Methodologic Approaches to Examining Home Health Using Traditional Medicare and Medicare Advantage Claims Data

Abstract

Objective

To describe a claims-based methodology for constructing new home health stays using traditional Medicare (TM) claims data and Medicare Advantage (MA) encounter data.

Study Setting and Design

To demonstrate our methodology’s performance, we assessed the percentages of TM and MA beneficiaries with one and two or more stays, and the mean length of a stay (LOS) among home health recipients. We compared 2019 and 2021 results to evaluate the methodology’s feasibility pre- and post-implementation of the Patient-Driven Groupings Model (PDGM).

Data Sources and Analytic Sample

We used 2019 and 2021 TM and MA home health claims and 2019 Outcome and Assessment Information Set for a nationally representative 20% sample of Medicare beneficiaries.

Principal Findings

In 2019, a lower percentage of MA beneficiaries had new home health stays than TM (5.9% vs. 6.5%). Among home health recipients, approximately 90% had a single stay. The mean LOS in MA was 39 days, compared with 44 days in TM. The statistics from the 2021 data were similar, except that the mean LOS in TM increased to 46 days.

Conclusions

Our claims-based new home health stay methodology is feasible both pre- and post-PDGM and would enable direct comparisons of home health utilization in TM and MA.

INTRODUCTION

Medicare-covered home health (HH) care delivers skilled nursing and rehabilitation therapy directly to homebound individuals in their homes. Each year, over four million Medicare beneficiaries receive HH care.¹ Medicare Advantage (MA)—the Medicare component administered by private insurers with public subsidies—now enrolls over half of the Medicare population and continues to grow.² MA and traditional Medicare (TM) differ in their coverage of HH. In TM, a physician must certify a beneficiary’s HH needs, and there is no cost-sharing for HH recipients. MA plans can further impose prior authorization for HH,³ can require cost-sharing, and capitated payments may affect their incentive to authorize HH.^4,5

The existing literature examining HH utilization suggests that enrollment in MA is associated with lower HH use and shorter lengths of stay than in TM.^1,4,6–11 Much of this work uses the Outcome and Assessment Information Set (OASIS), a required clinical assessment completed for HH patients, to estimate the receipt and use of HH services. Researchers have relied on OASIS in part because MA encounter records were not readily available until recently. However, using OASIS data to measure utilization has three important limitations. First, while the length of stay can be derived from OASIS, it does not include detailed utilization (e.g., dates of visits and services provided). Second, there are concerns about the completeness of OASIS data for MA beneficiaries.^1,12 While HH agencies are required to submit OASIS assessments for all Medicare beneficiaries receiving HH, only TM payments are tied to the submission of assessments. Third, OASIS also experienced higher rates of missing beneficiary identifiers during 2020–2023,¹³ especially for MA,¹ which may lead to an underestimation of utilization. Other work has focused on a large national MA carrier or a large national HH agency to examine utilization with more detailed organizational records.^6,14,15 These studies provide insights into HH delivery across TM and MA, but generalizability may be limited.

The nationally representative MA encounter data, with availability dating back to 2015, represents a new opportunity for detailed comparisons of HH utilization in TM and MA. While the completeness of the MA HH encounter remains a concern, recent analyses have shown that completeness has been steadily improving.¹² Emerging studies using the data have compared utilization in TM and MA at the beneficiary level,^1,10,16 which offers important insights but should be interpreted with caution where beneficiaries have multiple stays. These beneficiaries can be of particular interest to researchers and policymakers due to their higher disease burden and greater overall health care utilization. To use the data for stay-level comparisons, however, one major challenge researchers face is that HH services are documented differently in MA and TM data—and even across MA plans—as a result of variation in payment methodologies. TM payments are episode-based (with multiple episodes possible across a single stay), whereas, depending on their contracts, MA plans can pay HH agencies by episodes or bundles of visits.^1,4,15 Such differences have presented a challenge to direct comparisons of HH utilization at the stay level for researchers and policymakers.

Therefore, in this paper, we introduce a novel claims-based methodology for constructing new HH stays in TM and MA that would enable direct comparisons at the stay level. We present summary statistics of the constructed new HH stays in 2019 and 2021 to demonstrate this methodology’s feasibility both pre- and post-implementation of the Patient-Driven Groupings Model (PDGM) in 2020, which represented major revisions to traditional Medicare’s HH payment system.

METHODS

Data and Sample

Our primary data sources were the 2019 and 2021 HH claims (TM) and encounter records (MA) and 2019 OASIS file for a nationally representative 20% sample of Medicare beneficiaries. Hereafter, we will refer to both TM and MA records as claims data. For both years, the study sample included Medicare beneficiaries 65 years and older who were continuously enrolled in TM or MA with both Part A and Part B coverage. We further restricted to those who resided in the same county in one of the 50 states or the District of Columbia (DC) during the year. We also required that MA beneficiaries stay in the same plan. Our beneficiary sample included 3,210,086 MA beneficiaries and 4,846,644 TM beneficiaries in 2019, and 3,795,359 MA and 4,666,785 TM in 2021.

Constructing Claims-Based New HH Stays

Differences in HH payment methods between TM and MA result in different claim structures. In TM, Medicare reimburses HH agencies by episodes of care. An HH stay is a sequence of payment periods between admission and discharge.¹⁷ A payment period lasted for 60 days pre-2020, and the implementation of PDGM in 2020 shortened it to 30 days. In contrast, in MA, payment methods vary and can be based on episodes or bundles of visits.^1,15 As a result, an HH stay in TM typically consists of one or more consecutive claims, each documenting care provided during a payment period, with clearly documented admission and discharge dates. However, an HH stay in MA comprises claims of various durations, with gaps in between and no reliable admission or discharge dates.

We constructed claims-based new stays to facilitate direct comparisons between TM and MA (Figure 1). For TM, all claims of a HH stay would have the same admission date, and the discharge date would be populated on the last claim. We first ensured that for each beneficiary, claims were displayed chronologically, using the claim through date. Then, to ensure that we identified a new stay, we required that the beneficiary had no HH visits during the 60-day (30-day for 2021) lookback period prior to the admission date. The 2021 TM data required a modified approach because the admission date field was unavailable from 2020 to 2023.¹⁸ We applied the methodology for MA, as described below, to the 2021 TM claims.

Figure 1.

Constructing Claims-Based New Home Health Stays Using Traditional Medicare and Medicare Advantage Claims

Note: TM, traditional Medicare; MA, Medicare Advantage; PT, physical therapy; OT, occupational therapy; HH, home health; SW, social work. This figure illustrates the methodology for years prior to the implementation of the Patient-Driven Groupings Model (PDGM).

For MA, we first excluded chart review records and retained only the final action records. Then, we deduplicated the records, considering a unique combination of beneficiary ID, claim from (start) date, claim through (end) date, and organization national provider identifier (NPI) as a unique service record. Next, for a beneficiary, we connected claims less than 60 days (30 for 2021) apart from each other to form a new stay such that the beneficiary had no HH visits during the 60-day (30-day for 2021) lookback period prior to the first claim and no visits during the 60 days (30 days for 2021) after the last claim. The admission date would then be the first claim from date, and the discharge date would be the last through date. For example, suppose that a beneficiary had three claims in 2019: the first spanned from April 5 to April 25, the second from July 10 to July 16, and the third from July 25 to August 30. The second and third claims would be grouped into a new stay, with an admission date of July 10 and a discharge date of August 30.

The Completeness of HH Claims and OASIS

First, we examined the completeness of claims and OASIS for TM and MA in 2019 in two ways. Respectively for TM and MA, we identified beneficiaries who appeared in either claims or OASIS (i.e., beneficiaries who likely received HH services at some point), and calculated the percentages that appeared in claims, OASIS, and both. Additionally, among beneficiaries with new HH stays documented in either claims or OASIS, we calculated the percentages that had new stays documented in claims, OASIS, and both. To identify new HH stays in OASIS, we focused on start-of-care assessments. We considered a start-of-care assessment to represent a new stay if there were no assessments during the 60 days prior to the start-of-care assessment date. We only examined 2019 due to a considerably higher rate of missing beneficiary identifier reported by Chronic Conditions Warehouse in the 2021 OASIS.¹³ Because we used a single year of data to demonstrate our methodology, we were unable to identify new stays that started during the first 60 days of the year. Our new stays thus only included those that started on or after March 2. However, the window of observation is extendable when a longer panel of claims is used.

Assessment of Feasibility

While there is no available gold standard against which we can test the validity of our methodology for MA, we assessed its performance with the 2019 TM claims, taking advantage of the fact that the 2019 data reported admission and discharge dates. To do that, we applied the methodology for MA—including identifying new HH stays and imputing admission and discharge dates—to the 2019 TM claims and evaluated to what extent we identified the same stays as we did when using the original admission and discharge dates on the claims. To assess the feasibility of our constructed new HH stay measure both pre- and post-PDGM, for both 2019 and 2021, we calculated the percentages of TM and MA beneficiaries with one and two or more stays, and the mean length of a HH stay (LOS) for those with a new stay. Because of the 60-day lookback period, for 2019, we could only ascertain a discharge date for an MA stay if it ended at least 60 days prior to the end of the year. As a result, for both TM and MA, our LOS analysis only included those that concluded on or before November 1. Although 2021 only required a 30-day lookback period, for the purpose of balanced comparisons, we applied the same admission and discharge date restrictions to 2021. For beneficiaries with multiple stays in one year, the analysis only included their first stay.

RESULTS

Among our sample of Medicare beneficiaries in 2019, 270,080 (8.4%) MA beneficiaries and 471,800 (9.7%) TM beneficiaries had either HH claims or OASIS and likely received HH services. Among likely HH recipients, 91.1% of MA beneficiaries appeared in HH claims (Figure 2), comparable to that of TM beneficiaries (92.8%). However, 83.5% of MA beneficiaries were found in OASIS, in contrast to the 91.3% among TM beneficiaries. Driven primarily by less complete OASIS for MA than for TM, 74.6% of MA beneficiaries were in both claims and OASIS, while 84.1% of TM beneficiaries were in both files. These patterns were consistent with findings reported by MedPAC,¹² and they held when we focused on TM and MA beneficiaries’ new HH stays recorded in claims and OASIS (Appendix Figure 1).

Figure 2.

Percentages in Home Health Claims, in OASIS, and in Both Among Beneficiaries That Were in Either Home Health Claims or OASIS, by MA and TM in 2019

Note: TM, traditional Medicare; MA, Medicare Advantage; OASIS, Outcome and Assessment Information Set.

When applied to the 2019 TM claims, our new HH stay construction yielded very similar stays, compared with directly using the claim-reported admission and discharge dates. The two approaches identified the same new HH stays for 95.8% of the time.

Table 1 presents the results from our new HH stay construction. In 2019, MA beneficiaries in our sample were slightly less likely than TM beneficiaries to have new HH stays (5.9% vs. 6.5%). Among those with any new stays, about 90% of them had exactly one stay. The mean LOS among MA HH recipients was 39.0 days, which was 5 days (12%) shorter than the mean among TM recipients (44.3 days). The statistics from the 2021 data were similar to those from 2019, except that the average new TM HH stay was 1.7 days longer (46.0 days), which is consistent with findings in a recent study.¹⁹

Table 1.

The Distribution of Number of New Stays per Beneficiary and Mean Length of a Stay

	2019		2021
	MA	TM	MA	TM
No new staya	94.1%	93.5%	93.8%	93.4%
With new staysa	5.9%	6.5%	6.2%	6.6%
1	5.4%	6.0%	5.3%	5.8%
2 or more	0.5%	0.5%	0.8%	0.8%
Mean length of a stay (days)b	39.0	44.3	39.2	46.0
N	3,210,086	4,846,644	3,795,359	4,666,785

Note: TM, traditional Medicare; MA, Medicare Advantage.

^aNew stays only include those started on or after March 2 of each year.

^bThe sample for length of a stay includes new stays that started on or after March 2 of each year and ended on or before November 1. Only the first one is included when a beneficiary had multiple new stays in a year.

DISCUSSION

This paper introduces a novel claims-based methodology for constructing new HH stays in TM and MA claims data. This methodology would enable researchers to directly compare HH utilization, including LOS and the type, number, and timing of visits, in TM and MA, as well as across MA plans with various payment structures. Using our methodology, we find that MA beneficiaries are less likely than TM beneficiaries to have a new HH stay in a year. For HH recipients, the mean length of stay is more than 10% shorter in MA than in TM. These summary statistics are consistent with those reported in the literature.^1,6,20 Results are also consistent in 2019 and 2021, indicating the feasibility of our methodology both pre- and post-PDGM implementation, a significant revision to the HH payment system for TM.

Consistent with a report by the Medicare Payment Advisory Commission (MedPAC),¹² we find issues with OASIS completeness for MA. MA HH encounter data is reasonably complete overall, although there could be plan-level variation. The findings have important implications for various types of studies examining HH utilization in Medicare Advantage. For studies that examine rates of HH utilization, researchers should combine encounter data and OASIS. But using encounter data alone may be less subject to bias than OASIS alone. For studies that evaluate LOS and HH visits, it may be appropriate to use encounter data, especially given the recent improvement in completeness.¹² We recommend examining plan-level variation in record completeness and whether results are robust to the exclusion of plans with less complete records. OASIS assessments remain an important data source for HH research, with rich, patient-level data to inform utilization research, such as living arrangement and functional status. However, more transparency is needed regarding the missing or unlinked OASIS records. When using linked encounter and OASIS data, we recommend that researchers verify whether unlinked encounter and OASIS occur at random and whether results are driven by the exclusion of unlinked records.

Limitations

This study has three notable limitations. First, our new stay methodology is unable to identify resumption of HH care after admission to an inpatient facility. Second, we only examined the completeness of MA encounter and OASIS in 2019, due to limitations of our 2021 OASIS data. Third, plan-level variation in MA encounter record completeness was out-of-scope for this study, but it is an important topic for future work.

CONCLUSIONS

As enrollment in Medicare Advantage continues to grow, it is critical we understand how it affects care delivery. MA HH encounter data is an important new data source for studying HH among over half of Medicare beneficiaries. This paper describes a claims-based methodology for constructing new HH stays in TM and MA that would enable direct comparisons of HH utilization both between TM and MA and across MA plans. We also find issues with OASIS’ completeness for MA, which has important implications for studies seeking to examine HH in MA.

What is known on this topic

• Much of prior research comparing home health utilization in traditional Medicare (TM) and Medicare Advantage (MA) has relied on the Outcome and Assessment Information Set (OASIS).

• However, OASIS does not include detailed utilization information (e.g., dates of visits and services provided), and there are concerns about its completeness for MA.

• The MA encounter data represents a new opportunity for comparing home health utilization in TM and MA. But direct comparisons are challenging because services are documented differently in MA.

What this study found

• To enable direct comparisons between TM and MA, we present a novel claims-based methodology for constructing new home health stays using TM claims data and MA encounter data.

• Using our methodology, we find that compared with TM beneficiaries, MA beneficiaries are less likely to have new home health stays, and their mean length of stay is shorter.

• Our methodology is feasible both pre- and post-implementation of the Patient-Driven Groupings Model.