Validation of Medicare Advantage Claims for Long-Term Outcome Assessment in Low-Risk Aortic Valve Replacement.
Mar 29, 2025DOI: 10.1161/circoutcomes.125.011991Journal ArticleAHA
Written by: Christina LalaniFrank MedinaAndrew S OseranLichen LiangYang SongNeel M ButalaDhruv S KaziDavid J CohenJordan B StromRishi K WadheraRobert W Yeh
Find these authors on Roon
AO
Dr. Andrew Oseran
Advanced Heart Failure · BIDMC
DC
Dr. David Cohen
Interventional & Structural Cards · Cardiovascular Research Foundation
INTRODUCTION
Since the establishment of the Medicare Advantage (MA) program in 1997, there has been a marked increase in the number of individuals enrolled in MA plans versus traditional fee-for-service (FFS) Medicare. As of 2024, ~54% of the eligible Medicare population were enrolled in a MA plan.1 In contrast to FFS Medicare, MA plans are administered by private insurers. MA insurers receive a capitated, risk-adjusted payment from Medicare for providing care to each beneficiary and create a cap on beneficiaries’ out-of-pocket spending. Consequently, MA plans often have measures in place to control spending, such as preferred provider networks and prior authorizations.2 In addition, the capitated nature of MA may incentivize plans to code for more diagnoses in order to receive higher risk-adjusted payments.3,4 Unlike in FFS, MA plans can perform “chart reviews” separate from patient encounters to adjust the diagnosis codes associated with a patient, often leading to higher reimbursement.5 There is ongoing debate regarding whether these differences in payment structure between MA and FFS result in differences in healthcare resource use, high-value care delivery, and patient outcomes.2,6In FFS patients, Medicare claims that have been validated against adjudicated trial endpoints are often used to assess long-term outcomes in medical device post-approval studies.7–14 Although MA claims data are already being used by industry for medical device post-market surveillance, there is limited evidence evaluating the quality and validity of MA claims. As the shift toward privatization of Medicare continues, and MA plans become the dominant source of coverage, it will be critically important to evaluate if existing claims algorithms derived from Medicare FFS data perform similarly in MA to support the assessment of long-term clinical outcomes. Therefore, in this study, we sought to compare claims-based covariates and outcomes to “gold-standard” adjudicated trial variables among MA and FFS patients who were enrolled in the Evolut Low-Risk Transcatheter Aortic Valve Replacement (TAVR) Trial.15 We selected this trial to validate MA outcomes because, although TAVR has been FDA-approved for the treatment of severe aortic stenosis in low-risk patients since 2019, efforts to collect long-term safety and effectiveness data in low-risk TAVR are ongoing.16,17 In addition, the majority of patients undergoing aortic valve replacement likely qualify for Medicare, making them a suitable population in which to evaluate claims-based outcomes.
METHODS
The data that support the findings of this study are available from the corresponding author upon reasonable request. The Evolut Low-Risk trial randomized 1468 low-risk patients to undergo self-expanding TAVR vs. SAVR for the treatment of severe aortic stenosis between March 28, 2016 and November 27, 2018.15 Out of the 1139 U.S. patients ≥ 65 years old who were enrolled in the Evolut Low-Risk trial, only those patients who were successfully linked to complete Centers for Medicare & Medicaid Services (CMS) claims data were eligible for inclusion in the study. FFS and MA 100% inpatient claim files were used to identify the trial participants. Deterministic linkage, including sex, hospital identification, aortic valve replacement type, age, procedure date, and discharge date, was used to link patients to both Medicare FFS and MA claims data (Table S1). Patients were excluded if their comorbidity data were missing from claims at the month and year of their procedure. Patients were also excluded if their insurance status captured in the master beneficiary summary files (MBSF) at the month of the TAVR procedure was inconsistent with the inpatient claims source (MA vs. FFS) from which the aortic valve replacement procedures were identified. Medicare program type (MA vs. FFS) was determined based on enrollment status at the time of TAVR, and patients were followed regardless of whether they switched insurance during the follow-up period. Both FFS and MA claims were employed in searching for the covariates and outcomes. This study was approved by the Institutional Review Board at Beth Israel Deaconess Medical Center.
Covariates of interest included baseline rates of hypertension, prior myocardial infarction (MI), diabetes, and previous atrial fibrillation (AF)/atrial flutter, as these were the comorbidities that were available in both the Medicare Chronic Conditions Warehouse (CCW) and Evolut Low-Risk trial.18 For FFS patients, the Medicare CCW was used to identify comorbidities based on standardized algorithms. Since the CCW only searched comorbidity codes in FFS claims, MA claims were digitally searched using the same International Classification of Diseases-10 (ICD-10) claims codes and algorithms as used in the Medicare CCW to ensure the comprehensive identification of comorbidities for MA patients.
One-year outcomes of interest selected from the trial included stroke (ischemic or hemorrhagic), bleeding, permanent pacemaker implantation (PPI), transient ischemic attack (TIA), major vascular complications, and death. Aortic valve reintervention could not be validated due to having no events in the MA group. Death was ascertained from the MBSF. The remaining outcomes were identified using claims algorithms that have been validated in a population of FFS patients or are currently being used in post-approval studies (Table S2).19,20 Since bleeding is an outcome that can be difficult to ascertain using claims, we trialed use of both a parsimonious and comprehensive claims algorithm for bleeding (Table S2). Disabling stroke was defined as the presence of a claims code for stroke with a hospital discharge destination that was death or not home (e.g., acute rehab, skilled nursing facility, etc). Although this is an imperfect definition for disabling stroke because of the potential for a patient’s discharge to a facility to be driven by something other than a stroke, it has been previously validated in Medicare fee-for-service using six large TAVR trials. 21 In contrast to FFS Medicare, MA plans can add or delete diagnoses from beneficiary encounters through chart reviews. Consequently, MA patients have encounter claims, which are claims linked to a specific episode of care, linked chart review claims, which are chart reviews conducted in the context of a patient encounter, and unlinked chart reviews, which are chart reviews conducted independent of any patient encounter. To account for this, the analysis was repeated with each of the following subsets of claims to determine the optimal dataset to use for validation: 1) all claims 2) only encounter claims; 3) encounter claims and all chart review claims; 4) encounter claims and encounter-linked chart review claims. For methods 2, 3, and 4, duplicate claims were eliminated using the method described by Jung et al.6
Baseline characteristics were compared between linked and unlinked U.S. patients ≥ 65 years old using Student’s T-test for continuous variables and Fisher’s Exact test for categorical variables to ensure that findings from the linked patient sample would be generalizable to the total patient population.
Covariates and outcomes were compared between trial data and claims for both FFS and MA patients. To compare the frequency of covariates and outcomes identified in claims to the adjudicated trial outcomes, sensitivity, specificity, positive predictive value, negative predictive value, and kappa were calculated for MA and FFS patients. The simple kappa coefficient was utilized to measure the agreement between the claim-derived and trial-adjudicated events. The simple kappa statistic and asymptotic variance of kappa were computed using the SAS “PROC FREQ” “agree” command, which calculates kappa as (Po-Pe)/(1-Pe), where Po is the level of observed agreement and Pe is the level of agreement expected by chance.22 The 95% confidence intervals were estimated using the normal approximation approach. The z-test was used to determine whether the difference in the kappa for MA vs. FFS was statistically significant at p<0.05. Using event-level matching, claims submitted within 14 days of an adjudicated outcome event were considered a match. Medicare claims were accessed through the CMS Virtual Research Data Center (VRDC) secure environment under an existing data use agreement rather than through Medicare Provider Analysis and Review (MedPAR) files, as our group has done previously.19–21 Although there are differences in the structure of FFS and MA claims, the process for requesting and accessing FFS and MA data through the VRDC was identical. All analyses were conducted using SAS version 9.4 (SAS Institute Inc, Cary, NC).
RESULTS
Of the 1139 U.S. patients ≥ 65 years old in the Evolut Low-Risk TAVR Trial, 884 patients (77.6%) were successfully linked to Medicare claims (Figure 1). After exclusion of individuals with incomplete or inconsistent data, 782 patients were included, of which 77.6% (n = 607) were enrolled in Medicare FFS and 22.4% (n = 175) were enrolled in MA. Only 2 MA patients (1.1%) and 12 FFS patients (2.0%) changed insurance during the one-year follow-up period. Linked and unlinked patients had a similar distribution of most baseline characteristics. However, linked patients had more advanced symptoms (27.7% NYHA III/IV vs. 22.1%, p=0.032), had lower rates of peripheral artery disease (7.1% vs. 11.0%, p=0.036) and prior percutaneous coronary intervention (12.7% vs. 18.2%, p=0.018), and were less likely to receive surgical valve replacement (44.6% vs. 57.1%, p <0.001). (Table 1).
Medicare claims had the highest sensitivity for identifying diabetes (MA: 95%, FFS: 95%) and hypertension (MA: 99%, FFS: 95%). Although the claims algorithm did not successfully identify any of the MA patients with a trial-documented history of MI, the specificity of the algorithm for prior MI was high in both MA and FFS (99% and 98%, respectively). Otherwise, specificity was highest for atrial fibrillation/flutter (MA: 96%, MA: 95%). Compared to gold-standard trial-adjudicated comorbidities, the CCW-claims based algorithm performed better in MA than FFS for identifying diabetes (kappa: 0.77 [95% CI: 0.67, 0.87] vs. 0.64 [95% CI: 0.58, 0.71], p =0.0327) and hypertension (kappa: 0.68 [95% CI: 0.49, 0.86] vs. 0.44 [95% CI: 0.34, 0.54], p=0.0253), although the confidence intervals were wide (Table 2).
There were minimal differences in the number of outcome events identified in MA, regardless of the inclusion or exclusion of linked/unlinked chart review claims (Table S3).6 Ultimately, both encounter and chart review claims were included in the primary analysis for MA patients to optimize use of all available data (Method 3).
In MA, sensitivities were low for TIA (25%), major vascular complication (43%), and bleeding (11%) (Table 3). Although using the comprehensive bleeding algorithm resulted in a significant decrease in specificity from 100% to 55% in MA compared to the parsimonious algorithm, it did improve sensitivities significantly and resulted in a sensitivity of 100% in MA. There was perfect identification of death in both MA and FFS claims. The selected claims algorithms had specificities ≥ 98% for all outcomes except for bleeding (comprehensive) which had specificities of 50% and 55% in FFS and MA respectively. Compared to gold-standard trial adjudicated outcomes, the claims algorithms performed similarly in MA and FFS for all outcomes (p > 0.05 for difference in kappas). There was high agreement (kappa > 0.70) for stroke (kappa: 0.79 [95% CI: 0.52, 1.00] vs. 0.77 [95% CI: 0.64, 0.89]), disabling stroke (kappa: 1.00 [95% CI: 1.00, 1.00] vs. 0.85 [95% CI: 0.69, 1.00]), permanent pacemaker implantation (kappa: 0.84 [95% CI: 0.91, 0.96] vs. 0.91 [95% CI: 0.96, 0.96], and death (kappa: 1.00 [95% CI: 1.00, 1.00] vs. 1.00 [95% CI: 1.00, 1.00]) in both MA and FFS (Figure 2).
DISCUSSION
In this study, we evaluated the performance of previously validated FFS claims algorithms in the MA population by comparing the accuracy of comorbidity and outcome identification in MA vs. FFS claims for patients enrolled in the Evolut Low-Risk TAVR Trial. We selected this trial for validation because claims data are currently being used to support FDA-mandated post-approval studies for low-risk TAVR and evidence development for national coverage determinations through CMS.17,23–26 Furthermore, it is especially important to evaluate long-term real-world outcomes in the low-risk TAVR population in whom valve durability could impact the decision to pursue TAVR versus surgery.
In our comparison of MA and FFS, our study had three main findings. First, claims-based algorithms performed at least as well in MA as FFS for identifying all comorbidities except prior myocardial infarction. Second, claims performed similarly in MA and FFS for all outcomes —although confidence limits were wide for many of these parameters given the modest sample size of the MA cohort. Third, as has been demonstrated previously in FFS, we found that claims perform best for identifying procedures, such as permanent pacemaker implantation, and death, and performed reasonably well for endpoints such as stroke, including disabling stroke.20 Our findings support the use of certain claims algorithms as complementary tools for long-term outcome assessment in MA patients, as has been previously done in FFS.
Concurrent with the rise in patients insured by MA comes an increasing need to evaluate long-term outcomes of cardiovascular devices in MA patients. Although Medicare claims are a cornerstone for post-market medical device surveillance, to our knowledge, this is the first study of its kind to evaluate whether claims algorithms can be used with similar validity in MA patients in whom there is an incentive to code for more diagnoses for risk-adjustment and reimbursement, thus impacting both comorbidity assessment and outcomes ascertainment.
One limitation of this study is that the relatively low event rates in the Evolut Low-Risk TAVR Trial limit our ability to fully validate all claims outcomes in the MA population and result in wide confidence intervals for the kappa statistic. Furthermore, MA patients made up only 22% of the study population. Nonetheless, it is reassuring that the algorithms for stroke, disabling stroke, and death performed well in MA, as these are some of the most important outcomes for long-term TAVR follow-up. In addition, we are only able to draw conclusions about the patients that we successfully linked to Medicare claims, and it is possible that linked patients may differ from unlinked patients in ways that are not measured in our study. Looking forward, it will be important for future studies to validate the use of MA claims for long-term assessment of aortic valve reintervention and of other outcomes that may be prone to upcoding in MA relative to FFS.
AO
Dr. Andrew Oseran
Cardiology · Advanced Heart Failure & TransplantationBeth Israel Deaconess Medical Centerandrewo