Medicare advantage and dialysis facility choice

2.1 Data and study population

We used several data sources. First, we used 20% fee-for-service (FFS) outpatient claims and 100% MA outpatient encounter data from 2018 to identify dialysis visits by ESKD beneficiaries in TM and MA, respectively. Although there are concerns about the completeness of MA encounter data, past research has demonstrated that the quality of dialysis encounter data is relatively high.¹¹ We also used the 100% Medicare Beneficiary Summary File to obtain beneficiary characteristics and enrollment information. Second, we obtained quality and ownership information on dialysis facilities for the performance periods ending December 31, 2018 from Dialysis Facility Compare. We also used data from Dialysis Facility Reports, combined with information derived from FFS claims, to identify treating facilities based on the National Provider Identifier (NPI) listed on MA encounters. Finally, we used publicly reported data on MA plans and contracts, including the plan type, the contract quality rating, and an indicator for whether the plan was an ESKD SNP.¹² We linked this plan and contract data to the plan and contract of MA patients as of July 2018.

We used outpatient dialysis claims and encounters (bill type 72) that began before and ended after July 15, 2018 to identify point prevalent MA and TM dialysis patients. We attributed each patient to one treating facility and excluded patients with encounters or claims without a corresponding dialysis facility in Dialysis Facility Compare (N = 2264). We excluded patients treated with dialysis for acute kidney injury rather than ESKD (N = 524), patients outside of the 50 states and the District of Columbia (N = 1968), patients with both a TM claim and MA encounter (N = 45), and those not in the master beneficiary summary file (N < 11). In each regression, we exclude those with insufficient data on the outcome variables of interest. Our main sample consisted of 142,050 patients (weighted N = 385,034).

2.2 Outcomes

The outcomes of interest were the ownership and quality of the treating facility. First, we used data on facility ownership as of December 31, 2018 to classify facilities by ownership. We classified facilities as owned by Fresenius, DaVita, other dialysis organizations, or independent organizations. We considered facilities as owned by other dialysis organizations if they were owned by American Renal Associates, Dialysis Clinics, Inc., or US Renal Care, Inc.

Second, distance traveled is an important measure of quality and access because patients often travel to a dialysis facility three times a week for treatment. Even for home dialysis patients who often go to a facility monthly, distance remains an important consideration when selecting a facility.¹³ We calculated the distance, in miles, between a dialysis facility and the centroid of the patients' ZIP code tabulation area using the Haversine formula. Patients with over 100 miles to their treating facility were coded as missing because this may indicate the patient ZIP code is inaccurate or the patient is away from home.

Finally, we used four Centers for Medicare and Medicaid Services (CMS) publicly-reported measures of facility quality to ensure that our results are robust to different definitions of quality. The measures are: risk-adjusted mortality rate, five-star rating, end-stage renal disease quality incentive program (QIP) total performance core (TPS), and an indicator for whether the facility was penalized under the QIP. The facility mortality rate is case-mix adjusted for age, sex, race, ethnicity, cause of ESKD, duration of ESKD, body mass index, and comorbidities. The resulting variable can be interpreted as the number of deaths per 100 patient-years at risk.¹⁴ Although this rate may be affected by unobserved differences in facilities or random variation, the high mortality rate among dialysis patients and the difficulty of “gaming” mortality statistics makes it a useful measure of the quality of care. The five-star rating is a patient care rating of a dialysis facility relative to others. The rating is a summary of quality care measures such as avoiding death, hospitalizations, and hospital readmissions. The rating ranges from 1 to 5 stars with 3-stars as the average. A 5-star rating means a facility is “much above average” compared with others and a 1 or 2-star rating means a facility is below average. The QIP TPS is a composite of several quality measures, including dialysis adequacy, vascular access, and measures of health care utilization. Facility scores range from 0 to 100 and are calculated based on achievement (facility performance relative to prospectively set benchmarks) or improvement (facility performance relative to own past performance). QIP TPS is used to determine reductions of up to 2% of TM payments.¹⁴ Over 40% of facilities were penalized based on 2018 data.¹⁵

2.3 Statistical analysis

We first present summary characteristics of TM and MA patients. We then estimated ordinary least squares regressions that compared the quality and ownership of the treating dialysis facilities of MA and TM patients living in the same ZIP codes. Patients living in the same ZIP code likely face the same set of potential facility choices, regardless of their MA/TM status. We captured this in our model with patient ZIP code fixed effects. Among those with the same set of nearby facilities, differences in TM and MA facility choice may reflect MA plan design factors like networks. We accounted for differences in patient characteristics by adjusting for dual eligibility status, race and ethnicity, sex, original reason for Medicare entitlement, age, and age squared in all regressions. We do not adjust for dialysis modality because it is endogenous to facility choice.

Because MA carriers may pursue different network formation strategies, we also examined differences between the ownership of facilities treating TM and MA patients stratified by MA carrier. We used publicly reported MA parent organization to classify plans as operated by United Healthcare, Humana, Aetna, Blue Cross Blue Shield, Kaiser, or a parent organization that is not one of the largest five carriers, as determined by overall enrollment. In addition to carriers with Blue Cross Blue Shield in the parent organization name, we also considered Anthem, Highmark, and Premera to be Blue Cross Blue Shield plans. In addition to our primary analyses of differences in facility quality, we also considered whether facility quality differs between TM and different types of MA plans, stratified by large carrier status and plan type (Health Maintenance Organization (HMO) or Preferred Provider Organization (PPO)). We considered each of these stratifications by plan characteristic in separate regressions.

Because we used a 20% sample of FFS claims and 100% of MA encounters, we weighted TM and MA patients by the inverse probability of being in our sample. This research was determined to be exempt by the Johns Hopkins Bloomberg School of Public Health Institutional Review Board.

3.1 Sample description

Our sample included 142,050 dialysis patients (weighted N = 385,034), with 60,746 enrolled in TM (weighted N = 303,730) and the remaining 81,304 enrolled in MA (weighted N = 81,304) (see Table 1). On average, MA patients (mean age = 70.0) were older than TM patients (mean age = 63.3). TM patients were more likely to have originally enrolled in Medicare because of ESKD or ESKD and disability. These differences in age and enrollment likely reflect the policies restricting MA enrollment for ESKD patients during the study period. MA patients were also less likely than TM patients to be dually eligible for Medicaid and less likely to use peritoneal dialysis.

3.2 Differences in dialysis facility ownership

Compared to TM patients living in the same ZIP code, MA patients were 1.84 percentage points (95% CI: 1.34, 2.33) more likely to be treated at a facility owned by Fresenius or DaVita. The probability that an MA patient was treated at an independently owned facility was 1.85 percentage point lower (95% CI: −2.26, −1.44) than TM patients in the same ZIP code. There was no difference in the probability of being treated at a facility owned by other dialysis organizations (see Figure 1).

These aggregate results, however, mask heterogeneity across MA carriers, which we investigate in Table 2, while also considering each large dialysis organization separately. For example, compared to TM patients in the same ZIP code, MA patients in United Healthcare plans were 1.85 (95% CI: 1.07, 2.63) and 0.73 (95% CI: 0.24, 1.22) percentage points more likely to be treated at Fresenius and other dialysis organization facilities, respectively, but 1.00 (95% CI: −1.78, −0.22) and 1.58 (95% CI: −2.14, −1.03) percentage points less likely to be treated at DaVita and independent facilities, respectively. Patients in Humana and Blue Cross Blue Shield MA plans were more likely to be treated at DaVita facilities and less likely to be treated at Fresenius facilities. The magnitude of estimates for Kaiser plans was particularly large. Enrollees in Kaiser MA plans were 15.82 (95% CI: 14.43, 17.21) percentage points more likely to visit a Fresenius facility and less likely to visit facilities owned by other providers. Overall, these results suggest that each of the five largest health insurance carriers may steer patients toward facilities owned by one of the two largest dialysis organizations and away from independently owned facilities. There may also be additional heterogeneity within carrier, such as differences across plans that our analysis was not able to capture.

3.3 Differences in dialysis facility distance and quality

On average, MA patients were treated at facilities 0.15 miles (95% CI: 0.06, 0.25) farther than TM patients living in the same ZIP code (Table 3). This estimate is modest given that the average distance was 7.79 miles. Across the four measures of facility quality, MA patients were treated at worse-quality dialysis facilities compared to TM patients living in the same ZIP code. The magnitude of these results depends on the dialysis facility quality measure used. The mortality rate estimate of 0.67 deaths per 100 patient-years (95% CI: 0.61, 0.73) implies a 3.1% difference relative to the mean of 21.85. The QIP total performance score estimate of 0.15 (95% CI: −0.28, −0.01) implies a 0.3% difference on the mean of 59.85.

In Table 4, we investigate heterogeneity in quality differences by plan characteristics. First, patients in MA plans not owned by the largest five carriers traveled 0.61 miles further than TM patients in the same ZIP code (95% CI: 0.46, 0.75). There was no difference in facility distance between TM and MA patients in large carrier plans in the same ZIP code. The fact that differences in distance traveled by MA and TM enrollees for dialysis is evident only for plans not owned by the largest five carriers may reflect the stronger bargaining position of the largest five firms. Here, the point estimate of 0.61 would imply that MA patients in these plans going to dialysis 3 days a week would travel 3.66 miles further per week than those in TM. We also find that MA patients went to lower-quality facilities than TM patients regardless of carrier size. While the difference in facility mortality rate was higher for large carriers, this difference was not robust across all measures of dialysis facility quality (see Table 4 panel A).

Second, HMO patients traveled 0.23 miles further to dialysis than TM patients in the same ZIP code, while there was no statistically significant difference for PPO patients. This may reflect the fact that HMOs often have more restrictive networks. Differences in QIP TPS and mortality were higher for HMO than PPO patients, though this difference was not observed when examining facility five-star rating (see Table 4 panel B). In Table S1 of the supplemental materials, we examine heterogeneity by MA plan quality and ESKD SNP status. We found no consistent differences in facility distance or quality across plan types.

3.4 Sensitivity analyses

Studies comparing TM and MA patients are often limited by unobserved selection by risk. This is particularly true of studies on health outcomes, where unobserved factors correlated with MA enrollment may also be correlated with health status. Unlike health outcomes, facility choices may be less likely to be affected by this issue. Patients living in an area must choose from the same set of facilities. MA plans may influence facility choice directly via network design or indirectly through care management practices or the referral patterns of in-network nephrologists. Differences in risk across MA and TM would affect facility choice if patient preferences vary by risk or if facilities were able to select patients based on risk. We tested the robustness of our main results by including an adjustment for patient risk profile. The extent of upcoding in MA makes obtaining prevalent comorbidities from claims and encounter data unreliable.^{16, 17} Instead, because dialysis patients have high risk of hospitalization, we used a measure of hospitalization as a proxy for the underlying health status of a patient. Hospitalization is an imperfect proxy for risk because it is potentially modifiable by MA plans, but it is informative because it is not affected by coding practices in MA. We used two data sources to calculate the number of hospitalizations per month at risk (i.e., months with a dialysis claim or encounter) from January through July 2018. First, we used Medicare Provider and Review (MedPAR) data, which contains information on both TM and MA patients and is often used to compare hospital utilization between TM and MA.¹⁸ However, a small number of hospitals may not report MA discharges so the data may not be complete.¹⁹ Second, we used claims and encounter data. There are concerns that inpatient encounter data may not be complete. Past research has demonstrated some inconsistency between hospitalizations reported in encounter data and those reported in MedPAR.¹¹ Although each of these methods has its limitations, together they provide useful sensitivity analyses to test the robustness of our results to adjustment for different measures of patient risk. We find that across ownership, distance, and facility quality outcomes, the main results are stable with additional adjustments for patient hospitalization rates (see Table S2 of Supplemental Materials).

3.5 Subgroup analyses

We also conducted subgroup analyses stratifying by modality, urbanicity, and market concentration. Estimates were generally similar across these subgroups. However, differences in DaVita and Fresenius ownership appeared to be larger in urban and less concentrated markets. Similarly, differences in driving distance were statistically significant in urban areas but not in rural areas. MA plans may be less able to engage in steering in rural and more concentrated markets, where there are fewer available choices (see Table S3 of Supplemental Materials).