Financial Incentives to Facilities and Clinicians Treating Patients With End-stage Kidney Disease and Use of Home Dialysis

Importance Home dialysis rates for end-stage kidney disease (ESKD) treatment are substantially lower in the US than in other high-income countries, yet there is limited knowledge on how to increase these rates. Objective To report results from the first year of a nationwide randomized clinical trial that provides financial incentives to ESKD facilities and managing clinicians to increase home dialysis rates. Design, Setting, and Participants Results were analyzed from the first year of the End-Stage Renal Disease Treatment Choice (ETC) model, a multiyear, mandatory-participation randomized clinical trial designed and implemented by the US Center for Medicare & Medicaid Innovation. Data were reported on Medicare patients with ESKD 66 years or older who initiated treatment with dialysis in 2021, with data collection through December 31, 2021; the study included all eligible ESKD facilities and managing clinicians. Eligible hospital referral regions (HRRs) were randomly assigned to the ETC (91 HRRs) or a control group (211 HRRs). Interventions The ESKD facilities and managing clinicians received financial incentives for home dialysis use. Main Outcomes and Measures The primary outcome was the percentage of patients with ESKD who received any home dialysis during the first 90 days of treatment. Secondary outcomes included other measures of home dialysis and patient volume and characteristics. Results Among the 302 HRRs eligible for randomization, 18 621 eligible patients initiated dialysis treatment during the study period (mean [SD] age, 74.8 [1.05] years; 7856 women [42.1%]; 10 765 men [57.9%]; 859 Asian [5.2%], 3280 [17.7%] Black, 730 [4.3%] Hispanic, 239 North American Native, and 12 394 managing clinicians. The mean (SD) share of patients with any home dialysis during the first 90 days was 20.6% (7.8%) in the control group and was 0.12 percentage points higher (95% CI, −1.42 to 1.65 percentage points; P = .88) in the ETC group, a statistically nonsignificant difference. None of the secondary outcomes differed significantly between groups. Conclusions and Relevance The trial results found that in the first year of the US Center for Medicare & Medicaid Innovation–designed ETC model, HRRs assigned to the model did not have statistically significantly different rates in home dialysis compared with control HRRs. This raises questions about the efficacy of the financial incentives provided, although further evaluation is needed, as the size of these incentives will increase in subsequent years. Trial Registration ClinicalTrials.gov Identifier: NCT05005572


Introduction 24
Treatment for end stage renal disease (ESRD) requires a large amount of health care resources: the ESRD 25 population accounts for 7% of Medicare fee-for-service spending despite making up less than 1% of total 26 Medicare enrollment (USRDS 2020). The most common treatment for ESRD is dialysis, for which there 27 are two options: home dialysis (which usually takes the form of peritoneal dialysis) and facility dialysis 28 (which is usually hemodialysis). Home dialysis is often associated with lower spending and better 29 outcomes than facility-based dialysis (GAO 2015). However, only 12% of patients receive dialysis at 30 home in the US, significantly below that of other developed countries (USRDS 2020). 31 We propose to study the ESRD Treatment Choice (ETC) program, a nationwide randomized-controlled 32 trial (RCT) of ESRD treatment that was designed by the Centers for Medicare and Medicaid Services 33 (CMS) to encourage greater use of home dialysis. The RCT creates payment adjustments for ESRD 34 facilities and clinicians based on the rates of home dialysis and transplantation. The RCT was first 35 announced in July 2019 for a start date of January 1, 2020 which was later delayed to January 1, 2021. In 36 September 2020, 30% of hospital referral regions (HRRs) were assigned to treatment through a stratified 37 randomization procedure. The RCT is scheduled to last until June 2027. 38 This analysis plan is for the first year of the program, which covers calendar year 2021. 2 We will examine 39 the impact of payment incentives for home dialysis on home dialysis use among ESRD patients. We 40 examine both the average impacts, as well as heterogeneity across provider types and patient 41 characteristics. 42

Experimental Design 43
The ETC program is a nation-wide, HRR-level, mandatory RCT. All Medicare-certified ESRD facilities and 44 Medicare-enrolled managing clinicians in the selected HRRs are required to participate. In this section, 45 we summarize the details of the program and describe our study sample. 46

Program Incentives 47
The ETC model makes two types of payment adjustments for facilities and managing clinicians, an 48 adjustment to the reimbursement rate for home dialysis, and a performance adjustment. Although they 49 are implemented together and we analyze them jointly, we expect the performance payment 50 adjustment to have a bigger effect given the greater downside and upside risks. 51 The first adjustment, Home Dialysis Payment Adjustment (HDPA), raises the reimbursement rate for 52 home dialysis for the first three years of the program; the amount of increase is 3% in 2021, and reduces 53 to 2% in 2022, and 1% in 2023 (CMS 2020). 54 The second adjustment, Performance Payment Adjustment (PPA), is an increase or decrease in the 55 reimbursement rate based on the home dialysis rate and the transplant rate attributable to the 56 participating facility or clinician. The PPA, unlike the HDPA, has separate measurement years (during 57 which CMS assess the performance of each participant) and performance payment adjustment periods 58 (during which payments are adjusted based on the performance in the corresponding measurement  59 year). The first measurement year is the calendar year 2021, which affects payments made to providers 60 in the period between July and December of 2022. 3 The magnitude of the adjustment depends on the 61 modality performance score (MPS Providers with an MPS between 2 (exclusive) and 3.5 (inclusive) do not receive a payment adjustment; 77 providers with an MPS below 2 (inclusive) receive a negative payment adjustment and providers with an 78 MPS above 3.5 (exclusive) receive a positive payment adjustment. The magnitude of the adjustment 79 increases over time, from -5% to +4% in the first year to -10% to +8% in the last year (CMS 2020). 80 In our analysis, we will focus on home dialysis rates rather than transplant rates for several reasons. First home. We therefore expect this substitution effect to increase the number of patients in home dialysis 99 and decrease the number of patients in facility dialysis. Second, increased payments in HDPA and 100 (conditional on performance) PPA provide additional income to the providers, which can have an impact 101 on both the quantity and quality of all the services they provide, regardless of the treatment modality. 102 For example, the additional income could affect the number of dialysis stations the facility acquires, the 103 number of clinicians it hires, and the amount of time devoted to each patient, among many other things. 104 Conversely, reduced revenue from poor performance on PPA can also have an impact on all patients 105 through similar channels. We have no a priori hypothesis of the sign of this income effect on home 106 dialysis rates. 107 The computation of improvement scores, which compares one's performance during the measurement 108 year with that in a benchmark year that is 18 months prior, could result in ratchet effects. With the 109 exception of the first calendar year (2021) of the program, all subsequent years use at least part of a 110 previous performance measurement year as a benchmark; such design can dampen the incentive to 111 improve quickly since doing so will make the target harder to meet in later years. In the period between 112 the announcement of the program in 2019 and its final implementation in 2021, there was an additional 113 incentive for providers to lower their home dialysis and transplant rates as they can do so without 114 incurring any penalty while achieving a favorable benchmark rate in anticipation of the program launch. 115 To examine such effects, we analyze home dialysis rate in 2020 as one of our outcomes. 116 Finally, providers have the incentive to "cream skim" or attract patients who are more suitable for home 117 dialysis (as opposed to facility dialysis) to improve their home dialysis rates. These incentives are 118 exacerbated by the fact that currently there is no risk adjustment mechanism when computing home 119 dialysis rates. 120 121 2.3 Randomization 122 Any HRR with 20% or more zip codes in Maryland are excluded from the randomization and directly 123 assigned to the program. 30% of the remaining HRRs are randomly selected to participate in the 124 program, through a stratified randomization by Census-defined regions (Northeast, South, Midwest, and 125 West) (CMS 2020). 126

Program Eligibility and Study Sample 127
Our baseline sample includes all Medicare-certified ESRD facilities and Medicare-enrolled managing 128 clinicians located in HRRs that are eligible for randomization. Following the program rules (CMS 2020), 129 we make the following sample restrictions. In 2016 7 , a total of 394,323 Medicare beneficiaries with 130 Medicare as their primary insurance received dialysis. For each group of restrictions, we report the 131 number of beneficiaries excluded from our sample. 132 We begin by imposing three restrictions that the ETC program imposes: 133 beneficiaries) 159 7 Our analysis sample will be based on patients in 2021, but for purposes of the pre-analysis plan, we have implemented our sample definitions on the 2016 patients. 8 Maryland is assigned by CMS to participate in ETC. However, since the assignment is not based on randomization, we exclude Maryland from our analysis. 9 Since we only have a 20% Medicare carrier file at the moment, we do not have an estimate for how many patients are excluded from these criteria. 10 The restrictions marked with * only apply to the calculation of PPA based on the official rules. (CMS 2020) However, since HDPA and PPA are introduced at the same time and we expect most of the effect to come from PPA due to the stronger financial incentives, we restrict our baseline sample to the set of patients meeting the requirements for both adjustments. We may explore heterogeneity by looking separately at patients included in HDPA but not PPA. 11 The official rule applies on a month-by-month basis to all dialysis patient-months. However, since we are only looking at 90-day outcomes for new patients we modify these to apply to the first 90 days of dialysis treatment. This analysis plan is for the first year of the program, which begins on Jan 1, 2021, and ends on Dec 31, 171 2021. We will conduct our baseline analysis using the 100% Medicare claims data aggregated to the HRR 172 level. Our main regression specification examines the direct impact of the ETC program on rates of home 173 dialysis rates. 174 Let denote HRRs and let denote calendar years. Let denote the average outcome in HRR in the 175 first year of the program. Let be a binary variable that takes on a value of 1 for the HRRs assigned to 176 treatment and 0 for the HRRs assigned to control. Our baseline reduced form specification is given by 177 The coefficient of interest is . , ( ) denotes strata fixed effects; we control for strata fixed effects 179 because, as described above, randomization was performed within strata (specifically, within four 180 census regions). , denotes lagged outcome from prior to the ETC program, which helps improve 181 statistical power; specifically, we control for outcomes from three years prior to the first program year 182 (i.e. 2018) to avoid anticipatory effects from biasing our estimates, as the ETC program was announced 183 two years prior to its launch. denotes HRR-level covariates, which are also designed to improve 184 power; these consist of average demographics and baseline health among patients in our baseline 185 sample. Specifically, the demographic variables include indicators for fully interacted age-race-sex bins. 12 186 Baseline health includes separate indicators for the 31 Elixhauser chronic conditions measured in the 12 187 months prior to dialysis treatment. All measures are averaged at the HRR level. As shown in Appendix 188 and above who start dialysis in either modality in our baseline sample divided by the number of 218 Traditional Medicare patients aged 66 and above. 219 13 We count the share of weeks in home dialysis rather than the share of claims or treatment sessions because hemodialysis and peritoneal dialysis both require sessions each week but differ in the number of sessions required. 14 We identify sessions by revenue code lines. From Chapter 8 of the Medicare Claims Processing Manual, "effective April 1, 2007, the implementation of ESRD line item billing requires that each dialysis session be billed on a separate line." The 3/7 adjustment reflects the difference in average treatment frequencies between the modalities and is the ratio used in setting Medicare reimbursement rates. See Section 80.4 of the Medicare Claims Processing Manual. Specifically, we count the number dialysis sessions recorded in the revenue file and label sessions as hemodialysis or peritoneal dialysis based on the revenue center code. 15 To examine the causal impact of home dialysis on downstream outcomes, we could use the random assignment to treatment as an instrument for home dialysis. In this instrumental variable specification, the minimum detectible effect size (MDE) under homoskedasticity with a power of 80% at a significance level of 5% is given by 2.8 √ where is the standard deviation of the outcome in the sample, is the R-squared of the first stage, and is the sample size. Compared with the MDE in an OLS setting (2.8 √ ), the effect size is scaled by √ under the instrumental variable specification. We have estimated that =0.0017, which makes detecting changes in downstream outcomes challenging. Specifically, for 90-day hospitalizations we have computed a MDE of 41 percentage points (relative to a mean of 22.6%), and for 90-day mortality rate a MDE of 8 percentage points (relative to a mean of 6.9%). Based on these numbers, we believe we lack the power to detect meaningful changes in downstream outcomes. and about one third is non-white. 238 In addition, we will examine the heterogeneous impact on patients of different baseline propensities to 239 use home dialysis. Specifically, following the method laid out in Einav et al. (2020), we will define , the 240 predicted probability of home dialysis use, through a machine learning procedure using data on patient 241 demographics, distance to dialysis facilities, and health status prior to the program. One can interpret 242 this predicted probability of home dialysis use as a compliance propensity score (Follmann 2000). We 243 will document the key demographic, clinical, and other predictors of home dialysis use. 244 We will then use this predicted probability to divide our baseline sample into subgroups of different 245 propensities for home dialysis, and explore how the ETC program differentially affects those with high vs 246 low predicted probabilities. We will also consider a variant of this predicted probability by incorporating 247 each patient's distance to the nearest dialysis facility into our prediction model. 248 To understand heterogeneity in impact at the provider level, we will conduct the analysis in We will examine robustness of our results to alternative specifications. Specifically, we will explore 259 robustness of our results to different controls by estimating versions of equation (1) without controlling  260 for the lagged dependent variable and/or patient demographic and pre-dialysis health. We will also 261 explore robustness of our results to alternative time horizons by estimating equation (1) (1) using 2016 Medicare claims data (and lags from 2013). The column "MDE" references the minimum detectable effect size, and "% Control Mean" represents the ratio of MDE over Control Mean. Effect sizes and MDE calculated for an alpha of 0.05 and power of 0.8. Control standard deviations are calculated from residuals of a weighted least squares regression weighting by the average number of new patients in an HRR over the past two years with controls of indicators for Census Region Strata, average patient demographics, pre-dialysis health, and the lagged outcome from 2013. All outcomes are measured as 90-day outcomes, measuring from the first day of dialysis to 89 days afterwards. The sample includes all patients who meet the sample criteria: they are at least 66 years old on the first day of the first month of their dialysis treatment, start dialysis prior to 90 days before the end of 2016, and they are a "new patient" with no dialysis claims for 12 months prior. Please note that the power calculations -unlike our planned analysis -does not include any carrier claims in defining dialysis or previous dialysis use since we only have carrier claims for 20% of Medicare beneficiaries and dialysis claims are almost always recorded in the outpatient file. Sample sizes are N=211 HRRs for control, and N=91 HRRs for treatment, except where noted. *Sample sizes are N=206 HRRs for control, and N=90 HRRs for treatment. **Sample sizes are N=197 HRRs for control, and N=86 HRRs for treatment.
These strike us as promising MDEs. An increase of 1.98 percentage points or larger for home dialysis 273 seems reasonable to expect from the RCT, given the current low levels of home dialysis use in our study 274 sample (15.6%) relative to its potential. Studies have estimated that about 25-40% of patients would 275 choose home dialysis if presented as an option, and 85% of patients with advanced chronic kidney 276 disease are eligible for home dialysis. However, only one third of patients are currently informed of 277 peritoneal (home) dialysis as an option (Rivara and Mehrotra 2014). It also seems a reasonable effect 278 size given that home dialysis rates are significantly higher in many other countries and regions, such as 279 Hong Kong (74%), Mexico (61%), Guatemala (57%), and New Zealand (47%) (USRDS 2018). 280 281 282

Conclusion 283
This is an analysis plan of the impact of the first year of a nation-wide randomized controlled trial on 284 home dialysis use. Our analysis will help inform the design of health care payment policies for ESRD. 285 286