Post-Acute Care in Inpatient Rehabilitation Facilities Between Traditional Medicare and Medicare Advantage Plans Before and During the COVID-19 Pandemic

Abstract

Objectives

Compare post-acute care (PAC) utilization and outcomes in inpatient rehabilitation facilities (IRF) between beneficiaries covered by Traditional Medicare (TM) and Medicare Advantage (MA) plans during the COVID-19 pandemic relative to the previous year.

Design

This multiyear cross-sectional study used Inpatient Rehabilitation Facility–Patient Assessment Instrument (IRF-PAI) data to assess PAC delivery from January 2019 to December 2020.

Setting and Participants

Inpatient rehabilitation for stroke, hip fracture, joint replacement, and cardiac and pulmonary conditions among Medicare beneficiaries 65 years or older.

Methods

Patient-level multivariate regression models with difference-in-differences approach were used to compare TM and MA plans in length of stay (LOS), payment per episode, functional improvements, and discharge locations.

Results

A total of 271,188 patients were analyzed [women (57.1%), mean (SD) age 77.8 (0.06) years], among whom 138,277 were admitted for stroke, 68,488 hip fracture, 19,020 joint replacement, and 35,334 cardiac and 10,069 pulmonary conditions. Before the pandemic, MA beneficiaries had longer LOS (+0.22 days; 95% CI: 0.15–0.29), lower payment per episode (−$361.05; 95% CI: −573.38 to −148.72), more discharges to home with a home health agency (HHA) (48.9% vs 46.6%), and less to a skilled nursing facility (SNF) (15.7% vs 20.2%) than TM beneficiaries. During the pandemic, both plan types had shorter LOS (−0.68 day; 95% CI: 0.54–0.84), higher payment (+$798; 95% CI: 558–1036), increased discharges to home with an HHA (52.8% vs 46.6%), and decreased discharges to an SNF (14.5% vs 20.2%) than before. Differences between TM and MA beneficiaries in these outcomes became smaller and less significant. All results were adjusted for beneficiary and facility characteristics.

Conclusions and Implications

Although the COVID-19 pandemic affected PAC delivery in IRF in the same directions for both TM and MA plans, the timing, time duration, and magnitude of the impacts were different across measures and admission conditions. Differences between the 2 plan types shrank and performance across all dimensions became more comparable over time.

Post-acute care (PAC) provides patients with continuous recovery after hospital discharge.¹ According to the Medicare Payment Advisory Commission, more than 40% of Medicare beneficiaries used PAC after hospital discharge in 2015.² Among all types of PAC facilities, inpatient rehabilitation facilities (IRFs) offer the most intensive rehabilitation with the highest average cost per episode. The higher cost growth in IRF makes it the target of cost containment for both public and private payers as well as institutional care providers and/or health care systems.^{1 , 2}

The Traditional Medicare (TM) and Medicare Advantage (MA) plans are the 2 predominant payer types for aging care, which makes the 2 important comparators to each other on care quality and efficiency.3, 4, 5, 6, 7 More recently, with the development of MA plans, MA market share increased from 31% in 2015 to 42% in 2019 out of all Medicare beneficiaries.^{8 , 9} An IRF is paid by TM for the whole episode of care via Prospective Payment Systems depending on a patient's primary admission diagnosis and case-mixed group (CMG).^{10 , 11} In contrast, MA plans receive a risk-adjusted monthly capitated payment per enrollee to cover all health care services. As a result, MA plans may choose to provide less IRF overall, shorter length of stay (LOS), or lower payment per episode.^{5 ,} 12, 13, 14 Further, MA plans also use the provider networks, patient cost-sharing, and utilization authorization to manage and control costs.^{15 , 16}

Previous research comparing TM and MA plans focused more on general utilization, spending, preventive care, or acute care,^{4 , 8 ,} 17, 18, 19 but less on PAC or among IRFs.5, 6, 7 Within a few studies that focused on PAC, findings suggested that MA enrollees used fewer PAC services, shorter LOS,^{6 , 7 , 13} and were less likely to receive care from high-quality facilities or agencies.^{7 , 20} After controlling for patient and facility characteristics, however, MA enrollees on average had no worse functional improvements during IRF stays and even higher likelihood to return to community after discharge than TM beneficiaries.^{6 , 20} Studies that leveraged more detailed MA plan attributes suggested that TM and MA differences were also influenced by MA plan types (preferred provider organization vs health maintenance organization), contracting and administrative approaches (eg, networks, cost-sharing, utilization review and reauthorization, use of more nonprofit agencies).^{7 , 14 , 16} Studies that focused on certain diagnoses or specific PAC facility types, such as home health or skilled nursing facilities (SNFs), also had similar findings.^{20 , 21}

The outbreak of the COVID-19 pandemic since 2020 changed the non-COVID care utilization,22, 23, 24, 25 use of PAC,26, 27, 28 and choices among PAC facility types substantially.^{26 , 29} For example, hospitals shortened LOS for patients with non-COVID cases and substituted with PAC to alleviate the burden caring for the COVID cases.^{26 , 27} Patients were more likely to choose home health services rather than SNFs for PAC to reduce the risk of infection.^{27 , 28 , 30} Less is known of care utilization and outcomes in the IRF settings. IRF provides more intensive care than other PAC facilities such as SNFs or home health agencies (HHAs), and is more likely to be used to supplement acute care if needed.^{26 , 29} Yet IRF is also the most expensive among all PAC facility types. As a result, differentiated changes in utilization and outcomes were more likely to happen during the pandemic between TM and MA plans because of administrative and operational differences.^{6 , 7 , 14} All these call for the need for a close comparison between TM and MA in the IRF settings during the pandemic. Understanding the changes before and during the COVID-19 pandemic period and the potentially disproportional changes between patients with TM and MA plans provide important insights on the impacts of the pandemic on PAC services. Stakeholders such as policy makers, providers, and patients will benefit from this evidence to target on alleviating the pandemic impacts and more proactively planning for care improvement beyond the COVID-19 pandemic.27, 28, 29, 30

The objective of this study was to provide a comparison in the IRF settings before and during the first year of the COVID-19 pandemic between Medicare patients with TM and MA plans. This study focused on the 5 most common admission conditions for the aging population and compared across measures of care utilization (LOS), costs, and care outcomes, such as functional improvements and discharge locations. We hypothesize that PAC services utilization and outcomes changed during the COVID-19 pandemic relative to the pre-pandemic period. Yet the magnitude, timing, and duration of changes were different between TM and MA plans even after risk adjustment and sample matching among the patient population.

Methods

Data Sources

Data used in this study are the IRF–Patient Assessment Instrument (IRF-PAI) offered by the Uniform Data System for Medical Rehabilitation.³¹ IRF-PAI is a tool for patient assessment required by the Centers for Medicare and Medicaid Services (CMS) for payment reimbursement.³² IRF-PAI includes information on patient socioeconomic status, clinical conditions such as primary diagnosis and comorbidities, and IRF facility characteristics, as well as utilization, costs, and outcomes of the IRF stays, such as LOS, payment, discharge location, and functional assessments, which were evaluated within 72 hours of IRF admission and discharge.33, 34, 35 Previous studies have shown the validity and representativeness of this dataset.36, 37, 38, 39 This study is exempt for review by the Institutional Review Board of University of Wisconsin-Madison. Informed consent was waived due to data anonymity. This study follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

Study Sample

The unit observation of the study sample is 1 patient-episode. The study sample included Medicare (both TM and MA) beneficiaries 65 years or older who were admitted to IRFs for inpatient rehabilitation services between January 2019 and December 2020 for stroke (Impairment Group Code-IGC 01.1-9), hip fracture (08.11-12), joint replacement (08.51-52, 08.61-62, 08.71-72), and cardiac (09) and pulmonary conditions (10.1, 10.9), the 5 common admission conditions for the aging population.^{5 , 6 , 40} Excluded from the sample were those patients who were not admitted for initial rehabilitation (eg, transfers) or died during the rehabilitation stay, whose pre-hospitalization living settings were non-home, whose LOSs were longer than 30 days or shorter than 3 days, whose rehabilitation programs were interrupted, or who were discharged against medical advice. These patient-episodes were deemed to be different from others due to complicated clinical concerns and hence were less comparable.36, 37, 38, 39 Supplementary Figure 1 shows the flowchart of sample derivation and the number of excluded episodes in each step.

Supplementary Fig. 1.

Flowchart for sample derivation.

Main Variables

The outcome variables included LOS, payment for the episode, discharge locations, and functional improvements.^{38 , 39 , 41} The discharge locations included home with self-care (as default), home with home health care, SNFs, and others.^{42 , 43} Functional improvements were measured as the difference in functional scores at admission and discharge and were each measured separately for self-care (Section GG130) and mobility (Section GG170) in the IRF-PAI assessment data.^{37 , 44} Section GG is a standardized assessment implemented by CMS in PAC. The assessment measures a patient's need for assistance with self-care and mobility while also documenting the patient's prior level of function.33, 34, 35

The main exposure variables were Medicare coverage type and the pandemic timing. Medicare coverage type is indicated by whether a beneficiary's primary payment source was TM or MA plans. The pandemic timing was defined by a patient's IRF admission date. The pre-pandemic period was defined from January 2019 to February 2020. The pandemic period was defined by 3-month intervals to differentiate the initial start (March–May 2020), the recovery time (June–August 2020), and the recurring time (September–December 2020). Separation of time interval during the pandemic allowed us to capture the timing of changes and recovery of IRF utilization and outcomes relative to the pre-pandemic period and compare between TM and MA plans for the differentiated COVID responses if any.

The control variables included patient demographic characteristics, clinical conditions, IRF facility characteristics, and monthly fixed effects. The patient demographic characteristics included age, sex, marital status, and Medicare and Medicaid dual coverage.^{40 , 45} Clinical conditions included patients’ comorbidity tiers (None, Minor, Moderate, and Major) and CMGs¹¹ as defined by CMS for prospective payment purposes.¹⁰ The facility characteristics included bed size, IRF facility type (as a unit within an acute care hospital or a standalone facility), and the CMS region within which the IRF belonged to.

Study Design and Statistical Analysis

The study adopted a difference-in-differences design to conduct 2 steps of analysis. In the first step, the study used the full sample to estimate the predicted outcomes before and during the pandemic period for episodes covered by TM and MA plans, respectively. Using the full study sample allowed us to obtain the average predictions of outcome measurements with the most comprehensive national scope. In the second step, the study used a 2-round matching process to arrive at a balanced sample between TM and MA plans both before and during the pandemic period. This balanced sample was used to estimate the differences between TM and MA plans during the pandemic relative to the pre-pandemic period. This matched sample alleviated the concerns for sample selection issues that happened between TM and MA (eg, selection to MA plans for coverage, selection to use IRF in MA plans), as well as between before and during the pandemic period (eg, selection to acute care and/or subsequently to IRF).

First, with the full sample, we first showed the monthly admission volume by admission condition as a general trend before and during the pandemic period. Risk-adjusted outcome variables for each admission condition were then predicted by regressions of each outcome on the Medicare coverage type, the pandemic time indicator, the interaction between the two, and the control variables. LOS was predicted using Poisson models. Logarithm of payment and the absolute values of functional improvements were predicted using linear regressions. Discharge location was predicted by multinomial regressions with discharges to home with self-care as the default group. The analysis was conducted separately by admission condition. Separating each admission condition for analysis accounted for the clinical differences across conditions and allowed for condition-specific outcome estimates. This separation process also alleviated the sample selection concerns that happened at the admission condition level between TM and MA plans, as well as during the pandemic period relative to before.

Second, to obtain a balanced sample for each admission condition, we used a 2-round matching process. In the first round, we conducted a 1:1 exact match between episodes covered by TM and MA plans based on all control variables except the admission month. This matching process was done separately for time periods before and during the pandemic. The matched sample had balanced patient and facility characteristics between TM and MA plans for each time interval. In the second round, we used the obtained balanced sample in the first round to conduct a second round of matching between patient-episodes that happened before and during the pandemic. This second round of matching process yielded a further balanced sample with episodes comparable before and during the pandemic period.

To report the results, the predicted average outcomes by Medicare coverage type and time interval were calculated from the full sample using the margin command in Stata (StataCorp LLC). The predicted differences between plan type and across time intervals were calculated from the matched sample using the margin, contrast command in Stata. Sensitivity analysis was conducted by including the excluded episodes for both full- and matched-sample analysis, as well as estimation of the predicted differences using the unmatched full sample. All analysis was conducted using Stata SE 17 from May to August 2022.

Results

Study Sample Description

The final study sample included 271,188 patient-episodes over 24 months (2019 and 2020), among which there were 138,277 for stroke, 68,488 for fracture, 19,020 for hip or joint replacement, and 35,334 for cardiac and 10,069 for pulmonary conditions during the study period. The sample had mean (SD) age of 77.8 (0.06) years and included 57.1% women, 85.8% nonminority White, 44.9% married or living with a life-partner, 18.4% with MA coverage, and 8.4% with Medicare and Medicaid dual coverage (Table 1 ).

Table 1.

Patient Demographic and Facility Characteristics Before and During COVID-19 by Admission Condition

	Stroke				Fracture				Replacement				Cardiac				Pulmonary
	Pre-Pandemic		During		Pre-Pandemic		During		Pre-Pandemic		During		Pre-Pandemic		During		Pre-Pandemic		During
	n = 84614		n = 51040		n = 38286		n = 28990		n = 12800		n = 6054		n = 21791		n = 13053		n = 6527		n = 3403
Age (m/SD)	77.1	0.0	76.9	0.0	80.1	0.0	80.3	0.0	75.6	0.1	76.0	0.1	77.9	0.1	77.6	0.1	77.7	0.1	77.0	0.1
Sex (n/%)
Male	40,893	48.3	24,918	48.8	11,680	30.5	8944	30.9	4437	34.7	2167	35.8	11,995	55.1	7091	54.3	2840	44.0	1599	47.0
Female	43,719	51.7	26,122	51.2	26,606	69.5	20,046	69.2	8363	65.3	3887	64.2	9795	45.0	5962	45.7	3687	56.0	1804	53.0
Race (n/%)
White	67,155	79.4	40,746	79.8	34,513	90.2	26,270	90.6	10,675	83.4	5213	86.1	18,837	86.4	11,293	86.5	5835	89.0	2918	85.8
Black or Hispanic	14,742	17.4	8731	17.1	3010	7.9	2192	7.6	1727	13.5	726	12.0	2573	11.8	1543	11.8	598	9.0	425	12.5
Asian or Other	2717	3.2	1563	3.1	763	2.0	528	1.8	398	3.1	115	1.9	381	1.8	217	1.7	94	1.4	60	1.8
Marital status (n/%)
Married/with partner	41,046	48.5	24,243	47.5	16,483	43.1	12,159	41.9	6111	47.7	2798	46.2	10,412	47.8	5907	45.3	2647	41.0	1382	40.6
Single/live alone	40,568	47.9	24,641	48.3	20,564	53.7	15,773	54.4	6332	49.5	3039	50.2	10,736	49.3	6665	51.1	3658	56.0	1857	54.6
Unknown	3000	3.6	2156	4.2	1239	3.2	1058	3.7	357	2.8	217	3.6	643	3.0	481	3.7	222	3.4	164	4.8
MA (n/%)
No	58,267	68.9	32,841	64.3	33,714	88.1	23,587	81.4	11,257	88.0	5099	84.2	19,015	87.3	10,552	80.8	5933	90.9	2803	82.4
Yes	26,347	31.1	18,199	35.7	4572	11.9	5403	18.6	1543	12.1	955	15.8	2776	12.7	2501	19.2	594	9.1	600	17.6
Medicaid coverage (n/%)
No	77,303	91.4	46,468	91.0	35,671	93.2	27,002	93.1	11,938	93.3	5672	93.7	20,049	92.0	11,946	91.5	5797	88.8	2987	87.8
Yes	7311	8.6	4572	9.0	2615	6.8	1988	6.9	862	6.7	382	6.3	1742	8.0	1107	8.5	730	11.2	416	12.2
Comorbidity tier (n/%)
None	44,309	52.4	25,036	49.1	22,638	59.1	15,788	54.5	7461	58.3	3355	55.4	7269	33.4	3862	29.6	2173	33.3	1010	29.7
Major	1812	2.1	1149	2.3	953	2.5	740	2.6	123	1.0	65	1.1	1311	6.0	881	6.8	284	4.4	201	5.9
Medium	1060	1.3	748	1.5	2456	6.4	2162	7.5	345	2.7	194	3.2	2021	9.3	1367	10.5	905	13.9	509	15.0
Minor	37,133	43.9	24,086	47.2	12,204	31.9	10,299	35.5	4836	37.8	2439	40.3	11,157	51.2	6941	53.2	3161	48.4	1682	49.4
Facility type (n/%)
Freestanding	34,912	41.3	21,828	42.8	17,307	45.2	12,904	44.5	7205	56.3	3146	52.0	10,151	46.6	5845	44.8	2911	45.0	1207	35.5
Unit in hospital	49,702	58.7	29,212	57.2	20,979	54.8	16,086	55.5	5595	43.7	2908	48.0	11,640	53.4	7208	55.2	3616	55.0	2196	64.5
No. of certified beds (m/SD)	47.71	0.1	47.82	0.2	46.84	0.2	46.00	0.2	51.37	0.3	48.78	0.4	49.82	0.2	48.72	0.3	48.88	0.4	43.04	0.5

Admission Volume, Patient Composition, Facility Characteristics, and Share of MA Beneficiaries

During the pandemic, the IRF admission volume declined in 4 of the 5 common diagnoses, with the average monthly decline ranging between 16% and 34% in March 2020. Fracture was the only condition that had a small increase in monthly admission (6%). Larger changes in admission volume were found during the first 3 months since the pandemic started. By the end of 2020, monthly admission volume was only slightly lower than the pre-pandemic period for stroke, fracture, replacement, and pulmonary conditions, and slightly higher for fracture (Figure 1 ).

Fig. 1.

Number of admissions by month and condition.

Patient demographics and socioeconomic status were comparable before and during the pandemic in general, except for gender and race/ethnicity for some diagnoses. Proportionally more male and more minority patients were admitted for pulmonary conditions during the pandemic than the previous year (male during vs pre-: 47% vs 44%; minority 14.2% vs 11.0%). Fewer minorities were admitted for joint replacement during the pandemic than before (13.9% vs 16.6%) (Table 1).

However, patients admitted during the pandemic were more likely to have comorbidities than before (ie, 3 to 4 percentage points higher across all admission conditions) and proportionally more patients were classified in more severe CMGs for the corresponding admission condition (eg, CMG 103–106 for stroke, 704 for fracture, 804–805 for replacement, and 1403–1404 for cardiac and 1503–1504 for pulmonary conditions). The percentage shares of IRF admissions with MA plans were also significantly higher during the pandemic than before (by 3 to 9 percentage points for all admission conditions). The geographic distribution across 10 CMS regions did not change significantly (Table 1).

Regarding the facility characteristics, patients admitted for joint replacement and pulmonary conditions were more likely to be treated in the rehabilitation facilities within the same acute hospitals (as in-unit rather than standalone facilities) during the pandemic than before (48% vs 44% for replacement and 65% vs 55% for pulmonary conditions), whereas patients admitted for the other 3 elective conditions had comparable distributions in facility type before and during the pandemic. The average number of certified beds (as a measure of the facility size) was also significantly lower for replacement and pulmonary admissions during the pandemic than before (49 vs 51 and 43 vs 49 beds, respectively) but were comparable for the other 3 nonelective conditions (Table 1).

Length of Stay

Before risk adjustment, the average LOS per episode was 0.2 to 0.5 days longer during the pandemic than before for all 5 conditions (P < .05). Admissions for joint replacement and pulmonary conditions experienced larger increases in LOS during the pandemic than the other 3 conditions (ie, increased by 0.5 days/episode, representing 4.8% and 4.2% relative increase from the previous year, respectively). These increases in LOS before risk adjustment reflected the changes in the distribution of patient composition (eg, across CMGs and comorbidity tiers) during the pandemic relative to the pre-pandemic period for all conditions.

After adjusting for patient- and facility-level characteristics, the estimated LOSs were even shorter during the pandemic period than the previous year. Significantly shorter LOSs were found among the stroke admissions for both TM and MA coverage types (0.7 days shorter, P < .01) and among the cardiac admissions for MA plans (0.6 days shorter, P < .01) (Table 2 and Supplementary Figure 2).

Table 2.

Predicted LOS by Pandemic Time and Medicare Coverage Type

	Unadjusted	Adjusted		Difference (95% CI)
	All	TM	MA
Stroke
Pre-pandemic	14.63	14.65	14.87	0.22 (0.15 to 0.29)
Mar–May 2020	14.83	14.57	14.52	−0.06 (−0.22 to 0.10)
Jun–Aug 2020	14.93	14.61	14.68	0.07 (−0.07 to 0.20)
Sep–Dec 2020	14.46	13.97	14.02	0.06 (−0.11 to 0.22)
Fracture
Pre-pandemic	13.15	13.24	13.29	0.04 (−0.08 to 0.17)
Mar–May 2020	13.47	13.15	13.19	0.04 (−0.14 to 0.22)
Jun–Aug 2020	13.39	13.25	13.17	−0.09 (−0.27 to 0.10)
Sep–Dec 2020	13.12	12.91	13.08	0.17 (−0.06 to 0.41)
Replacement
Pre-pandemic	10.24	10.32	10.66	0.34 (0.15 to 0.53)
Mar–May 2020	10.54	10.29	10.54	0.25 (−0.20 to 0.70)
Jun–Aug 2020	10.71	10.48	10.59	0.11 (−0.20 to 0.43)
Sep–Dec 2020	10.76	10.22	10.43	0.21 (−0.23 to 0.64)
Cardiac
Pre-pandemic	10.92	10.93	11.36	0.43 (0.27 to 0.59)
Mar–May 2020	11.11	11.04	11.10	0.06 (−0.21 to 0.33)
Jun–Aug 2020	11.20	10.98	11.26	0.28 (0.02 to 0.55)
Sep–Dec 2020	11.03	10.66	10.72	0.06 (−0.27 to 0.39)
Pulmonary
Pre-pandemic	10.98	10.97	11.72	0.75 (0.38 to 1.11)
Mar–May 2020	11.50	11.22	11.59	0.37 (−0.18 to 0.93)
Jun–Aug 2020	11.57	11.30	11.31	0.01 (−0.65 to 0.66)
Sep–Dec 2020	11.17	10.82	11.72	0.90 (−0.02 to 1.82)

Supplementary Fig. 2.

Risk-adjusted average LOS.

Comparing between TM and MA admissions, before the pandemic, MA admissions had longer LOS than TM admissions for 4 conditions (except fracture), ranging from 0.8 days' difference in pulmonary condition to 0.2 days’ difference in stroke (6.8% and 1.5% relative difference out of the average). However, during the pandemic, differences in LOS between the 2 insurance types were no longer significant after adjusting for patient- and facility-level characteristics (Table 2).

Payment per Episode

Before risk adjustment, payment per episode was higher during the pandemic than the previous year for both TM and MA admissions. Stroke and pulmonary conditions had larger unadjusted payment increases ($1700 to $2100 per episode, P < .01) than the other 3 conditions ($700 to $800, P < .01).

After adjusting for patient- and facility-level characteristics, changes in average payments per episode varied by time and Medicare coverage type. During the first 3 months since the pandemic started (March to May 2020), average payment per episode significantly increased in all 5 conditions (by $1500/episode for stroke, $1300 for fracture, $1100 for pulmonary, and $800 to $850 for replacement and cardiac conditions; P < .01) (Table 3 ). During June to August 2020, payment per episode slightly decreased for most conditions and both TM and MA coverage except for cardiac and pulmonary conditions with MA coverage, which experienced even higher average payments than the previous months. By the end of 2020, only stroke and pulmonary conditions had significantly higher average payments per episode than the pre-pandemic period (Table 3 and Supplementary Figure 3).

Table 3.

Predicted Payment ($) per Episode by Pandemic Time and Medicare Coverage Type

	Unadjusted	Adjusted		Difference (95% CI)
	All	TM	MA
Stroke
Pre-pandemic	25,123	26,200	25,839	−361 (−573 to −149)
Mar–May 2020	26,182	27,727	27,075	−651 (−1143 to −159)
Jun–Aug 2020	26,146	27,569	27,072	−497 (−921 to −73)
Sep–Dec 2020	26,844	26,998	26,804	−194 (−721 to 333)
Fracture
Pre-pandemic	22,321	22,609	22,172	−436 (−738 to −135)
Mar–May 2020	23,190	23,906	23,456	−450 (−925 to 25)
Jun–Aug 2020	23,007	23,733	23,153	−580 (−1070 to −91)
Sep–Dec 2020	23,200	22,890	22,133	−757 (−1370 to −144)
Replacement
Pre-pandemic	17,218	17,573	16,490	−1083 (−1378 to −788)
Mar–May 2020	18,255	18,417	17,959	−457 (−1246 to 331)
Jun–Aug 2020	17,585	18,111	17,186	−925 (−1480 to −371)
Sep–Dec 2020	18,056	16,875	15,882	−993 (−1677 to −309)
Cardiac
Pre-pandemic	18,944	19,515	19,170	−345 (−712 to 22)
Mar–May 2020	19,581	20,343	20,218	−125 (−827 to 577)
Jun–Aug 2020	19,785	20,105	20,677	572 (−110 to 1254)
Sep–Dec 2020	19,843	19,637	19,415	−222 (−1038 to 595)
Pulmonary
Pre-pandemic	20,498	21,170	21,115	−55 (−899 to 787)
Mar–May 2020	21,705	22,303	21,844	−459 (−1823 to 904)
Jun–Aug 2020	21,019	21,669	22,727	1058 (−594 to 2710)
Sep–Dec 2020	22,549	21,225	22,533	1308 (−659 to 3276)

Supplementary Fig. 3.

Risk-adjusted average payment ($) per episode.

Comparing between TM and MA admissions, before the pandemic, MA admissions had significantly lower payments per episode for stroke, fracture, and replacement ($25,839 vs $26,200, $22,172 vs $22,609, $16,490 vs $17,573, respectively; P < .01 all 3). During the pandemic, payments for both coverage types had similar changing patterns but different magnitude or timing of changes. By the end of 2020, the significantly lower payments for MA admissions than the TM admissions remained only for fracture and replacement conditions (Table 3).

Functionality Improvement

The average patient functionality improvements for self-care/independence and mobility were not significantly different between TM and MA admissions before the pandemic for all conditions. There were no significant changes during the pandemic either, except for stroke or fracture conditions (Supplementary Tables 1 and 2, Supplementary Figures 4 and 5).

Supplementary Fig. 4.

Risk-adjusted average functional improvements on self-care.

Supplementary Fig. 5.

Risk-adjusted average functional improvements on mobility.

Additional regressions of functional scores at admission on Medicare coverage type and the pandemic timing suggested that the admitted patients had comparable functional status before and during the pandemic period. As a result, the observed changes in functional improvements during the IRF stays for stroke and fracture conditions were not due to the changes in patients’ functional status at admission.

Discharge Locations

Before the pandemic and without risk adjustment, slightly more than 20% of the patients were discharged from IRF to home with self-care (ie, no home health care), more than 50% were discharged to home with care from HHAs, fewer than 20% to SNFs, and approximately 5% to 8% to other places (eg, transfers, long-term care hospitals, hospice). The actual percentages varied across admission conditions. Joint replacement had the highest percentage of discharges to home with self-care (>30%), followed by stroke and cardiac (>20%), and then fracture and pulmonary (<20%). Stroke had the lowest percentage of discharges to home with HHA (48%), whereas the other 4 conditions had approximately 60%. Stroke and fracture had higher percentages of discharges to SNFs (16% to 18%) than the other 3 conditions (5% to 6%) (Table 4 ).

Table 4.

Risk-Adjusted Probabilities of Discharge Locations

	Home		HHA		SNF		Others
	TM	MA	TM	MA	TM	MA	TM	MA
Stroke
Pre-pandemic	25.2	26.7	46.6	48.9	20.2	15.7	8.0	8.7
Mar–May 2020	20.7	19.6	57.2	60.8	13.8	10.6	8.3	9.1
Jun–Aug 2020	22.3	22.4	54.7	57.6	14.4	10.6	8.6	9.4
Sep–Dec 2020	22.5	23.6	52.8	54.4	14.5	11.4	10.2	10.7
Fracture
Pre-pandemic	17.2	19.6	59.5	63.5	17.9	10.8	5.5	6.0
Mar–May 2020	14.9	15.1	69.4	70.5	10.2	8.1	5.5	6.4
Jun–Aug 2020	13.8	15.0	68.6	71.6	12.0	7.2	5.6	6.3
Sep–Dec 2020	14.4	15.0	66.4	71.8	12.5	7.0	6.7	6.3
Replacement
Pre-pandemic	32.1	35.9	58.9	57.0	5.6	4.4	3.4	2.7
Mar–May 2020	27.3	26.6	65.3	64.6	4.8	3.4	2.6	5.4
Jun–Aug 2020	27.3	32.2	65.2	61.4	4.6	4.2	2.9	2.3
Sep–Dec 2020	28.4	29.5	62.9	61.0	4.5	5.2	4.2	4.3
Cardiac
Pre-pandemic	21.8	22.5	61.4	61.3	7.1	4.9	9.8	11.2
Mar–May 2020	18.9	16.6	66.3	69.2	5.4	3.8	9.4	10.4
Jun–Aug 2020	18.2	18.5	65.9	66.0	4.9	3.9	11.0	11.5
Sep–Dec 2020	19.8	18.6	64.7	64.2	4.6	3.5	10.9	13.7
Pulmonary
Pre-pandemic	20.0	18.3	62.1	60.0	6.1	6.4	11.8	15.3
Mar–May 2020	18.1	12.5	65.1	73.3	4.0	3.1	12.7	11.1
Jun–Aug 2020	17.3	20.8	66.0	63.3	5.5	6.6	11.3	9.4
Sep–Dec 2020	20.2	16.2	60.6	64.8	4.9	4.2	14.3	14.9

All values are percentages.

During the pandemic and after adjustment for patient and facility characteristics, there were significantly more discharges to HHAs, fewer to SNF or home, and almost no differences for discharges to other locations. Further, these changes were mainly driven by a few conditions. For example, stroke and fracture were the main drivers for increases in discharges to HHAs. The absolute increases during the pandemic relative to before were 6 to 8 percentage points, representing 12% to 15% relative increases. Stroke and fracture were also the main drivers for decreases in discharges to SNFs. The absolute decreases were 4 to 6 percentage points, roughly 27% to 35% relative decreases. Alternatively, fracture and joint replacement were the main drivers for decreases in discharged to home with self-care. These changes remained by the end of 2020 (Table 4 and Supplementary Figure 6).

Supplementary Fig. 6.

Risk-adjusted percentages of discharge locations.

Comparing between TM and MA, the differences in discharge patterns varied by admission conditions and became smaller during the pandemic. Before the pandemic, MA cases had significantly higher percentages of discharges to home with self-care for stroke, fracture, and joint replacement, and higher percentages of discharges to home with HHA for stroke and fracture. MA cases also had lower percentages to SNF than TM cases for 4 admission types except the pulmonary condition. During the pandemic, these differences became smaller. By the end of 2020, only discharges to SNFs were still significantly different between the 2 insurance types (Supplementary Figure 6).

Discussion

Using a national sample of Medicare beneficiaries, this study estimated the changes in PAC utilization and outcomes among the aging population 65 years or older who were admitted to the IRFs during the COVID-19 pandemic relative to the previous year. The study also compared between beneficiaries who were covered by the TM and those with MA plans. There was a 16% to 34% annualized decrease in IRF admission volume during the period of March to December 2020 across the 5 common admission conditions (stroke, hip fracture, joint replacement, cardiac and pulmonary conditions) compared with the previous year. MA plans experienced less reduction in IRF admissions than the TM during the pandemic. After adjusting for patient and facility characteristics, by the end of 2020, IRF admissions had shorter LOS, higher payment per episode but lower functional improvements than that before the pandemic period for most admission conditions. Discharges to home with home health services increased, whereas discharges to home without home health care and to SNF decreased during the pandemic. Compared with the pre-pandemic period, differences in LOS, payment per episode, and discharge patterns between TM and MA plans became smaller and insignificant during the pandemic period up to December 2020.

This study contributed to the literature by highlighting a few key points. First, changes in PAC utilization and patient composition during the COVID-19 pandemic varied a lot by admission condition. Although the PAC sector showed an average of 51% decline in admission volume and 32% decline in the IRFs since March 2020,²⁶ our study showed that among the 5 most common admission conditions for the Medicare aging beneficiaries, joint replacement and pulmonary conditions were the 2 categories that experienced the most decline (27% to 34%), followed by stroke and cardiac conditions (15% to 16%). Fracture had even a slight increase of 6% in admission volume during the same period. These variations in volume change were partly due to the elective status of the procedures, as well as the need to move (less severe and/or non-COVID) patients from acute care to PAC settings to free up resources in the hospitals for the COVID cases.^{27 , 29} The variation of changes during the pandemic across admission conditions were also reflected in patient composition and IRF characteristics. In addition to the general changing trends, joint replacement and pulmonary conditions were also the 2 admission categories that had the biggest changes in percentages of male, non-White minorities, beneficiaries with MA coverages, severe CMGs, and those using in-unit and/or smaller IRFs compared with the other 3 admission conditions. Furthermore, IRF utilization after hospital discharge could also depend on insurance coverage, patient preferences, and caregiving, and these factors could also be more (or less) influential during the pandemic period than before.

Second, prediction of the changes in average LOS, payment, functional improvement per episode, and the discharge patterns should consider changes in patient composition and severity of the admissions (by CMGs and comorbidity tiers). For example, the average LOS was longer during the pandemic than the pre-pandemic period before any risk adjustment. However, after adjusting for patient demographics and the clinical conditions, results showed that the average LOS was even shorter for some diagnoses (eg, stroke and cardiac conditions). The shorter LOS remained throughout the study period. For payment per episode, there was an increase during the first 3 months of the pandemic (March to May 2020) for all 5 common conditions both before and after risk adjustment. But by the end of the study period, only stroke still had higher average payment than the pre-pandemic period, whereas all the other 4 conditions recovered to their pre-pandemic levels.^{26 , 28 , 46 , 47}

Last, the differences between IRF admissions covered by TM and MA plans varied by time and became smaller during the pandemic compared with the pre-pandemic period. These changes in differences between the 2 plan types over time could be explained by the different coping and administrative strategies of the 2 plan types.^{3 , 9} First, compared with TM, MA plans did not have significant decline in admission volume since the pandemic started. MA patients on average were less sensitive to the pandemic shock than TM patients when it came to the decision on whether to use IRF for PAC after hospital discharge. Further, for LOS, although admissions had shorter average LOS over time for both plan types, admissions covered by the MA plans had a larger variance of LOS during the pandemic than before compared with the TM admissions, which made the 2 plan types insignificantly different from each other by December 2020. In addition, for payment, although the general trends suggested that average payment per episode first increased then decreased (or recovered) during the pandemic, MA plans had a larger magnitude, longer duration (slower recovery), and larger variation of these changes than TM. As a result, the payment differences between TM and MA admissions became smaller and less significant during the pandemic than before. Last, for discharges, before the pandemic, MA plans had higher percentages of discharges to home with or without HHA and less to SNFs than TM. During the pandemic, the TM admissions had larger increases in discharges to home with HHA (and smaller decreases in discharges to home without HHA) than MA admissions. As a result, the differences between TM and MA in discharge patterns were no longer significant except for discharges to SNFs.

Limitation

This study has a few limitations. First, the changing trends in PAC delivery over time and between TM and MA plans did not imply causal relationships or any underlying mechanisms. The findings instead suggested a joint change of a series of beneficiary selections—at the stages of plan enrollment and hospital admission, as well as IRF admission, over the whole study period. The study used a 2-round matching process to control for the sample selection issues between TM and MA plans before and during the pandemic period to the extent that the data allowed. We admitted that the selection issues could be alleviated but cannot be fully controlled. Second, comparisons between TM and MA plans in the study were presented at the national average level. Regional variations such as MA penetration, care system formulation and development, and market consolidation/competitiveness could yield different region-specific findings. Third, this study did not include the COVID-19 outbreak data at the community level due to data limitation on patient or facility location in the assessment data. The varying COVID-19 timing and severity (eg, case outbreaks or deaths) by region could further refine the analysis. Last, the data used in the study did not include information on the COVID-19 status. Yet, the COVID-19 status of each IRF admission was expected to have impacts on the study findings at least for some diagnoses, such as pulmonary conditions.

Conclusions and Implications

This study compared between Medicare beneficiaries who were covered by TM and MA plans for PAC in the IRF settings during the COVID-19 pandemic relative to the pre-pandemic period. The IRF sector had an overall decrease in LOS and increase in payment per episode during the pandemic for both TM and MA plans. Patients were also more likely to be discharged to home with home health services and less likely to home with self-care or to an SNF. However, the differences in these outcomes between TM and MA became smaller during the pandemic than before. In summary, although the 2 Medicare coverage types had different managerial settings, both experienced similar changes in PAC delivery within IRF during the pandemic. By the end of 2020, the differences between TM and MA in LOS, payment per episode, and discharge patterns were smaller than those of 2019.

Supplementary Material

Supplementary Table 1.

Functional Score on Self-Care at Admission and Improvements During the IRF Stay

	Unadjusted Self-Care Score at Admission	Adjusted Self-Care Score Improvements
		TM	MA	Difference (MA-TM) (95% CI)
Stroke
Pre-pandemic	20.65	9.71	9.82	0.11 (−0.01 to 0.23)
Mar–May 2020	19.85	10.19	10.01	−0.18 (−0.45 to 0.09)
Jun–Aug 2020	19.97	9.83	9.88	0.05 (−0.18 to 0.29)
Sep–Dec 2020	19.96	9.29	9.48	0.19 (−0.11 to 0.49)
Fracture
Pre-pandemic	21.03	12.63	12.48	−0.16 (−0.42 to 0.10)
Mar–May 2020	20.12	12.88	12.82	−0.05 (−0.44 to 0.33)
Jun–Aug 2020	20.36	12.55	12.57	0.02 (−0.38 to 0.41)
Sep–Dec 2020	20.41	12.16	12.40	0.24 (−0.27 to 0.75)
Replacement
Pre-pandemic	23.94	13.88	13.78	−0.10 (−0.48 to 0.28)
Mar–May 2020	22.84	13.93	12.65	−1.29 (−2.48 to −0.10)
Jun–Aug 2020	23.02	13.59	14.05	0.46 (−0.21 to 1.13)
Sep–Dec 2020	22.79	13.45	13.50	0.05 (−0.99 to 1.09)
Cardiac
Pre-pandemic	23.30	10.91	10.66	−0.24 (−0.65 to 0.17)
Mar–May 2020	22.58	11.21	10.95	−0.26 (−1.02 to 0.50)
Jun–Aug 2020	22.74	10.52	10.53	0.00 (−0.69 to 0.70)
Sep–Dec 2020	22.70	10.66	9.62	−1.03 (−2.00 to −0.06)
Pulmonary
Pre-pandemic	24.03	9.54	9.05	−0.49 (−1.44 to 0.46)
Mar–May 2020	23.24	9.72	10.52	0.80 (−0.61 to 2.20)
Jun–Aug 2020	23.56	9.42	9.75	0.33 (−1.23 to 1.89)
Sep–Dec 2020	23.03	8.83	8.50	−0.33 (−2.37 to 1.72)

Supplementary Table 2.

Functional Score on Mobility at Admission and Improvements During the IRF Stay

	Unadjusted Mobility Score at Admission	Adjusted Mobility Score Improvements
		TM	MA	Difference (95% CI)
Stroke
Pre-pandemic	34.07	24.16	24.22	0.06 (−0.20 to 0.32)
Mar–May 2020	32.95	24.94	24.95	0.01 (−0.57 to 0.59)
Jun–Aug 2020	32.74	24.38	24.53	0.14 (−0.35 to 0.63)
Sep–Dec 2020	32.92	23.36	23.63	0.28 (−0.35 to 0.90)
Fracture
Pre-pandemic	27.83	31.70	30.83	−0.87 (−1.43 to −0.32)
Mar–May 2020	26.44	33.09	32.54	−0.54 (−1.37 to 0.29)
Jun–Aug 2020	26.97	32.39	32.18	−0.22 (−1.09 to 0.66)
Sep–Dec 2020	27.14	30.27	30.25	−0.02 (−1.10 to 1.06)
Replacement
Pre-pandemic	34.30	38.53	38.05	−0.49 (−1.38 to 0.41)
Mar–May 2020	32.54	38.30	35.18	−3.12 (−5.76 to −0.48)
Jun–Aug 2020	32.64	38.11	38.52	0.41 (−1.24 to 2.07)
Sep–Dec 2020	32.31	36.51	37.47	0.96 (−1.46 to 3.37)
Cardiac
Pre-pandemic	37.09	28.73	28.27	−0.46 (−1.30 to 0.38)
Mar–May 2020	35.99	29.13	28.85	−0.27 (−1.80 to 1.25)
Jun–Aug 2020	35.54	28.17	28.08	−0.09 (−1.49 to 1.31)
Sep–Dec 2020	35.92	27.98	26.16	−1.82 (−3.73 to 0.08)
Pulmonary
Pre-pandemic	38.21	25.89	23.98	−1.90 (−3.73 to −0.08)
Mar–May 2020	36.39	26.70	27.68	0.98 (−1.73 to 3.69)
Jun–Aug 2020	36.89	25.64	24.43	−1.21 (−4.46 to 2.05)
Sep–Dec 2020	36.25	24.98	24.78	−0.21 (−4.15 to 3.73)