in the intervention period. Relative to comparison beneficiaries, the difference in mean payments was a net decrease of $114 PBPM (95% CI, −$202 to −$26; P = .01) for CEC beneficiaries, amounting to a savings of 1.8% (Table 2). However, when shared savings payments of $247 PBPM are considered, Medicare experienced net losses of $78 PBPM (95% CI, −$8 to $164; P = .07). Relative to the comparison group, inpatient payments for CEC beneficiaries decreased by $68 PBPM (95% CI, −$112 to −$24; P = .003), and readmission payments decreased by $29 PBPM (95% CI, −$57 to −$2; P = .04). Payments for hospitalizations owing to ESRD complications also decreased by $10 PBPM (95% CI, −$19 to −$0.49; P = .04) for CEC beneficiaries relative to the comparison group. Total dialysis payments increased for CEC beneficiaries relative to the comparison group by $15 PBPM (95% CI, $7-$24; P < .001). All payment results were associated primarily with wave 1 ESCOs.
Table 2. Effect Estimates for CEC Medicare Payments, PBPM, by Wave and PYa.
| Measure | CEC facility mean before start of CEC model, $ | Difference-in-differences estimate (95% CI), $a,b | Change relative to mean before start of CEC model, % | P value |
|---|---|---|---|---|
| Total Medicare payments Part A and Part B | ||||
| All ESCOs (PY1 and PY2) | 6315 | −114 (−202 to −26)c | −1.8 | .01 |
| Wave 1 | ||||
| PY1 | 6316 | −123 (−242 to −5)c | −2.0 | .04 |
| PY2 | 6316 | −176 (−296 to −56)c | −2.7 | .004 |
| Wave 2 PY2 | 6329 | −31 (−126 to 65) | −0.5 | .53 |
| Acute inpatient | ||||
| All ESCOs (PY1 and PY2) | 1634 | −68 (−112 to −24)c | −4.1 | .003 |
| Wave 1 | ||||
| PY1 | 1635 | −64 (−124 to −3)c | −3.9 | .04 |
| PY2 | 1635 | −106 (−166 to −46)c | −6.5 | .001 |
| Wave 2 PY2 | 1657 | −9 (−25 to 6) | −2.0 | .23 |
| Readmissionsd | ||||
| All ESCOs (PY1 and PY2) | 563 | −29 (−57 to −2)c | −2.8 | .04 |
| Wave 1 | ||||
| PY1 | 563 | −32 (−68 to 4) | −2.7 | .28 |
| PY2 | 563 | −45 (−84 to −6)c | −5.0 | .03 |
| Wave 2 PY2 | 573 | −5 (−42 to 33) | −0.2 | .94 |
| Hospitalizations for ESRD complications | ||||
| All ESCOs (PY1 and PY2) | 149 | −10 (−19 to −0.49)c | −6.5 | .04 |
| Wave 1 | ||||
| PY1 | 149 | −16 (−27 to −5)c | −10.5 | .006 |
| PY2 | 149 | −9 (−22 to 5) | −5.8 | .20 |
| Wave 2 PY2 | 152 | −0.19 (−14 to 13) | −0.1 | .98 |
| Total dialysis | ||||
| All ESCOs (PY1 and PY2) | 2595 | 15 (7 to 24)c | 0.6 | <.001 |
| Wave 1 | ||||
| PY1 | 2595 | 14 (4 to 24)c | 0.6 | .006 |
| PY2 | 2595 | 19 (7 to 32)c | 0.8 | .003 |
| Wave 2 PY2 | 2595 | 11 (−1 to 24) | 0.4 | .07 |
Abbreviations: CEC, Comprehensive ESRD Care; ESCO, ESRD Seamless Care Organization; ESRD, end-stage renal disease; PBPM, per beneficiary per month; PY, performance year.
Risk-adjusted means and differential changes in Medicare payments, use, and quality measures for the CEC population relative to their pre-CEC period and to the comparison population. The pre-CEC period was from January 2014 to March 2015 for CEC facilities that began participation in October 2015 and from January 2014 to June 2016 for CEC facilities that began participation in January 2017. About 33% of facilities had 9 quarters of CEC participation (October 2015-December 2017); the remaining 67% participated in CEC from January 2017 to December 2017 (4 quarters). Medicare payment outcomes are standardized to remove the effect of geographic and other adjustments. All results were adjusted for patient, market, and facility characteristics that are outside the control of ESCOs. Market and facility variables are representative of the facility to which the beneficiary was assigned based on first-touch assignment.
Percentage change is defined as the difference-in-differences estimate divided by the margins-predicted CEC baseline mean for facilities participating in the CEC model.
Significant at P < .05.
Readmission payments are also included in inpatient payments. To accommodate a longer claim lag needed to identify readmissions after 30 days of discharge, the intervention period for this outcome ended September 2017.
The effect estimates for use were consistent with the effect estimates for payment. Comprehensive ESRD Care beneficiaries experienced −5.0 fewer hospitalizations per 1000 beneficiaries per month (95% CI, −8.5 to −1.6; P = .004), a 4% relative decrease, also associated primarily with wave 1 ESCOs (Table 3). Overall, there was a trend toward fewer emergency department visits among CEC beneficiaries relative to the comparison group, although it was not statistically significant. There was no overall change in hospital readmission rates or emergency dialysis sessions. Outpatient dialysis sessions increased by 71.3 (95% CI, 27.4-115.3; P = .001) per 1000 beneficiaries per month overall, an increase of 0.58%. Results were primarily due to wave 1 ESCOs.
Table 3. Effect Estimates for CEC Use, per 1000 Beneficiaries per Month, by Wave and PYa.
| Measure | CEC facility mean before start of CEC model | Difference-in-differences estimate (95% CI)a,b | Change relative to mean before start of CEC model, % | P value |
|---|---|---|---|---|
| Hospitalizations | ||||
| All ESCOs (PY1 and PY2) | 126.6 | −5.0 (−8.5 to −1.6)c | −4.0 | .004 |
| Wave 1 | ||||
| PY1 | 126.7 | −5.1 (−9.8 to −0.3)c | −4.0 | .04 |
| PY2 | 126.7 | −6.6 (−11.3 to −2.0)c | −5.2 | .005 |
| Wave 2 PY2 | 126.8 | −2.6 (−6.6 to 1.5) | −2.0 | .21 |
| Readmissionsd | ||||
| All ESCOs (PY1 and PY2) | 347.9 | −6.2 (−16.8 to 4.5) | −1.8 | .26 |
| Wave 1 | ||||
| PY1 | 347.9 | −4.0 (−17.3 to 9.3) | −1.2 | .56 |
| PY2 | 347.9 | −8.6 (−24.6 to 7.4) | −2.5 | .29 |
| Wave 2 PY2 | 346.6 | −9.4 (−25.5 to 6.6) | −2.7 | .25 |
| ED visits | ||||
| All ESCOs (PY1 and PY2) | 137.7 | −3.9 (−8.7 to 1.0) | −2.8 | .12 |
| Wave 1 | ||||
| PY1 | 137.8 | −3.7 (−10.3 to 3.0) | −2.7 | .28 |
| PY2 | 137.8 | −6.9 (−13.2 to −0.6)c | −5.0 | .03 |
| Wave 2 PY2 | 140.0 | 0.2 (−5.5 to 6.0) | −0.2 | .94 |
| Emergency dialysis | ||||
| All ESCOs (PY1 and PY2) | 1.8% | −0.01 (−0.2 to 0.1) | −0.7 | .87 |
| Wave 1 | ||||
| PY1 | 1.8% | 0.1 (−0.1 to 0.3) | 5.5 | .38 |
| PY2 | 1.8% | −0.1 (−0.2 to 0.1) | −3.4 | .51 |
| Wave 2 PY2 | 1.8% | −0.3 (−0.4 to −0.1)c | −14.7 | .001 |
| Dialysis sessions | ||||
| All ESCOs (PY1 and PY2) | 12 254 | 71.3 (27.4 to 115.3)c | 0.6 | .001 |
| Wave 1 | ||||
| PY1 | 12 253 | 76.1 (12.3 to 139.9)c | 0.6 | .02 |
| PY2 | 12 253 | 88.0 (38.2 to 137.8)c | 0.7 | .001 |
| Wave 2 PY2 | 12 264 | 37.8 (−4.7 to 80.4) | 0.3 | .08 |
Abbreviations: CEC, Comprehensive ESRD Care; ED, emergency department; ESCO, ESRD Seamless Care Organization; PY, performance year.
Risk-adjusted means and differential changes in Medicare payments, use, and quality measures for the CEC population relative to their pre-CEC period and to the comparison population. The pre-CEC period was from January 2014 to March 2015 for CEC facilities that began participation in October 2015 and from January 2014 to June 2016 for CEC facilities that began participation in January 2017. About 33% of facilities had 9 quarters of CEC participation (from October 2015 to December 2017); the remaining 67% participated in CEC from January 2017 to December 2017 (4 quarters). Medicare payment outcomes are standardized to remove the effect of geographic and other adjustments. All results were adjusted for patient, market, and facility characteristics that are outside the control of ESCOs. Market and facility variables are representative of the facility to which the beneficiary was assigned based on first-touch assignment.
Percentage change is defined as the difference-in-differences estimate divided by the margins-predicted CEC baseline mean for facilities participating in the CEC model.
Significant at P < .05.
To accommodate a longer claim lag needed to identify readmissions after 30 days of discharge, the intervention period for this outcome ended September 2017.
The CEC model was associated with improvements in quality-of-care indicators. Comprehensive ESRD Care beneficiaries with a catheter as vascular access for periods more than 90 days decreased by 0.78 percentage points (95% CI, −1.36 to −0.19 percentage points; P = .01) relative to the comparison group, an 8.3% difference per month (Table 4). Comprehensive ESRD Care beneficiaries were 0.11 percentage points less likely (95% CI, −0.20 to −0.02 percentage points; P = .01) to be hospitalized in a given month owing to an ESRD complication, a 6.4% difference per month. Results were associated primarily with wave 1 ESCOs in PY1. The CEC model was not associated with hospitalizations for vascular access complications.
Table 4. Effect Estimates per Month for CEC Quality, by Wave and PYa.
| Measure | CEC facility mean before start of CEC model, % | Difference-in-differences estimate (95% CI)a,b | Change relative to mean before start of CEC model, % | P value |
|---|---|---|---|---|
| Catheter | ||||
| All ESCOs (PY1 and PY2) | 9.4 | −0.8 (−1.4 to −0.2)c | −8.3 | .01 |
| Wave 1d | ||||
| PY1 | 9.4 | −1.0 (−1.8 to −0.2)c | −10.7 | .01 |
| PY2 | 9.4 | −0.9 (−1.6 to −0.1)c | −9.2 | .02 |
| Wave 2 PY2 | 9.6 | −0.2 (−0.8 to 0.4) | −1.5 | .64 |
| Vascular access complications | ||||
| All ESCOs (PY1 and PY2) | 0.6 | 0.0 (−0.1 to 0.1) | 0.5 | .91 |
| Wave 1d | ||||
| PY1 | 0.6 | −0.02 (−0.1 to 0.1) | −3.1 | .60 |
| PY2 | 0.6 | 0.00 (−0.1 to 0.1) | 0.1 | .99 |
| Wave 2 PY2 | 0.6 | 0.04 (−0.02 to 0.1) | 7.2 | .21 |
| Hospitalizations for ESRD complications | ||||
| All ESCOs (PY1 and PY2) | 1.8 | −0.1 (−0.2 to −0.02)c | −6.4 | .01 |
| Wave 1d | ||||
| PY1 | 1.8 | −0.2 (−0.3 to −0.03)c | −8.3 | .01 |
| PY2 | 1.8 | −0.1 (−0.2 to 0.02) | −6.2 | .09 |
| Wave 2 PY2 | 1.8 | −0.03 (−0.2 to 0.1) | −1.9 | .60 |
Abbreviations: CEC, comprehensive ESRD care; ESCO, ESRD Seamless Care Organization; ESRD, end-stage renal disease; PY, performance year.
Risk-adjusted means and differential changes in Medicare payments, use, and quality measures for the CEC population relative to their pre-CEC period and to the comparison population. The pre-CEC period was from January 2014 to March 2015 for CEC facilities that began participation in October 2015 and from January 2014 to June 2016 for CEC facilities that began participation in January 2017. About 33% of facilities had 9 quarters of CEC participation (from October 2015 to December 2017); the remaining 67% participated in CEC from January 2017 to December 2017 (4 quarters). Medicare payment outcomes are standardized to remove the effect of geographic and other adjustments. All results were adjusted for patient, market, and facility characteristics that are outside the control of ESCOs. Market and facility variables are representative of the facility to which the beneficiary was assigned based on first-touch assignment.
Percentage change is defined as the difference-in-differences estimate divided by the margins-predicted CEC baseline mean for facilities participating in the CEC model.
Significant at P < .05.
Wave 1 results failed parallel trend tests. However, visual inspection of the trend graph that compared trends between the treatment (CEC) and comparison groups yielded no obvious differences.
Discussion
The CEC model is a unique iteration of the Medicare ACO concept that focuses on a particular clinical population (patients with ESRD receiving dialysis) and assigns accountability to specialty health care providers (dialysis facilities, nephrologists, and other providers and suppliers) that deliver care to these patients. This structure allows the CMS to test whether specialty providers can improve the coordination of care for vulnerable, high-cost beneficiaries to achieve better quality at a lower cost.
Overall, the first 2 years of experience under the CEC model are promising, with lower provider payments and improvements in some use and quality outcomes. We observed a nearly 2% reduction in total Medicare payments, associated largely with reductions in payments for inpatient hospitalizations and readmissions. However, shared savings payments were greater than decreases in provider payments, resulting in net losses to the Medicare program. This finding underscores the challenge of calibrating payments to providers that encourage participation in Innovation Center models without offsetting anticipated cost reductions.
Use outcomes paralleled the payment reductions, with significant decreases in hospitalizations. The increase in dialysis payments and treatments imply that CEC beneficiaries were more adherent to dialysis regimens, which could explain the lower number of hospitalizations for complications due to ESRD and the subsequent cost savings. Qualitatively, information collected through structured interviews at site visits confirm the ESCOs’ focus on improving dialysis care through enhanced efforts to avoid or reschedule missed dialysis sessions and improved communication with hospital emergency departments to divert ESCO patients back to their facilities for additional treatment.10 The site visits also indicated that new investments in information technology and hiring care coordinators were nearly universal and that steps taken to engage and coordinate with providers extended beyond care directly related to dialysis (eg, arranging other specialty visits and performing medication reconciliation), but the lack of risk-bearing partners outside of dialysis facilities and nephrologists may have limited the model participants’ ability to effect care more broadly.
In general, changes made by ESCOs that entered the model in wave 1 were associated with the effect estimates, and wave 1 ESCOs’ results improved from PY1 to PY2. Several factors may account for the better performance of wave 1 ESCOs relative to wave 2 ESCOs. First, wave 1 ESCOs had more time to prepare for the model owing to delays in the PY1 start date that created a longer delay between application and start than in wave 2. There also were key differences between wave 1 and wave 2 facilities that may have been associated with performance. Facilities in wave 1 ESCOs had higher historical Medicare payments and higher standardized hospitalization and readmission rates than facilities in wave 2 ESCOs, whereas those joining in wave 2 had lower payments and lower standardized hospitalization and readmission rates than non-CEC facilities (eTable 3 in the Supplement). This finding suggests that the facilities in wave 2 ESCOs may have had less ability to improve on their pre-CEC performance. Furthermore, wave 1 ESCO facilities, as early adopters, may have been more motivated to make changes. Nephrologists in wave 1 facilities, for example, began participation in the CEC model before the implementation of the Quality Payment Program, which was created by the Medicare Access and CHIP [Children's Health Insurance Program] Reauthorization Act of 2015, which was signed into law on April 16, 2015, with an initial performance period in 2017. Under the Quality Payment Program, clinicians are able to earn a 5% incentive payment through participation in qualifying Advanced Alternative Payment Models. Wave 2 facilities may have been aware of the potential incentive payments for participation in an Advanced Alternative Payment Model beginning in 2017, which may have motivated nephrologists participating in wave 2.
Limitations
The findings presented in this study have several limitations. First, the 37 ESCOs are not representative of all Medicare dialysis providers, limiting our ability to generalize the results. Second, although the analysis used matching methods to select an appropriate comparison group to infer counterfactual outcomes, the characteristics selected for matching and the specificity of the data may not adequately account for all differences between CEC and comparison facilities. Third, while the data establish net losses in total Medicare payments, they do not allow us to calculate costs to providers accounting for implementation costs. Fourth, the current analysis does not compare outcomes for patient with ESRD in the CEC with those in other ACO programs, an area in which further research would be of interest to policy makers.
Patients with ESRD receiving dialysis are a particularly appropriate target for specialty-oriented ACOs because in-center patients receiving hemodialysis have regular contact with dialysis facilities (3 times weekly) and nephrologists (3-4 times monthly). Patients receiving dialysis at home also have regular, albeit less frequent, contacts with these providers. Therefore, these positive outcomes for the CEC model might not be directly generalizable to populations with other chronic illnesses (eg, diabetes, HIV, or congestive heart failure). Nonetheless, the CEC experience could provide lessons about the potential benefits of specialty providers increasing their responsibilities in an ACO context, whether that ACO is composed entirely of a population with a particular chronic condition or represents a subpopulation within an ACO.
The burden of chronic kidney disease and ESRD remains very high for Medicare beneficiaries. This fact was recently recognized in President Trump’s Executive Order on Advancing American Kidney Health,11 which calls for future models to further enhance the care of beneficiaries with ESRD and reach beneficiaries currently excluded from the CEC (eg, beneficiaries with advanced kidney disease who are not yet receiving dialysis and those who receive kidney transplants). These new models, which include better alignment of financial incentives to delay the onset of dialysis and increase uptake of home dialysis and transplants, build on the CEC model and were informed by this analysis.
Conclusions
During the first 2 performance years of the CEC, Medicare payments decreased by 2% while quality improved, which suggests that a specialty-centered ACO model can deliver more efficient care to a clinically complex and vulnerable population. The data suggest that ensuring adequate dialysis, thereby reducing hospitalizations, was a likely mechanism. Further analysis is required to assess the longer-term outcomes of the CEC model and to consider the applicability of these results to populations with other complex chronic conditions.
eAppendix. Facilities Eligible to be Included in the Comparison Group Pool
eFigure 1. Treatment and Comparison Facility Identification and Exclusions
eFigure 2. Log Odds Propensity Score Density of CEC and Comparison Facilities
eFigure 3. Sensitivity Analysis
eTable 1. Risk-Adjustment Covariates Used in Claims Analysis
eTable 2. Baseline Parallel Trend Tests
eTable 3. Characteristics of Matched CEC and Non-CEC Facilities 2012-2014
eTable 4. Baseline Characteristics of CEC and Non-CEC Beneficiaries (January 2014-March 2015)
eTable 5. Reason for De-alignment of Beneficiaries by CEC Participation
eTable 6. De-aligned Beneficiary Characteristics by CEC Participation
References
- 1.Weisbord SD, Mor MK, Sevick MA, et al. Associations of depressive symptoms and pain with dialysis adherence, health resource utilization, and mortality in patients receiving chronic hemodialysis. Clin J Am Soc Nephrol. 2014;9(9):1594-1602. doi: 10.2215/CJN.00220114 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Knauf F, Aronson PS. ESRD as a window into America’s cost crisis in health care. J Am Soc Nephrol. 2009;20(10):2093-2097. doi: 10.1681/ASN.2009070715 [DOI] [PubMed] [Google Scholar]
- 3.Abbasi MA, Chertow GM, Hall YN. End-stage renal disease. BMJ Clin Evid. 2010;2010:2002. [PMC free article] [PubMed] [Google Scholar]
- 4.Foster BJ, Mitsnefes MM, Dahhou M, Zhang X, Laskin BL. Changes in excess mortality from end stage renal disease in the United States from 1995 to 2013. Clin J Am Soc Nephrol. 2018;13(1):91-99. doi: 10.2215/CJN.04330417 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.US Renal Data Systems USRDS 2006 Annual Data Report: Atlas of End-Stage Renal Disease in the United States. Bethesda, MD: National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases; 2018. [Google Scholar]
- 6.Centers for Medicare & Medicaid Services, Center for Medicare and Medicaid Innovation Comprehensive ESRD Care model: request for applications. Accessed January 1, 2019. https://innovation.cms.gov/Files/x/cec-py2-rfa.pdf
- 7.Centers for Medicare & Medicaid Services Definition of attribution-eligible beneficiary for the Comprehensive ESRD Care (CEC) model under the Quality Payment Program. Accessed January 1, 2019. https://innovation.cms.gov/Files/fact-sheet/cec-attributionfs.pdf
- 8.National Kidney Foundation 2006. Updates: clinical practice guidelines and recommendations: hemodialysis adequacy, peritoneal dialysis adequacy, vascular access. Accessed January 1, 2019. https://www.kidney.org/sites/default/files/docs/12-50-0210_jag_dcp_guidelines-hd_oct06_sectiona_ofc.pdf
- 9.Cameron AC, Gelbach JB, Miller DL. Robust inference with multiway clustering. J Bus Econ Stat. 2011;29(2):238-249. doi: 10.1198/jbes.2010.07136 [DOI] [Google Scholar]
- 10.Centers for Medicare & Medicaid Services Comprehensive End-Stage Renal Disease Care (CEC) model: performance year 1 annual evaluation report. Accessed January 1, 2019. https://innovation.cms.gov/Files/reports/cec-annrpt-py1.pdf
- 11.Advancing American kidney health: Executive Order 13879 of July 10, 2019. Fed Regist. 2019;84:33817-33819. [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
eAppendix. Facilities Eligible to be Included in the Comparison Group Pool
eFigure 1. Treatment and Comparison Facility Identification and Exclusions
eFigure 2. Log Odds Propensity Score Density of CEC and Comparison Facilities
eFigure 3. Sensitivity Analysis
eTable 1. Risk-Adjustment Covariates Used in Claims Analysis
eTable 2. Baseline Parallel Trend Tests
eTable 3. Characteristics of Matched CEC and Non-CEC Facilities 2012-2014
eTable 4. Baseline Characteristics of CEC and Non-CEC Beneficiaries (January 2014-March 2015)
eTable 5. Reason for De-alignment of Beneficiaries by CEC Participation
eTable 6. De-aligned Beneficiary Characteristics by CEC Participation
