Impact of a Comprehensive Transitional Care Management Model on Use of Community‐Based Rehabilitation After Stroke

Abstract

Background

Community‐based physical and occupational therapy (PT/OT) are critical for stroke recovery but are underused. We conducted a secondary analysis of the COMPASS (Comprehensive Post‐Acute Stroke Services) study, a pragmatic trial of comprehensive postacute transitional care (TC) to investigate whether TC programs increase PT/OT use.

Methods

Forty hospitals were randomized to implement COMPASS‐TC or maintain usual care for patients with stroke/transient ischemic attack. In a crossover phase, usual care hospitals implemented COMPASS‐TC. We linked participants to administrative claims to assess PT/OT use after stroke. Adjusted generalized estimating equations compared COMPASS‐TC to usual care within the trial and crossover cohorts on 30‐/90‐day PT/OT use, time to first visit, number of visits, and receipt of PT and OT versus single therapy. Per protocol analysis estimated complier average causal effects.

Results

COMPASS enrolled 8377 patients from July 2016 to March 2019; 5261 were linked to administrative claims. Thirty‐day PT/OT ranged from 22.6% in usual care to 37.5% in COMPASS‐TC. Therapy use was similar between groups in the trial cohort, and COMPASS‐TC was associated with increased use in crossover analysis (9.4% [95% CI, 5.6–13.3%] at 30 days). COMPASS‐TC was consistently associated with a shorter time to therapy (mean difference, −0.16 [95% CI, −0.03 to −0.29]). Per protocol results were larger for most outcomes.

Conclusions

COMPASS‐TC was associated with shorter time to PT/OT and with greater therapy receipt in the crossover, but not the trial, analysis. Inconsistencies may reflect confounding or differences in hospitals that chose to adopt the intervention in Phase 2. Implementation studies to improve care transitions after stroke are needed to enhance use of postacute rehabilitation.

Nonstandard Abbreviations and Acronyms

CMS

Centers for Medicare and Medicaid Services

COMPASS

Comprehensive Post‐Acute Stroke Services

COMPASS‐TC

Comprehensive Post‐Acute Stroke Services Transitional Care

NC

North Carolina

OT

occupational therapy

TCM

transitional care model

Research Perspective.

What Is New?

• We found evidence that a comprehensive postacute transitional care model versus usual care was associated with increased use, timeliness, and intensity of community‐based physical and occupational therapy after stroke and evidence to suggest that greater adherence to the model improved its effectiveness.

• Use of community‐based physical and occupational therapy among survivors of stroke discharged home overall was low (23% in usual care hospitals), possibly due to gaps in referral, insufficient health insurance coverage, lack of transportation, or patient and family preferences.

What Question Should Be Addressed Next?

• Future work should investigate how to implement transitional care interventions in clinical practice with greater fidelity and specifically test model components aimed at enhancing patient participation in postacute community‐based rehabilitation.

Stroke is a major cause of long‐term disability and a complex condition with high comorbidity, associated health care costs, and hospital readmission rates. ¹ , ² , ³ , ⁴ Over 9 million adults in the United States are currently living with the effects of a stroke, and ∼795 000 people experience a new or recurrent stroke each year. ⁵ Many patients with stroke who are discharged home have residual physical and cognitive deficits and often have unmet social needs. ⁶ These deficits, unmet needs, and other behavioral factors can hinder stroke recovery. ⁷

Coordinating health care in the community as the patient with stroke transitions from hospital to home can help the patient better manage the complexities of their condition and recovery. Community‐based rehabilitation services, including home health and outpatient physical therapy (PT) and occupational therapy (OT), are critical and effective for survivors of stroke and transient ischemic attack discharged home with impairments in mobility, activities of daily living (eg, feeding, bathing, personal hygiene), or cognition. ⁸ Clinical practice guidelines and hospital quality metrics underscore the critical importance of assessing rehabilitation needs and making timely referrals. ⁸ , ⁹ Timely rehabilitation is consistently associated with improved functional outcomes and quality of life and reduced disability and secondary complications. ¹⁰ , ¹¹ Early initiation of rehabilitation can prevent further functional decline and decrease the risk of adverse events such as falls and immobility, thereby supporting optimal recovery. ¹² Unfortunately, poststroke rehabilitation in the community remains underused. ¹³ The reasons for this are complex and are driven by multifactorial barriers, including person‐level factors that limit access to services, personal preferences and motivation, hospital‐related challenges in the evaluation for and referral process to rehabilitation, and the coordination of these services after stroke. ¹⁴ Improving access to rehabilitation for patients with stroke is vital as set forth in a recent policy statement from the American Heart Association. ¹⁵

Effective coordination of posthospital care is key to improving outcomes and reducing health care costs but has been a persistent challenge in the US health care system. ¹⁶ , ¹⁷ To address this gap, in 2013 the Centers for Medicare and Medicaid Services (CMS) introduced transitional care management (TCM) reimbursement policies for hospital‐to‐home transitional care services for complex patients. Whether TCM improves use of rehabilitation in US‐based health systems is unclear. TCM entails coordinating patient care across providers for up to 30 days after hospital discharge. To qualify for reimbursement by CMS, TCM must include both early telephone follow‐up and an in‐person visit with an advanced practice provider or physician. This visit must assess rehabilitation needs and include appropriate communication and coordination with rehabilitation providers. Although evidence from international settings suggest that TCM programs can improve functional outcomes, ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² , ²³ overall uptake of TCM in the United States remains poor, with wide variability across hospitals and few studies demonstrating clear effectiveness. ²⁴ , ²⁵ , ²⁶ , ²⁷

The COMPASS (Comprehensive Post‐Acute Stroke Services) study was the first pragmatic trial to compare the effectiveness of comprehensive TCM versus usual care in improving functional status among survivors of stroke. The COMPASS‐TC (COMPASS Transitional Care) intervention was designed to explicitly incorporate rehabilitation professionals into the acute care discharge planning process and to promote early referral to therapy during postacute follow‐up. ²⁸ , ²⁹ , ³⁰ The trial found no differences in functional status between arms but significantly increased use of preventive behaviors (home blood pressure monitoring) and ambulatory care use. Rehabilitation therapy use was not an end point for the trial, which focused on 90‐day functional impact and readmissions. In this analysis, we evaluate the degree to which the COMPASS‐TC intervention, versus usual care, was associated with increased rehabilitation use, timeliness, type, and intensity in a subset of trial participants linked to claims data. We also explore whether these effects vary between patients with stroke and TIA and across insurance types.

Methods

Data Availability

In order to minimize the possibility of unintentionally sharing information that can be used to reidentify private information, a subset of the data generated for this study are available at the Patient‐Centered Outcomes Data Repository and can be accessed at https://www.icpsr.umich.edu/web/pcodr/studies/38185. Because of the sensitive nature of the administrative claims data used for this study, requests to access these data from qualified researchers trained in human subject confidentiality protocols may be sent to the Research Data Assistance Center at www.resdac.org.

Institutional Review Board Approval

This study was reviewed by the Institutional review board at the University of Pittsburgh and was determined to be exempt as it was a secondary analysis of data from a prior study. The parent study (COMPASS Study) met criteria for a waiver of consent and Health Insurance Portability and Accountability Act authorization. Eligible patients were enrolled at hospital discharge without consent. At 90 days post discharge, patients or their proxies provided verbal informed consent over the telephone for collection of outcomes.

Study Design and Population

This study is a secondary analysis of the COMPASS study, a cluster‐randomized pragmatic trial conducted in North Carolina (NC) between July 2016 and March 2019. ²⁹ The study examined the effectiveness of COMPASS‐TC versus usual care and enrolled 11 601 racially, socioeconomically, and geographically diverse adults with stroke or TIA who were discharged home after hospitalization. ²⁹ In Phase 1, 40 hospitals were randomized to implement COMPASS‐TC or maintain their usual care for patients with stroke (participants enrolled in this phase are referred to as the trial cohort; Figure 1). Randomization was stratified according to certification as a comprehensive or primary stroke center (yes, no) and stroke census volume (high, low). Hospitals were assigned 1:1 to intervention within each of the 4 resultant strata. In Phase 2, 16 of the hospitals assigned to usual care switched to implement COMPASS‐TC (participants enrolled in these hospitals are referred to as the crossover cohort). Transition to Phase 2 occurred in waves, determined by date of randomization and achievement of sample size targets for Phase 1. Average duration of Phase 1 was 14 months (range 8–20) and 12 months (range 8–16) for Phase 2. This study performs comparative analyses of both the trial and the crossover cohort.

Figure 1. COMPASS study design overview.

Numbers in parentheses indicate participants included in secondary analysis. Hospitals were randomized to COMPASS‐TC or usual care in Phase 1. In Phase 2, usual care hospitals crossed over to implement the COMPASS‐TC intervention. Comparative effectiveness of COMPASS‐TC vs usual care was assessed within (1) the trial cohort, comparing patient outcomes from hospitals randomized to COMPASS‐TC (n=20 hospitals, with n=19 enrolling patients) vs usual care (n=20 hospitals); and (2) the crossover cohort, comparing patient outcomes within the set of usual care hospitals that implemented COMPASS‐TC in Phase 2 to patient outcomes from usual care in Phase 1 (n=16 hospitals). Data from sustainability (Phase 2 among hospitals initially randomized to COMPASS‐TC) were not used in this analysis. COMPASS‐TC indicates Comprehensive Post‐Acute Stroke Services Transitional Care.

Baseline patient demographic and clinical characteristics were collected by nursing staff via medical record abstraction and patient self‐report. For this study, we linked COMPASS study data for NC residents to administrative claims data including Blue Cross Blue Shield of NC, Medicare fee‐for‐service, Medicare Advantage, and NC Medicaid to assess rehabilitation use in the year following stroke. Participants residing outside of NC (n=324) were not included because several of our databases (provider supply and claims data) were available only for NC. Linkages to Medicare claims were performed via deterministic, iterative protocols using patient sex, date of birth, index discharge hospital, and dates of service. Linkages to NC Medicaid and Blue Cross Blue Shield of NC were performed at the beneficiary level using first, last, and middle name; sex; and date of birth. Details are available elsewhere. ³¹ , ³² , ³³ Claims data were available for 5261 of the 8053 enrolled participants comprising the trial and crossover cohorts (65%; see Figure S1 for detailed flow of participants). Reasons for not linking to administrative claims include lack of insurance (~10% of patients were uninsured); coverage by payors not available to our study (eg, Veteran's Administration, private payors); and potential failed linkage (eg, due to errors in data elements used for linkage). Table S1 shows patient characteristics according to linkage.

Our goal was to assess the comparative effectiveness of COMPASS‐TC versus usual care on rehabilitation use, timeliness, intensity, and therapy type (receipt of PT and OT versus only 1 type) and to explore modification of these effects by diagnosis (stroke, TIA) and insurance type. We examined these questions in both the trial cohort (parallel comparison of patients from hospitals randomized in Phase 1) and the crossover cohort (pre‐post comparison within the same hospitals of patients in Phase 2 versus Phase 1). Participants who did not have continuous coverage through 30 or 90 days, depending on the outcome, were excluded (<5% of the analytic sample, see Figure S1 for details).

The COMPASS‐TC Intervention

Details of the COMPASS‐TC intervention have been published. ²⁸ , ²⁹ Briefly, the intervention was designed to be consistent with the TCM management and reimbursement policies of CMS and was delivered by multidisciplinary teams at each site. Discharge planning encouraged physician follow‐up, referral to rehabilitation, and sharing of patient information across settings. The intervention was designed to include early telephone follow‐up (ideally within 2 business days) and an in‐person follow‐up clinic visit targeted to take place within 7 to 14 days as per TCM policy. During the 2‐day telephone follow‐up, nursing staff conducted standardized assessments of new stroke symptoms, medication reconciliation, and assessments of referral and receipt of home health and outpatient rehabilitation. In‐person follow‐up visits were delivered by an advanced practice provider (eg, nurse practitioner) or neurologist. Patients underwent comprehensive evaluation of medical, functional, psychological, and social needs. Table 1, ²⁹ summarizes assessments at these time points. Participants received an electronic personalized care plan (eCare plan) based on identified needs that was shared across postacute providers and included referrals to rehabilitation and community services. The eCare plan was organized around 4 messaging domains: (1) Know your Numbers, focused on risk factor knowledge and monitoring of blood pressure; (2) Engage Mind and Body, focused on falls prevention, physical activity and recovery, and cognition; (3) Willingness, focused on medication adherence and behavior change; and (4) Support, focused on caregiver, social, transportation and emotional support.

Table 1.

Domains Assessed in Postdischarge Follow‐Up After Stroke ²⁹

Postdischarge follow‐up call
Medication reconciliation	Home health or outpatient services
New or worsening symptoms	Caregiver assistance
Falls	PCP follow‐up appointment
Transportation	Stroke clinic follow‐up appointment
Clinic visit: poststroke functional assessment
Medication management	Spasticity
Financial barriers to medication management	Social support
Cognition	Physical mobility and safety
Depression	General health
Health literacy	Upper extremity
Access to PCP	Transportation
Emergency department visits or hospital readmissions	General health
Status of advance directive	Falls
Stress	ADL/instrumental ADL
Clinic visit: caregiver assessment
Caregiver assistance	Caregiver stress
Caregiver health
Clinic visit: advanced practice provider assessment
Lifestyle management (alcohol, smoking, drugs)	Risk factor management (blood pressure, low‐density lipoprotein, international normalized ratio (prothrombin time); glycated hemoglobin)
Modified Rankin Scale	Depression
Cognition	Communication
Physical activity

Reproduced from Duncan et al. ²⁹ under the terms and conditions of the Creative Commons Attribution (CC‐BY) license (https://creativecommons.org/licenses/by/4.0/).

ADL indicates activities of daily living; and PCP, primary care provider.

Outcomes

Rehabilitation use was abstracted from outpatient, carrier, and home health insurance claim files based on the revenue codes or CMS accepted Current Procedural Terminology/Healthcare Common Procedure Coding System codes. ³¹ The primary outcome of the analysis was receipt of any community‐based rehabilitation (any PT, OT, or both) in the home or outpatient setting within 30 and 90 days of discharge. To further explore use, we evaluated a series of secondary outcomes among individuals who received therapy within 90 days: (1) time to first therapy visit, (2) receipt of both PT and OT versus single‐type therapy, and (3) intensity of rehabilitation therapy (number of visits).

Covariates

Covariates were included in select models that were either strongly prognostic or meaningfully imbalanced by treatment arm. Imbalance was observed first because cluster randomization does not guarantee balance on individual‐level covariates and second because our sample was further restricted to a subset of enrolled participants for whom we had administrative claims data. Selected covariates were age (linear, quadratic), sex, race (White, Black, other [see Table 2]), insurance type (fee‐for‐service, Medicare Advantage, Medicaid only, Blue Cross Blue Shield of NC, dual Medicare‐Medicaid), rurality (metropolitan, micropolitan, smalltown/rural), stroke type (stroke, TIA), aphasia (yes, no), National Institutes of Health Stroke Scale score (0, 1–4, 5–42), admission status (emergency department visit/observation stay, inpatient stay), ambulatory status at discharge (ambulates independently, unable to ambulate/ambulates with assistance), history of stroke or TIA, county‐level home health supply (quartile), county‐level PT supply (quartile), and need for rehabilitation at discharge (yes, no). Data on county‐level provider supply were obtained from the NC Health Profession Database and the Area Health Resources file and were linked to participant data according to county of residence. ³⁴ , ³⁵ Need for rehabilitation was derived from data that nurses abstracted as part of the trial. Specifically, nurses indicated whether the patient needed postacute rehabilitation and whether the patient was referred to these services in the home or outpatient setting.

Table 2.

Characteristics of Hospitals and Patients by Treatment Group

Characteristic	Trial cohort		Crossover cohort*
	COMPASS‐TC intervention	Usual care	COMPASS‐TC intervention	Usual care
Hospital characteristics
Number of hospitals	19†	20	16
Joint Commission certified comprehensive or primary stroke center	12 (63.2)	11 (55.0)	9 (56.3)
Academic affiliation—n (%)a	3 (15.8)	5 (25.0)	3 (18.8)
Critical access hospital—n (%)	2 (10.5)	2 (10.0)	2 (12.5)
Annual stroke volume (2013)
<100	4 (21.1)	5 (25.0)	3 (18.8)
100–299	9 (47.4)	9 (45.0)	8 (50.0)
300+	6 (31.6)	6 (30.0)	5 (31.3)
Urban–rural classification—n (%)
Metropolitan (population ≥ 50 000)	9 (47.4)	12 (60.0)	10 (62.5)
Micropolitan (population 10 000‐49 999)	7 (36.8)	7 (35.0)	5 (31.3)
Small town or rural (population <10 000)	3 (15.8)	1 (5.0)	1 (6.3)
Ownership
For profit	2 (10.5)	1 (5.0)	1 (6.3)
Not for profit	9 (47.4)	9 (45.0)	9 (56.3)
Other	8 (42.1)	10 (50.0)	6 (37.5)
Hospital has stroke team	13 (68.4)	12 (60.0)	10 (62.5)
Physical therapist and occupational therapist on discharge planning team	11 (57.9)	8 (40.0)	7 (43.8)
Affiliated home health agency	3 (15.8)	3 (15.0)	3 (18.8)
Physical therapy/occupational therapy services provided by staff	14 (73.7)	14 (70.0)	11 (68.8)
Hospital delivers outpatient rehab services	11 (57.9)	14 (70.0)	10 (62.5)
Participant characteristics
Number of participants	1855	2077	1329	1849
Age, y, mean±SD	71.4 (12.44)	70.2 (12.87)	69.6 (12.69)	70.3 (12.96)
Sex, n (%)
Male	931 (50.2)	925 (44.5)	663 (49.9)	804 (43.5)
Female	924 (49.8)	1152 (55.5)	666 (50.1)	1045 (56.5)
Race, n (%)
White	1491 (80.9)	1465 (71.1)	896 (67.9)	1287 (70.2)
Black	312 (16.9)	551 (26.7)	378 (28.7)	503 (27.4)
Other‡	41 (2.2)	45 (2.2)	45 (3.4)	43 (2.3)
Missing	11	16	10	16
Urban–rural classification—n (%)
Small town or rural (population <10 000)	283 (15.3)	125 (6.0)	62 (4.7)	98 (5.3)
Micropolitan (population 10 000–49 999)	487 (26.3)	296 (14.3)	132 (9.9)	179 (9.7)
Metropolitan (population ≥50 000)	1084 (58.5)	1656 (79.7)	1135 (85.4)	1572 (85.0)
Missing	1	0	0	0
County PT supply per 10 000 population—n (%)
1st quartile (≤2.68)	215 (11.6)	143 (6.9)	358 (9.1)	117 (6.3)
2nd quartile (2.69–4.42)	350 (18.9)	444 (21.4)	794 (20.2)	406 (22.0)
3rd quartile (4.43–6.59)	322 (17.4)	690 (33.2)	1012 (25.7)	549 (29.7)
4th quartile (>6.59)	968 (52.2)	800 (38.5)	1768 (45.0)	777 (42.0)
County home health supply per 1000 population—n (%)
1st quartile (≤0.017)	818 (44.1)	1321 (63.6)	2139 (54.4)	1293 (69.9)
2nd quartile (0.018–0.028)	452 (24.4)	454 (21.9)	906 (23.0)	286 (15.5)
3rd quartile (0.029–0.044)	459 (24.7)	192 (9.2)	651 (16.6)	163 (8.8)
4th quartile (>0.044)	126 (6.8)	110 (5.3)	236 (6.0)	107 (5.8)
Insurance—n (%)
Medicare fee‐for‐service	916 (49.4)	1017 (49.0)	439 (33.0)	911 (49.3)
Medicare Advantage	488 (26.3)	481 (23.2)	495 (37.2)	425 (23.0)
Medicaid	157 (8.5)	221 (10.6)	135 (10.2)	191 (10.3)
Blue Cross Blue Shield	129 (7.0)	139 (6.7)	134 (10.1)	123 (6.7)
Dual coverage	165 (8.9)	219 (10.5)	126 (9.5)	199 (10.8)
Has a primary care provider, n (%)	1696 (91.4)	1866 (89.8)	1216 (91.5)	1652 (89.3)
Rehabilitation needed or referred at discharge, n (%)	789 (42.5)	819 (39.4)	627 (47.2)	715 (38.7)
National Institutes of Health Stroke Scale score, n (%)
0	702 (38.2)	725 (36.2)	401 (30.7)	651 (36.5)
1–4	872 (47.4)	977 (48.7)	615 (47.1)	875 (49.1)
5–42	266 (14.5)	303 (15.1)	290 (22.2)	257 (14.4)
Missing	15	72	23	66
Stroke diagnosis, n (%)
Stroke	1173 (63.2)	1284 (61.8)	1016 (76.4)	1121 (60.6)
TIA	682 (36.8)	793 (38.2)	313 (23.6)	728 (39.4)
Aphasia at presentation	408 (22.0)	526 (25.3)	323 (24.3)	494 (26.7)
Inpatient admission	1451 (78.2)	1645 (79.2)	1136 (85.5)	1487 (80.4)
Ambulatory status at discharge, n (%)
Ambulate independently	1438 (90.6)	1620 (90.7)	1172 (93.6)	1443 (90.6)
With assistance/unable to ambulate	149 (9.4)	166 (9.3)	80 (6.4)	149 (9.4)
Missing	268	291	77	257
Medical history and comorbidity
Previous stroke or TIA	542 (29.2)	670 (32.3)	361 (27.2)	605 (32.7)
Diabetes	660 (35.6)	754 (36.3)	489 (36.8)	664 (35.9)
Heart failure	144 (7.8)	200 (9.6)	108 (8.1)	183 (9.9)
Myocardial infarction or coronary artery disease	402 (21.7)	447 (21.5)	284 (21.4)	403 (21.8)
Treatment adherence
Receipt of 2‐d call	1338 (72.1)	NA	977 (73.5)	NA
Receipt of clinic visit and eCare plan	665 (35.8)	NA	496 (37.3)	NA

COMPASS‐TC indicates Comprehensive Post‐Acute Stroke Services Transitional Care; NA, not applicable; and TIA, transient ischemic attack.

^*Crossover analysis is carried out among the 16 hospitals randomized to usual care that crossed over to the COMPASS‐TC intervention in Phase 2.

^† Characteristics of hospitals are shown for the 19 of 20 randomized to the intervention that enrolled at least 1 patient in the study; 1 small critical access hospital did not have patients with stroke in the study and contributed no data to the analysis.

^‡ Racial groups included in this category are American Indian/Alaskan Native (n=30 trial cohort, n=27 crossover cohort); Asian (n=12, n=11); Native Hawaiian/Pacific Islander (n=0, n=1); multi‐race (n=12, n=10); and individuals who self‐described as other race (n=32, n=39).

Statistical Analysis

We performed primary analyses as intention‐to‐treat (ITT) such that all participants enrolled at a given hospital and for whom claims were available were analyzed according to the hospital's treatment assignment. Effects of COMPASS‐TC versus usual care were estimated in the trial (participants enrolled in Phase 1) and crossover (participants enrolled in Phase 1 and 2 within the set of usual care hospitals that implemented COMPASS‐TC in Phase 2) cohorts. We used generalized estimating equations to evaluate absolute differences in proportions for binary outcomes (ie, rehabilitation receipt within 30/90 days of discharge, receipt of dual therapy) using models fit with an identity link to estimate marginal effects. Days to first therapy visit and number of visits were log transformed before fitting with linear generalized estimating equations. Models were clustered by hospital to account for clustering by design.

We fit 4 models for each outcome in the ITT analysis. The first model was unadjusted and accounted only for stratification variables used for randomization (ie, stroke certification status and annual stroke volume). The second model, our primary analysis, was adjusted for the set of covariates described previously. Because referral to therapy may have been enhanced by the intervention, and not simply a reflection of differences in underlying need in the population, we fit a third model that omitted this covariate as a sensitivity analysis. In the fourth model, we further adjusted model 2 for hospital‐level variables as a sensitivity analysis. Finally, we explored interactions between the intervention and insurance type and stroke type in the primary ITT analysis. These were defined a priori as potential modifiers of COMPASS‐TC, first, because copays and caps on services that vary across payors may influence use and effectiveness of TCM and second, because diagnosis with stroke versus TIA reflects a difference in severity and potentially the participant's perceived need for therapy.

We performed per protocol secondary analyses to evaluate the effect of COMPASS‐TC receipt (defined as receiving an eCare plan at in‐person follow‐up within 30 days) on outcomes of interest. For these analyses, in keeping with the approach for the trial analysis, we estimated the complier average causal effect, which is the effect of COMPASS‐TC among those who would adhere to treatment recommendations. Details of this approach can be found in the statistical supplement to the primary trial paper. ³⁶ , ³⁷ Briefly, we employed an instrumental variables approach to complier average causal effect estimation using 2‐stage least squares regression. Randomization arm served as the instrument for this analysis as it is associated with intervention receipt and is hypothesized to have no effect on the outcomes of interest except through receipt of TC. We adjusted for the covariate set used in model 2 ITT analysis, which was our primary analysis model.

Covariates of interest were sometimes missing (as shown in Table 1). Missing data were addressed using multiple imputation with chained equations to construct 100 complete data sets via the fully conditional specification method. These data sets were analyzed as described and resultant estimates were combined using standard techniques. ³⁷ , ³⁸ All statistical tests were 2 sided with an alpha level of 0.05. SAS Version 9.4 was used for all analyses.

Results

Characteristics of hospitals participating in each study phase are shown in Table 2. There were slightly more Joint Commission certified stroke centers, hospitals with a stroke team, and hospitals with a physical therapist/occupational therapist on staff as part of the discharge planning team in the COMPASS‐TC arm versus usual care. Four of the hospitals randomized to usual care declined to continue into Phase 2 COMPASS‐TC. Characteristics of hospitals that continued to Phase 2 were largely similar to those randomized to usual care in Phase 1, although there was less representation of low‐volume hospitals. There were also minimal differences between the characteristics of patients in Phase 2 COMPASS‐TC and usual care hospitals (Table 2), suggesting that patients enrolled at hospitals that continued to Phase 2 were not different with regard to stroke severity or need for rehabilitation.

The mean age of study participants was 70.5 years, and 52% were female. In the trial cohort, there was a higher proportion of women, individuals who were Black, and residents of metropolitan areas in the usual care relative to the intervention arm (Table 2). The higher proportion of women in the usual care arm was also seen in the crossover cohort. In the crossover cohort there were fewer individuals insured through Medicare fee‐for‐service versus Medicare Advantage in the intervention arm than other groups. Need for rehabilitation services determined at discharge was higher in COMPASS‐TC versus usual care arm in both cohorts (42.5% versus 39.4% in the trial cohort; 47.2% versus 38.7% in the crossover cohort). Stroke severity as measured with the National Institutes of Health Stroke Scale was similar across groups, although a larger proportion of participants in the crossover COMPASS‐TC group had scores ranging from 5 to 42, likely reflecting the higher proportion of patients with stroke in this group.

Receipt of the intervention was modest and not appreciably different between the crossover and trial cohorts (Table 2). In the trial cohort, 72.1% of patients received telephone follow‐up and 35.8% received a follow‐up clinic visit and eCare plan. Adherence rates were similar, 73.5% and 37.3%, respectively, in the crossover cohort.

Receipt of rehabilitation therapy within 30 days of discharge was 22.9% in usual care, 25.1% in COMPASS‐TC in the trial cohort, and 37.9% in crossover COMPASS‐TC (Table 3). Ninety‐day receipt across these groups was 22.9% in usual care, 30.5% in COMPASS‐TC, and 43.8% in crossover COMPASS‐TC. Among patients who received therapy within 90 days, the average time to first therapy visit was shortest in the COMPASS‐TC arm of the crossover cohort (14.3 days versus 16.4–19.4 days across other groups). Mean number of visits was similar across groups, ranging from 9.4 to 10.0. Finally, receipt of both PT and OT as opposed to a single therapy ranged from 44.2% in usual care to 52.2% in crossover COMPASS‐TC.

Table 3.

Rehabilitation Service Receipt, Initiation Timing, Intensity, Setting, and Type

	Trial cohort		Crossover cohort
	COMPASS‐TC	Usual care	COMPASS‐TC	Usual care
Population with 30‐d follow‐up	1832	2047	1300	1823
Receipt of PT or OT by d 30, n (%)	460 (25.1)	469 (22.9)	492 (37.9)	412 (22.6)
Population with 90‐d follow‐up	1775	1985	1268	1773
Receipt of PT or OT by d 90, n (%)	542 (30.5)	588 (29.6)	555 (43.8)	523 (29.5)
Population receiving at least 1 PT/OT visit	542	588	555	523
Time to first therapy service within 90 d
Days (mean±SD	16.4 (21.0)	19.0 (22.3)	14.3 (18.8)	19.4 (22.8)
Median d (25th, 75th percentile)	7.0 (3.0, 20.0)	8.0 (3.0, 27.0)	6.0 (3.0, 17.0)	8.0 (3.0, 27.0)
Number of PT and OT visits within 90 d
Visits, mean±SD	9.5 (8.3)	9.6 (9.2)	10.0 (8.6)	9.4 (9.3)
Median visits (25th, 75th percentile)	8.0 (3.0, 13.0)	7.0 (3.0, 13.0)	8.0 (3.0, 15.0)	7.0 (3.0, 13.0)
Type of therapy services within 90 d, n (%)
PT only	264 (48.7)	294 (50.0)	235 (42.3)	268 (51.2)
OT only	34 (6.3)	27 (4.6)	25 (4.5)	24 (4.6)
Dual therapy (PT and OT)	244 (45.0)	267 (45.4)	295 (52.2)	231 (44.2)

COMPASS‐TC indicates Comprehensive Post‐Acute Stroke Services Transitional Care; OT, occupational therapy; and PT, physical therapy.

In adjusted ITT analyses of the trial cohort, no treatment effect was detected for therapy use at 30 or 90 days, nor were there differences in therapy type, number of therapy visits, or receipt of dual therapy (Figure 2). There was a 14.5% shorter time to first therapy visit relative usual care arm (mean difference in log‐days, −0.157 [95% CI, −0.287 to −0.027]). In contrast, in ITT analyses of the crossover cohort, COMPASS‐TC was associated with an absolute 9.4% higher use of rehabilitation at 30 days (95% CI, 5.6%–13.3%) and 8.9% higher use at 90 days (95% CI, 5.0%–12.9%). Time to first therapy was 20% shorter relative to usual care, with a mean difference in log‐days of −0.181 (95% CI, −0.374 to 0.013).

Figure 2. Effect of COMPASS‐TC intervention versus usual care on rehabilitation use, timeliness, and intensity, intention‐to‐treat, and per protocol results.

Estimands are the absolute difference in proportion for binary outcomes. For time to therapy and intensity, estimands represent the difference in log‐days to first therapy and difference in log‐number of visits, respectively. Estimates in the trial cohort are adjusted for age, sex, race, rurality, insurance, aphasia, diagnosis type, National Institutes of Health Stroke Scale score, history of stroke or transient ischemic attack, need for rehabilitation at discharge, ambulatory status at discharge, and area home health and physical therapy supply and are clustered by hospital. COMPASS‐TC indicates Comprehensive Post‐Acute Stroke Services Transitional Care; OT, occupational therapy; and PT, physical therapy.

Per protocol results, shown in Figure 2, represent the estimated effects of receipt of COMPASS‐TC (defined as receiving an eCare plan at a clinic visit by day 30). Estimands were larger in magnitude across all outcomes than in the ITT analyses and the pattern of differences between the trial and crossover cohorts is similar. There was an absence of significant associations in the trial cohort, except for time to therapy. In crossover analyses, COMPASS‐TC was associated with significantly greater therapy use, shorter time to therapy, and higher number of visits compared with usual care. The probability of 30‐day therapy use was 28.9% higher for COMPASS‐TC than usual care (95% CI, 19.5–38.4%). COMPASS‐TC decreased time to receipt of therapy by 56.2% (difference in log‐days, −0.576 [95% CI, −0.100 to −1.052]) and increased number of visits by 31.0% (difference in log number visits, 0.270 [95% CI, 0.015–0.525]) compared with usual care. In sensitivity analysis, we explored whether additional adjustment for hospital‐level characteristics such as having a stroke team, presence of a physical therapist or occupational therapist on the discharge planning team, and whether the hospital had an affiliated home health agency impacted effect estimates. Results were not meaningfully different (see Model 4 results in Table S2). We also performed analyses with a modified patient‐level covariate set, omitting need or referral to therapy at discharge as this could theoretically be affected by the intervention. Results were slightly larger in magnitude (ie, away from the null) for all analyses compared with our primary models but did not change interpretation of findings. Finally, we evaluated whether ITT effects of COMPASS‐TC varied according to stroke diagnosis or insurance type. No differences were seen across most outcomes; there was consistent evidence that the effect of the intervention on use of therapy was stronger in patients with stroke than with TIA (Tables S3 and S4). For example, the difference in 90‐day receipt was 11.0% (95% CI, 6.6%–15.4%) among patients with stroke and 3.3% (95% CI, −0.5% to 8.2%) among patients with TIA in the crossover analysis.

Discussion

In this analysis of data from the COMPASS Study, a pragmatic trial of 40 diverse hospitals across NC, we compared rehabilitation use outcomes between hospitals randomized to COMPASS‐TC or usual care (trial cohort) and post‐ versus preadoption of COMPASS‐TC among hospitals initially assigned to usual care (crossover cohort). To our knowledge, this study is the first to evaluate the impact of transitional care on community‐based rehabilitation among survivors of stroke. Receipt of rehabilitation at 30 and 90 days post discharge was ~9% higher in patients receiving COMPASS‐TC, but this finding was observed only in the crossover cohort. Among patients who received rehabilitation, we found that those who received COMPASS‐TC compared with usual postacute care experienced more timely receipt of community‐based rehabilitation. This effect was observed in both trial and crossover analyses. We also observed some support in the crossover analyses for more therapy visits and more comprehensive use of therapy (ie, both PT and OT) associated with COMPASS‐TC. Taken together, these findings suggest that comprehensive transitional care may improve the use, timeliness, and intensity of PT and OT after stroke, although compliance to COMPASS‐TC was suboptimal.

Rehabilitation Use

Our results were mixed on whether the intervention was associated with differences in rehabilitation use within 30 or 90 days. Although there is little evidence to date on the impact of TC interventions on therapyuse, a 2018 Medicare claims study found that patients with stroke seen at hospitals with Joint Commission accreditation and higher nurse staffing levels were more likely to receive continuity of therapist care into the community, suggesting possible effectiveness of transitional care components such as prehospital triage, systematization of assessment for rehabilitation need, and coordination of stroke care. ³⁹

Among hospitals that crossed over from usual care to the intervention, there were notable differences in use, including a 9% and 21% absolute increase (ITT analysis and per protocol analysis, respectively) associated with the intervention. The stronger intervention effects noted within the crossover cohort (versus the trial cohort) may be due to several factors. First, the quality with which hospitals delivered the program may have improved in Phase 2 compared with Phase 1. Not all hospitals opted to continue into Phase 2 of the study, and those that did may have been more willing and able to adapt and improve their transitional care. Although study meetings with sites were segregated by arm, usual care hospitals were aware they would be transitioning to the intervention and therefore had time to prepare. They may have initiated preparatory measures that led to faster adoption and better fidelity when they ultimately fully implemented COMPASS‐TC. Further, these hospitals may have delivered the program more successfully based on lessons learned by the study team in Phase 1 (eg, how to better engage with community rehabilitation providers) that were incorporated into training and ongoing implementation coaching in Phase 2. Although there was no increase in patient adherence between phases (as measured by telephone and 14‐day clinic follow‐up attendance), these measures may not reflect differences in the degree to with patients followed through with referrals and recommendations of the care plan or provider fidelity to TC processes and quality of care. A second reason for the stronger intervention effects noted in the crossover cohort may be differences in unmeasured confounding. The crossover design allows for control of hospital factors and may therefore be less subject to bias from system‐level factors than those of the trial cohort. Third, higher rehabilitation use in the crossover cohort could reflect an underlying temporal trend of increased use of postacute care. However, rates in a vanguard (pilot) hospital that maintained the same TC processes did not change appreciably over the same time period.

Time to Rehabilitation

We found evidence that the COMPASS‐TC intervention increased the timeliness of community‐based rehabilitation compared with usual care. This effect was observed in both the trial and crossover cohorts and in both the ITT and per protocol analyses. These findings are particularly important given the robust evidence linking timely initiation of rehabilitation following stroke to better overall recovery. ⁴⁰ , ⁴¹ , ⁴² Timely initiation of poststroke community‐based rehabilitation may also reduce the risk of adverse events such as falls, which are common among survivors of stroke, ⁴³ and inactivity/immobility, which can have consequences for both physical function and cardiovascular health. ⁴⁴ Our results on time to first therapy visit align with those of previous reported findings that the COMPASS‐TC intervention was associated with increased poststroke use of physician ambulatory care services—predominantly primary care services—and decreased time to the first ambulatory care visit. ³² Primary care follow‐up in particular may have facilitated rehabilitation use as Medicare and many private insurers require physician referral to therapy.

Therapy Intensity and Type

We found no effect of the intervention in the trial cohort in terms of intensity (ie, number of visits) or type (both versus single therapy). However, within the crossover cohort, in both adjusted ITT and per protocol analyses, COMPASS‐TC was associated with increased intensity. COMPASS‐TC was associated with 27% more visits within 90 days relative to usual care in the per protocol analyses as well as a 13.3% higher probability of receiving both PT and OT, although this latter finding was not statistically significant. Greater intensity of therapy after stroke has been shown to be associated with better functional outcomes. ⁴⁵

Transitional Care Interventions and Rehabilitation Use

Considerable progress has been made on improving acute care for stroke. ⁴⁶ However, important gaps remain in our understanding of effective strategies to improve postacute rehabilitation access in the community, which is critical for functional recovery and community reintegration. Several studies suggest rehabilitation care in the community is underused, despite evidence that it improves functional outcomes and is increasingly recognized as a priority in the United States. ¹³ , ³⁹ , ⁴⁵ , ⁴⁷ Barriers to rehabilitation including transportation, availability, affordability, and gaps in referral can be significant. Although reports suggest most patients (77%–90%) are assessed for rehabilitation before hospital discharge, ⁴⁸ , ⁴⁹ referral and use rates remain low, particularly among patients with lower stroke severity such as many of the patients in the COMPASS study. ¹³ , ³⁹ , ⁴⁵ , ⁴⁷ We previously reported that whereas 70% to 80% of patients with stroke have some degree of motor impairment, only approximately one third were discharged home with a referral to rehabilitation and there was notable variation in referral rates across hospitals. ¹⁴ One study found that 59% of patients with stroke discharged home received neither PT nor OT in the first 30 days after discharge. ⁴⁵ In our study, this was even higher, with 76% of patients in the trial cohort receiving no therapy in the community within 30 days and 70% within 90 days, in keeping with other reports. ¹³ Data on refusal of PT/OT are limited; reports suggest 6% to 28% of patients refuse recommended home health services. Reasons for therapy refusal include not fully understanding the need for services or benefits they can provide, privacy or cost concerns, and previous negative experiences. ⁵⁰ These reasons may be more common among patients with stroke with aphasia, depression, or cognitive deficits that limit their ability to communicate or engage in shared decision‐making, particularly when referring providers do not accommodate these individual needs when discussing rehabilitation and when these factors go unaddressed after stroke. ¹⁵

There is mixed evidence that TCM can improve patient outcomes such as readmission, disability, and mortality, and much of the research has been conducted outside the context of the US health care system. ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² , ²³ , ⁵¹ , ⁵² , ⁵³ A meta‐analysis of 31 stroke transitional care models (only 1 US study included) reported positive effects of these interventions on reducing mortality and improving independence. The largest effects were seen for transitional care programs with home visits led by multidisciplinary teams. ²³ Other evidence suggests that TCM interventions that incorporate elements of early supported discharge—an acute hospital discharge intervention aimed at linking inpatient and community‐based rehabilitation services—are effective for improving physical function and mortality. ²³ , ⁵³ More recent studies support the benefits of TC interventions with health coaching features on quality of life and activities of daily living. ⁵⁴ , ⁵⁵ However, little evidence exists on the impact of these interventions on rehabilitation use. Our study is one of the first to report on the impact of TC interventions on rehabilitation use and, to our knowledge, is the first to examine the impact of TCM on rehabilitation use in the community for patients with stroke. Rehabilitation use is an important potential mediator of TC interventions to achieve improvements in function and other health outcomes.

The COMPASS‐TC intervention incorporated some elements of early supported discharge, but notably not home visits as these are not reimbursed by CMS TCM billing. It was implemented by a multidisciplinary team, with the intervention being delivered by the hospital's postacute care nurse coordinator and an advanced practice provider or physician. Several components were specifically designed to support rehabilitation assessment, referral, and uptake. ³⁰ Patients and caregivers completed self‐reported assessments of functional and social determinants of health at the 2‐day call that informed early referral to rehabilitation that may have been insufficient at discharge. For patients who received therapy, most (~68%) received it before the 14‐day visit, suggesting that early TC processes such as hospital discharge planning and 2‐day follow‐up may have affected rehabilitation outcomes. Assessment of the patient's need for rehabilitation at the clinic visit was also thorough and incorporated nonmedical factors such as access to transportation, availability of a caregiver, insurance, and more. Referrals were specific to the patient's local community and included both rehabilitation providers and community‐based services that provide nonmedical needs to respond to the patient's social determinants of health.

Limitations

A notable limitation of this analysis is the presence of potential confounding. We made every effort to adjust for prognostic factors and imbalances between groups, but because we did not perform randomization on the individual level and only patients linked to insurance claims were included, the analysis is in effect observational. The possibility of unmeasured or residual confounding (eg, patient preferences), therefore, remains. Further, because we do not have outcomes on uninsured individuals, the results may not generalize to this group or to other private insurance plans that differ meaningfully from those in this study. Participants in the COMPASS Study were primarily mild stroke and TIA discharged directly home and results also may not generalize to patients who are discharged to inpatient rehabilitation or more severe strokes. Temporal changes may have affected the crossover results; for example, we noted a higher proportion of patients in Medicare Advantage in the crossover cohort reflecting an increase in Medicare Advantage enrollment over time. This change may have impacted rehabilitation use, although insurance was adjusted for in analyses and assessed as a modifier. Finally, the lack of detailed fidelity data limits our ability to fully understand differences between the trial and crossover results, and whether they were the result of improved delivery or quality of care in Phase 2. Unique strengths of this study include the representative and diverse inclusion of hospitals and survivors of stroke and the linkage to administrative claims. Together, these data provide real‐world evidence on the impact of TCM programs on use of community‐based rehabilitation services.

Future Directions

Comprehensive transitional care may improve use of rehabilitation, but as we learned, effective delivery requires dedication of hospital systems to reorganize care, meaningfully engage with multidisciplinary providers including therapists, and successfully coordinate services with home health agencies and outpatient rehabilitation facilities. ⁵⁶ TCM remains underused more than a decade after CMS introduced reimbursement codes to facilitate its adoption. ²⁴ , ²⁵ , ²⁶ , ²⁷ More viable financial models or additional quality measures may be needed to motivate systems to adopt and deliver TCM or other programs to improve postacute stroke care. Facilities that are part of integrated systems of care such as stroke centers may be more likely to have TCM processes in place.

Many barriers to TC implementation exist including inadequate staffing for care coordinators or for patient follow‐up visits and lack of tools for systematically assessing patients and producing care plans that are available to providers across settings. Fragmented systems where outpatient providers are disconnected and acute treatment of patients takes place at tertiary centers that may be far from where patients reside also present challenges for implementation. ⁵⁷ A further complication is that TC in practice and TC interventions tested in clinical trials are highly heterogeneous and complex. ²⁷ , ⁵⁸ This limits our understanding of core aspects necessary for efficacy. Implementation studies are needed to further develop and test strategies to improve care transitions after stroke that facilitate access to postacute services such as rehabilitation. Understanding the most effective components of these programs and the processes through which they act, such as increasing rehabilitation use, may allow for design of interventions that are optimized for both adoption and effectiveness. Candidate components may include standardized screening and improved education for families, goal setting, addressing transportation barriers, treating depression, and considerations for cognitive and language/communication impairments. ¹⁵ Incorporating lifestyle coaching or other interventions that increase patient activation may enhance engagement with and participation in referred therapy. ⁵⁹

Conclusions

COMPASS‐TC was consistently associated with a shorter time to first community‐based PT or OT visit, while findings for its effect on use were inconsistent. The meaningfully higher therapy use after stroke observed only in the crossover analysis may be because hospitals in this group were more willing and prepared to adapt their transitional care. Per protocol results supported that, when received, COMPASS‐TC may improve a variety of therapy use outcomes. TC is not yet widely adopted in practice. Continued research to test implementation and effectiveness of these and other approaches to improve care transitions after stroke are needed to address the concerning underuse of therapy.

Sources of Funding

The project was supported by the National Institutes of Health (award no. 1R01HD101493) through the Eunice Kennedy Shriver National Institute of Child Health and Human Development and the National Center for Medical Rehabilitation Research. The COMPASS study was supported through a Patient‐Centered Outcomes Research Institute Project Program Award (PCS‐1403‐14 532).

Disclosures

Drs Duncan and Bushnell report ownership interest in Care Directions, Inc. Dr Duncan is a research advisor for BQ Technologies. The remaining authors have no disclosures to report.

Supporting information

Tables S1–S4.

Figure S1.