Caregiver Engagement Enhances Outcomes among Randomized Control Trials of Transitional Care Interventions: A Systematic Review and Meta-Analysis

Abstract

Background:

Fluctuations in health among chronically ill adults result in frequent healthcare transitions. Some interventions to improve patient outcomes after hospitalization include caregiver engagement as a core component, yet there is unclear evidence of the effects of this component on outcomes.

Objective:

To synthesize evidence regarding the attention given to caregiver engagement in randomized control trials (RCTs) of transitional care interventions (TCIs), estimate the overall intervention effects, and assess caregiver engagement as a moderator of intervention effects.

Methods:

Three databases were systematically searched for RCTs of TCIs targeting adults living with physical or emotional chronic diseases. For the meta-analysis, overall effects were computed using the relative risk effect size and inverse variance weighting.

Results:

Fifty-four studies met criteria, representing 31,291 participants and 66 rehospitalizations effect sizes. Half (51%) the interventions lacked focus on caregiver engagement. The overall effect of TCIs on all-cause rehospitalizations was nonsignificant at 1 month (p=.107, k=29), but significant at ≥2 months (RR=0.89, 95%CI: 0.82, 0.97, p=.007, k=27). Caregiver engagement moderated intervention effects (p=.05), where interventions with caregiver engagement reduced rehospitalizations (RR=0.83, 95%CI: 0.75, 0.92, p=.001), and those without, did not (RR=0.97, 95%CI: 0.87, 1.08, p=.550). Interventions with and without caregiver engagement did not differ in the average number of components utilized, however, interventions with caregiver engagement more commonly employed baseline needs assessments (p = .032), discharge planning (p = .006), and service coordination (p = .035).

Discussion:

Future TCIs must consistently incorporate the active participation of caregivers in design, delivery, and evaluation.

Introduction

As adults age with chronic illness, negative changes in health status result in frequent movement across healthcare settings (e.g., hospital to home), hereafter referred to as healthcare transitions. On average, older adults will experience between 2–5 healthcare transitions in the year following a hospitalization.^1,2 Over the course of the chronic disease trajectory, the accumulated number of healthcare transitions is substantial. Despite efforts to improve care throughout these vulnerable transitions, the confluence of patient, family caregiver, and systems issues have made ineffective healthcare transitions the norm.^3–6 Numerous studies have documented the poor management of the complex healthcare needs of adults with chronic conditions across healthcare transitions, often with harmful human and economic consequences—up to 26% experience emergency department visits,⁴ 18% rehospitalizations,⁷ 66% adverse drug events,⁸ and 81% medication discrepancies.⁹ Transitional care has emerged out of a need to facilitate a range of time-limited services focused on care continuity and preventing poor outcomes between and across healthcare settings.¹⁰ Yet, a full complement of services to achieve effective transitional care is lacking, even across healthcare systems with healthcare transitions programs.¹¹

Rigorously tested evidence-based transitional care interventions (TCIs) targeting hospitalized adults—Better Outcomes for Older Adults Through Safe Transitions (BOOST),¹² Care Transitions Intervention (CTI),¹³ Re-Engineered Discharge (Project RED),¹⁴ Transitional Care Model (TCM)^15,16—commonly include disease self-management education, medication management, service coordination (i.e., home care), care team communication strategies (i.e., with outpatient primary or specialist providers), and caregiver engagement as core components of successful healthcare transitions. Informal caregivers—encompassing family members or friends¹⁷—play a critical role in the quality of transitional care, with evidence to support their impact on outcomes (e.g., medication discrepancies) when actively engaged across healthcare transitions.^18,19 Caregiver engagement involves, “patients, families, […] and health professionals working in active partnership at various levels across the healthcare system [...].”²⁰ The policy landscape reinforces the need for engaging caregivers, where the federal RAISE Family Caregivers Act of 2018 has called for the development of strategies that put caregivers at the center of care teams (e.g., involving family caregivers in patient assessments and service planning, including care transitions and coordination).²¹ The state-level Caregiver Advise, Record, Enable [CARE] Act also directs hospitals to incorporate caregivers into the discharge planning process and to provide them education on the care they will provide at home.²² Further, national organizations have identified caregiver engagement as a key aspect of improved health care delivery.^23,24 Despite growing awareness of the central role of caregivers in effective transitions, opportunities to systematically engage with caregivers are often missed (e.g., poor care plan communication; ineffective evaluation of caregiver needs to provide supportive care in the home), leaving patients and caregivers frequently feeling abandoned by the healthcare system across healthcare transitions.25

Despite frequent systematic reviews on healthcare transitions, attempts to synthesize the literature rarely focus on the caregiver’s perspective. Further, meta-analyses are less common, and when conducted, focus on certain disease groups (e.g., heart failure, depression)^26–28 or specific types of healthcare transitions (e.g., post-operative transitions)29 or types of interventions (e.g., pharmacist-led).30 Reported meta-analyses also often lack contextualization, the use of moderator analyses to explain sources of variation in intervention effects based on certain sample or study characteristics, contributing to the insufficient evidence base on how essential transitional care components, such as caregiver engagement, contribute to TCI intervention outcomes.^31,32 This paper addresses this critical knowledge gap by: 1) synthesizing available evidence of caregiver engagement among randomized controlled trials (RCTs) of TCIs for chronically ill adults transitioning from hospital to home; 2) estimating the overall relationship between TCIs and all-cause rehospitalizations; and 3) testing caregiver engagement in TCIs as a moderator of intervention effects. This paper addresses the question: What is the overall effect of caregiver engagement on TCI outcomes among adult patients living with chronic illness?

Methods

Search

This systematic review builds upon a previous review of TCIs.¹⁰ Consistent with the previous strategy,¹⁰ search terms included terminology consistent with three major concepts: healthcare transitions (e.g., “transition of care,” “transitional care model”), intervention design (e.g., “randomized control trial,” “experimental”) and hospital utilization (e.g., “readmission,” “emergency department,” “discharge”). Search terms were separated parenthetically into one search phrase and combined using the “AND” Boolean operator. As before, the search was conducted in CINAHL and PubMed, with a third database added (PsycINFO) to ensure thoroughness. Search limitations included peer-reviewed articles published since the prior review (2010). To identify additional eligible studies, reviews of reference sections of included studies and systematic reviews and meta-analyses captured by the search strategy were conducted.¹⁰

Eligibility Criteria

Consistent with the prior systematic review10 included studies: 1) employed a RCT design; 2) tested a TCI for patients transitioning from hospital to home, where the TCI was consistent with the definition set forth in the prior review—”time-limited services designed to ensure care continuity and promote the safe and timely transfer of patients from one setting to another”¹⁰; 3) targeted adults living with either physical or emotional chronic diseases, 4) were conducted in the United States, and 5) were published in English. Studies were excluded if the sample was <18 years of age. Additionally, both studies with and without caregiver engagement components were included to be able to address aim one and answer aim three (described in the introduction).

The prior systematic review descriptively summarized the effectiveness of TCI targeting chronically ill adults, where nine studies had a significant positive effect on at least one measure of all-cause hospital readmissions. Studies that met criteria for the prior review (Appendix) were still considered eligible for this meta-analysis, as the aim was to summarize the effectiveness of TCIs quantitatively, rather than descriptively, and addressed a new research question— focusing on the impact of caregiver engagement in TCI components on intervention outcomes (i.e., all-cause hospital readmissions).

Study Selection

Two authors (K.L., E.R.) screened studies, compared assessments, discussed discrepancies, and achieved consensus(Appendix, Supplemental Digital Content 1, http://links.lww.com/MLR/C448). A third author (K.B.H.) was retained for further adjudication when needed.

Data Collection and Synthesis

Sample characteristics, intervention characteristics, intervention components, caregiver engagement features, and outcomes were abstracted from each study. An overall determination of the level of caregiver engagement was made for each study using categories of engagement identified across a continuum by Carman and colleagues.²⁰ These categories ranged from no engagement (e.g. terms caregivers or family were not mentioned in relation to the intervention components or outcomes), to consultation (e.g. caregiver passively received information), to involvement (e.g. caregivers’ perspectives were explored), and finally, partnership and shared leadership (e.g., treatment plan accounted for caregiver perspectives).²⁰

Abstracted data were synthesized to determine trends and gaps in intervention design. Intervention components were synthesized in 2 ways: (1) proportions of interventions with individual intervention components; and (2) tabulations of the average number of intervention components across all studies as well as within the different levels of caregiver engagement. In addition, 2 authors (K.L., M.M.)independently abstracted the effect size data for the meta-analysis. As the effect size was the relative risk (RR), this abstraction included raw counts of intervention and control group patients with and without the outcome (ie, rehospitalizations). Uniform data abstraction sheets were compared, discrepancies discussed, and consensus achieved. Given our unique aims from the prior review,¹⁰this data collection and synthesis process was conducted de novo for this review.

Risk of Bias

Risk of Bias Cochrane Collaboration’s tool for assessing risk of bias was utilized to evaluate bias within studies using 7 indicators.³³Risk of bias assessments were also conducted de novo for this review by 2 authors (K.L., E.R.) and further assessed by a third author (M.M.). Any discrepancies were discussed and consensus achieved. Bias adjudication was made for each study according to recommended guidelines.³⁴

Publication bias was tested using Begg and Mazumdarrank correlation test. This test was used to detect funnel plot asymmetry, which involved a regression analysis of the rank correlation (Kendallτ) between the effects sizes and their associated SEs.³⁵A large deviation from zero indicated publication bias.

Quantitative Synthesis

Effect Size.

The effect size was the relative risk (RR). As the distribution of the RR values across studies was likely to be non-normal, the analysis was computed using the logarithm of the RR (LRR) and its associated variance (VOR).³⁶ Model parameters were back-transformed using an exponent function for more interpretable results.

Statistical Analysis.

The meta-analysis was conducted using the metafor package in R³⁷ and was based on the methods set forth by Hedges and Olkin38 and further described by Cooper, Hedges, and Valentine³⁶ (Appendix). The meta-analysis was conducted according to two different outcome groupings: 1) all-cause rehospitalizations at 1 month, and 2) all-cause rehospitalizations at 2 or more months, an approach similar to prior research.²⁶

To determine whether a fixed or random effects model was indicated, a test of homogeneity was conducted using a Q-statistic. A non-significant Q-statistic indicated a fixed effects model, which accounts for within study variation only; and a significant Q-statistic indicated a random effects model, which accounts for both within and between study (Tau²) variation. Between study variation was estimated using DerSimonian and Laird methods,³⁹ and its magnitude was interpreted using the I² value, where values of 25%, 50%, 75% indicated a low, medium, and high amount of between study variation, respectively.⁴⁰

To compute the overall effect, each effect size was weighted according to inverse variance. Then the overall effect size was estimated by taking the sum of the weighted individual effects and dividing them by the sum of the individual weights (Appendix). Significance testing for the overall effect size was conducted on a z-distribution under the null hypothesis that the overall effect size was zero. A priori significance was a p ≤ .05.

Moderator analysis.

Caregiver engagement was tested as a moderator of intervention effects. Due to the categorical nature of this moderator (i.e., interventions with or without caregiver engagement), a weighted ANOVA-like model was used. If the test of moderators (QM) was significant, indicating moderation, a test of residual between study variation (QE) was used to guide model selection. If there was a significant amount of residual between study variation left unexplained by the moderator (i.e., a significant QE), then a mixed effects model was indicated, otherwise a fixed effects model with predictors was utilized (Appendix). To compute effects, the group effect sizes within each study, according to the categories of the moderator, were first weighted using inverse variance (Appendix). The sum of the weighted effect sizes was then divided by the sum of the weights for each group. Hypothesis testing determined if the effect of the moderator (QM) was significant (i.e., the group effects were significantly different). If significant, group effect sizes were interpreted.

Results

Study Selection

The search yielded 1,256 articles, with 995 unique studies for title and abstract review (Figure 1). Of 311 studies reviewed in full text, 21 met criteria. Thirty-three additional studies were identified, either from reference section review or from the prior review,¹⁰ amounting to 54 studies in all (Appendix). Among these, three manuscripts^41–43 stemmed from parent studies in the review.^44,45 Thus, this review represents a synthesis of evidence from 51 unique RCTs of TCIs.

Figure 1.

PRISMA Diagram

Sample Characteristics

There were 31,291 participants across studies, with analyzed sample sizes ranging from 86 to 3,988 (Table 1). Across the studies reporting age characteristics (n=45), the weighted sample mean age was 59.2 years (range: 33–82 years). Studies’ inclusion criteria typically targeted patients with specific diagnoses (e.g., heart failure; COPD) (n=33, 65%), while the remainder (n = 18, 35%) targeted patients with certain risk factors (e.g., having received mechanical ventilation) or age ranges.

Table 1.

Study Characteristics

Citationa	Intervention Orientation	Dosage	Duration	Facilitator	Patient Population
Aboumater (2019)	Disease/case management	1+PRN inpatient visit PRN home visit PRN phone calls	3 months	RN	Adults 40+ with COPD (N = 240) M age 64.9 n = 120 intervention, n = 120 control
Allen (2009)	Disease/case management	1 home visit PRN phone calls	6 months	APN	Adults with ischemic stroke (N = 380) M age 68.0 n = 190 intervention, n = 190 control
Altfeld (2013)	Care coordination	1 phone call	1 month	SW	Adults 65+ with d/c risk factors (N = 720) M age 74.5 n = 360 intervention, n = 360 control
Balaban (2015), Balaban (2017), Galbraith (2017)	Comprehensive discharge planning and disease/case management	1 inpatient visit 4 phone calls	6 months	CHW	Adults with HF, COPD, or risk factors (N= 1921) M age 66.0 n = 739 intervention, n = 1182 control
Benzo (2016)	Self-management	1 inpatient visit 1 clinic visit 21 phone calls	12 months	RN RT	Adults with COPD (N = 214) M age 68.0 n = 108 intervention, n = 106 control
Bielaszka- DuVernay (2011)	Disease/case management	2+PRN home visits 12 phone calls	12 months	APN SW	Adults 65+ with low income (N = 951) M age and group sample sizes not reported
Biese (2018)	Care coordination	2 phone calls	1 month	RN	Adults 65+ (N = 1949) M age 74.1, n = 974 intervention, n = 975 control
Bowles (2011)	Disease/case management	5 home visits Up to 9 video calls	2 months	RN	Adults 55+ with HF (N = 217) M age 72.5, n = 101 intervention, n = 116 control
Calvert (2012)	Medication management	1 inpatient visit 1 phone call	6 months	PharmD	Adults with CAD (N = 143) Median age 62 n = 71 intervention, n = 72 control
Carroll (2007)	Self-management	1 home visit 3 APN phone calls 12 peer phone calls	3 months	APN Peers	Adults 65+ following MI or CABG surgery who were unpartnered (N = 247) M age 76.3 n = 121 intervention, n = 126 control
Chan (2015), Goldman (2014)	Comprehensive discharge planning	2 inpatient visits 2 phone calls	2 weeks	APN RN	Adults 55+ (N = 700) M age 66.2 n = 347 intervention, n = 352 control
Coleman (2006)	Self-management	1 inpatient visit 1 home visit 3 phone calls	1 month	APN RN	Adults 65+ with specific chronic conditions (N = 750) M age 76.2 n = 379 intervention, n = 371 control
Currier (2010)	Care coordination	1 clinic visit	1 week	Psychiatrist RN	Adults with suicidal ideation (N = 120) M age 32.7 n = 56 intervention, n = 64 control
Daly (2005)	Disease/case management	1 inpatient visit 8+PRN home visits or phone calls	2 months	APN	Adults with mechanical ventilation (N = 334) M age 62.9 n = 231 intervention, n = 103 control
Davis (2012)	Disease/case management	1 inpatient visit 1 phone call	1 week	Case Managers	Adults 21+ with HF and MCI (N = 125) M age 58.5 n = 63 intervention, n = 62 control
Dixon (2009)	Care coordination	1 inpatient visit PRN home visits PRN clinic visits	3 months	RN SW	Adults 18–70 with psychiatric illness (N = 135) M age 47.8 n = 64 intervention, n = 71 control
Englander (2014)	Disease/case management and care coordination	1 inpatient visit PRN home visits PRN phone calls	1 month	RN	Adults in general medicine or cardiology unit with risk factors (N = 382) Age not reported n = 209 intervention, n = 173 control
Farris (2014)	Medication management	3 inpatient visits 1 phone call (enhanced only)	1 week	PharmD	Adults in specific units with risk factors (N = 945) M age 61.0 n = 314 intervention 1, n = 316 intervention 2, n = 316 control
Finn (2011)	Comprehensive discharge planning	1 inpatient visit PRN phone calls	1 week	APN	Adults in general medicine units (N = 872) M age 63.0 n = 440 intervention, n = 432 control
Fitzgerald (1994)	Disease/case management	PRN clinic visits 2+ PRN phone calls	12 months	RN	Adults 45+ in general medicine units (N = 668) M age 64.5 n = 333 intervention, n = 335 control
Grahn (2019)	Disease/case management	1 inpatient visit 1 phone call	1 week	Not reported	Adults undergoing ileostomy (N = 100) M age not reported n = 49 intervention, n = 51 control
Gurwitz (2014)	Care coordination	1 transmission	n/a	Automated	Adults 65+ (N = 3661) M age 79.1 n = 1870 intervention, n = 1791 control
Heaton (2019)	Medication management	1 pharmacy visit 1 phone call	1 week	PharmD	Adults with MI, pneumonia, HF, COPD, or DM (N = 400) M age 61.7 n = 213 intervention, n = 187 control
Ho (2014)	Medication management	1 inpatient visit 2 clinic visits PRN phone calls	12 months	PharmD	Adults with acute coronary syndrome (N = 241) M age 63.9 n = 122 intervention, n = 129 control
Jack (2009)	Comprehensive discharge planning	PRN inpatient visits 1 phone call	1 week	RN	Adults (N = 749) M age 49.9 n = 373 intervention, n = 376 control
Jennings (2015)	Self-management	1 inpatient visit 1 phone call	1 week	RN	Adults 40+ with COPD (N = 172) M age 64.6, n = 93 intervention, n = 79 control
Kripalani (2012)	Medication management	1–2 inpatient visits 1 phone call	1 week	PharmD	Adults with ACS or HF (N = 851) M age 60.0 n = 423 intervention, n = 428 control
Laramee (2003)	Disease/case management	Daily inpatient visits 9+PRN phone calls	3 months	RN	Adults with HF and risk factors (N = 186) M age 70.7 n = 141 intervention, n = 145 control
Linden (2014)	Self-management	1 inpatient visit PRN phone calls	3 months	RN	Adults with HF or COPD (N = 512) M age 66.8 n = 253 intervention, n = 259 control
Liss (2019)	Disease/case management and care coordination	PRN clinic visits	PRN up to 6 months	Team	Adults (N = 654) M age 43.8 n = 490 intervention, n = 164 control
Magny-Normilus (2019)	Comprehensive discharge planning and disease/case management	PRN inpatient visits 1 home visit, 3 phone calls, 2 clinic visits	1 month	APN	Adults with DM on insulin and cardiac disease (N = 180) M age 64.4 n = 88 intervention, n = 92 control
McWilliams (2019)	Disease/case management and care coordination	PRN clinic visits 4+PRN phone calls	1 month	RN	Adults with discharge risk factors (N = 1,876) M age 58.9 n = 935 intervention, n = 941 control
Melton (2012)	Disease/case management	1 phone call	1 week	CM	Adults with specific diagnoses (N = 3,988) M age 50.0 n = 1,994 intervention, n = 1994 control
Mion (2003)	Care coordination	In-ED PRN phone calls	1 week	APN	Adults 65+ (N = 650) M age 74.4 n = 326 intervention, n = 324 control
Naylor (1994)	Comprehensive discharge planning	Inpatient - q48h 2+PRN phone calls	2 weeks	APN	Adults 70+ with cardiac disease (N = 174) M age 75.5 n = 140 intervention, n = 134 control
Naylor (1999)	Comprehensive discharge planning and disease/case management	Inpatient - q48h 2+PRN home visits 4+PRN phone calls	1 month	APN	Adults 65+ with specific conditions (N = 363) M age 75.4 n = 177 intervention, n = 186 control
Naylor (2004)	Comprehensive discharge planning and disease/case management	Daily inpatient visits 8+PRN home visits 12+PRN phone calls	3 months	APN	Adults 65+ with heart failure (N = 139) M age 76.0 n = 118 intervention, n = 121 control
Parry (2009)	Self-management	1 inpatient visit 1 home visit 3 phone calls	1 month	RN	Adult 65+ with specific chronic conditions (N = 86) M age 81.7 n = 44 intervention, n = 42 control
Pekmezaris (2012)	Disease/case management and care coordination	PRN home visits PRN video calls	2 months	RN	Adults with HF and home health referral (N= 168) M age 82.0 n = 83 intervention, n = 85 control
Phatak (2016)	Medication management	2 inpatient visits 3 phone calls	1 month	PharmD	Adults with specific medications (N = 278) M age 55.6 n = 137 intervention, n = 141 control
Reeves (2019)	Disease/case management and care coordination	PRN home visits PRN phone calls	3 month	SW	Adults with stroke (N = 178) M age 66.2 n = 88 intervention 1, n = 90 intervention 2, n = 87 control
Rich (1995)	Disease/case management and care coordination	Daily inpatient visits 3+PRN home visits PRN phone calls	3 months	RN	Adults with heart failure and risk factors (N = 282) M age 79.3 n = 142 intervention, n = 140 control
Riegel (2004)	Self-management	PRN inpatient visits 12 home visits and/or phone calls	3 months	Peer	Adults with heart failure and risk factors (N = 88) M age 73.0 n = 45 intervention, n = 43 control
Ritchie (2016)	Disease/case management	1 inpatient visit 28 auto-calls PRN phone calls	1 month	RN	Adults with heart failure or COPD (N = 478) M age 63.4 n = 233 intervention, n = 245 control
Saleh (2012)	Disease/case management and care coordination	1 inpatient visit 3 home visits 1 clinic visit	1.5 months	RN	Adults 65+ (N = 333) M age not reported n = 160 intervention, n = 173 control
Sales (2013)	Disease/case management	2 inpatient visits 4 phone calls	1 month	Pre-med students	Adults with heart failure (N = 137) M age 72.6 n = 70 intervention, n = 67 control
Siu (1996)	Comprehensive discharge planning	1 inpatient visit 4 home visits	2 months	APN	Adults 65+ with risk factors (N = 354) M age not reported n=178 intervention, n=176 control
Tuttle (2018)	Medication management	1 home visit	1 week	PharmD	Adults 65+ with CKD (N = 141) M age 69.0 n = 72 intervention, n = 69 control
Velligan (2017)	Disease/case management	1+PRN clinic visits	3 months	Team	Adults 65+ with SMI (N = 326) M age 38.0, n = 219 intervention, n = 107 control
Wakefield (2008)	Disease/case management and care coordination	13 calls (phone or video)	3 months	RN	Adults with HF (N = 137) M age 69.3, n = 47 intervention 1, n = 52 intervention 2, n = 49 control
Weinberger (1996)	Disease/case management and care coordination	2 inpatient visits 1+PRN calls 1+PRN clinic visits	6 months	RN	Adults 65+ with diabetes, COPD, or HF (N = 1,396) M age 62.8 n = 695 intervention, n = 701 control

^aFull citations for included studies can be referenced in the Supplemental References section of the Appendix

Abbreviations: ACS = acute coronary syndrome, APN = advance practice nurse, CAD = coronary artery disease, CKD = chronic kidney disease, COPD = chronic obstructive pulmonary disease, DM = diabetes, ED = emergency department, CHW = community health worker, MI = myocardial infarction, GI = gastrointestinal, HF = heart failure, M age = mean age, MCI = mild cognitive impairment, PharmD = pharmacist, PRN = as needed, RN = registered nurse, SMI = serious mental illness, SW = social worker, Team = interdisciplinary team shares facilitation role

Interventions

Characteristics.

While the 51 unique TCIs were multi-faceted, most were oriented toward certain aspect(s) of care transitions: disease/case management (n=13, 25%), disease/case management and care coordination (n=9, 17%), self-management (n=7, 14%), medication management (n=7, 14%), care coordination (n=6, 12%), comprehensive discharge planning (n = 5, 10%), and comprehensive discharge planning and disease/case management (n=4, 8%) (Table 1). Varied interdisciplinary team members facilitated interventions, but registered nurses (RNs) and advanced practice nurses (APRNs) were most common (n=33, 65%). Intervention dosage considerably differed, where some interventions were as simple as a single phone call⁴⁶ and others were more complex with varying combinations and frequencies of interactions (e.g., inpatient visits, home visits, follow-up calls). Inpatient visits (n=32, 63%) and follow-up calls (n=43, 84%) were most common across interventions, while outpatient clinic visits (n=11, 22%) and home visits (n=20, 39%) were less common. Interventions typically involved short-term follow-up—up to a month (n=26, 51%), while others involved longer-term follow-up—up to 2 months (n=5, 10%), 3 to 6 months (n=16, 31%), or a year (n=4, 8%).

Components.

Intervention components were diverse overall (Table 2; Table A, Appendix), with the majority utilizing disease- and self-management education (n=39, 76%), phone call/telemedicine follow-up (n=38, 75%), care team communication (e.g., communication with out-of-hospital providers) (n=37, 73%), baseline needs assessments (n=29, 57%), and service coordination (e.g., ensuring implementation of home health) (n=28, 55%). Interventions with and without caregiver engagement did not differ in the average number of components utilized (6 vs. 5). However, significant differences were identified in the component types, where the proportion of interventions with caregiver engagement using baseline needs assessments (72% vs. 42%, p=.032), discharge planning (52% vs. 15%, p=.006), and service coordination (68% vs. 42%, p=.035) was significantly greater than the proportion of interventions with no caregiver engagement using these components.

Table 2.

Intervention Components by Level of Caregiver Engagement

Component	By Levels of Caregiver Engagement						Dichotomized Engagement Comparison				X 2	Overall
	Partnership n = 8		Involvement n = 9		Consultation n = 8		Any Engagement n = 25		No Engagement n = 26			N = 51
	Total n (%)	Caregiver n (%)	Total n (%)	Caregiver n (%)	Total n (%)	Caregiver n (%)	Total n (%)	Caregiver n (%)	Total n (%)	Caregiver n (%)		Total n (%)	Caregiver n (%)
Baseline Needs Assessment	6 (75)	4 (50)	8 (89)	7 (78)	4 (50)	2 (25)	18 (72)	13 (52)	11 (42)	---	.032	29 (57)	13 (25)
Disease- or Self-management Education	6 (75)	6 (75)	7 (78)	6 (67)	7 (88)	7 (88)	20 (80)	19 (76)	19 (73)	---	.560	39 (76)	19 (37)
Discharge Planning	4 (50)	3 (38)	6 (67)	4 (44)	3 (38)	0 (0)	13 (52)	7 (28)	4 (15)	---	.006	17 (33)	7 (14)
Service Coordination	6 (75)	1 (13)	8 (89)	3 (33)	3 (38)	0 (0)	17 (68)	4 (16)	11 (42)	---	.035	28 (55)	4 (8)
Coaching or Motivation	4 (50)	2 (25)	3 (33)	1 (11)	2 (25)	1 (13)	9 (36)	4 (16)	11 (42)	---	.645	20 (39)	4 (8)
Medication Management	4 (50)	2 (25)	3 (33)	0 (0)	5 (63)	1 (13)	12 (48)	3 (12)	11 (42)	---	.683	23 (45)	3 (6)
Care Team Communication	5 (63)	2 (25)	8 (89)	0 (0)	6 (75)	0 (0)	19 (76)	2 (8)	18 (69)	---	.588	37 (73)	2 (4)
Red Flag Education	4 (50)	3 (38)	5 (56)	3 (33)	3 (38)	2 (25)	12 (48)	8 (32)	6 (23)	---	.063	18 (35)	8 (16)
Home Visits	6 (75)	2 (25)	2 (22)	2 (22)	4 (50)	1 (13)	12 (48)	5 (20)	8 (31)	---	.208	20 (39)	5 (10)
Phone calls or Telemedicine	7 (88)	3 (38)	8 (89)	5 (56)	4 (50)	0 (0)	19 (76)	8 (32)	19 (73)	---	.588	38 (75)	8 (16)
Emotional or Psychological Support	4 (50)	3 (38)	3 (33)	2 (22)	1 (13)	0 (0)	8 (32)	5 (20)	5 (19)	---	.296	13 (25)	5 (10)
Average number of components per study	7	4	7	4	5	2	6	3	5	0	---	---	---

Note: Each cell refers to the total number of interventions that included a given component divided by the total number of interventions in that caregiver engagement category. The total column refers to the number of interventions that used a given component, while the caregivers column refers to the number of studies that engaged caregivers in that component.

Caregiver engagement features.

Overall, most interventions were adjudicated with a no caregiver engagement determination (n=26, 51%). The remainder (n=25) had some form of caregiver engagement (consultation, n=8 (16%); involvement, n=9 (17%); partnership, n=8 (16%)). Among the 25 studies with caregiver engagement, the majority of caregivers were involved in the disease- and self-management education (n=18, 72%) or needs assessments (n=13, 52%) components of the interventions (Table 2; Table A, Appendix). At the higher levels of engagement (i.e., partnership, involvement), caregivers were involved in four intervention components on average, compared to two components at the consultation level (Table 2).

Outcomes.

Interventions emphasized healthcare system outcomes (n=46, 90%) over patient outcomes (n=32, 63%), and rarely evaluated caregiver outcomes (n=2, 4%) (Table 3). All-cause rehospitalizations were the most frequently monitored overall (n=43, 84%), and thus utilized for meta-analysis.

Table 3.

Monitored Outcomes by Level of Caregiver Engagement

Outcome	By Levels of Caregiver Engagement			Dichotomized Engagement Comparison		Overall
	Partnership n = 8 n (%)	Involvement n = 9 n (%)	Consultation n = 8 n (%)	Any Engagement n = 25 n (%)	No Engagement n = 26 n (%)	N = 51 n (%)
Healthcare system outcomes (n = 46, 90%)
All-cause readmissions	7 (88)	9 (100)	7 (88)	23 (92)	20 (77)	43 (84)
Emergency room/acute care visits	2 (25)	6 (67)	4 (50)	12 (48)	12 (46)	24 (47)
Costs	3 (38)	5 (56)	3 (38)	11 (44)	4 (15)	15 (29)
Adherence to outpatient follow-up	3 (38)	2 (22)	2 (25)	7 (28)	5 (19)	12 (24)
Patient outcomes (n = 32, 63%)
Quality of life	4 (50)	3 (33)	2 (25)	9 (36)	8 (31)	17 (33)
Care satisfaction	1 (13)	4 (44%)	2 (25)	7 (28)	6 (23)	13 (25)
Mortality	3 (38)	2 (22)	3 (38)	8 (32)	7 (27)	15 (29)
Caregiver outcomes (n = 2, 4%)
Caregiver stress	1 (14)	0 (0)	0 (0)	1 (< 1)	0 (0)	1 (2)
Caregiver perceived quality of discharge preparation	1 (14)	0 (0)	0 (0)	1 (< 1)	0 (0)	1 (2)

Bias

The majority of RCTs were assessed with an uncertain overall risk of bias (n=35, 69%), with the remaining a high (n=16, 31%) risk of bias (Table B, Appendix). Risk of bias was not a significant moderator of intervention effects on all-cause rehospitalizations at 2 or more months (p=.391) (Appendix). There was no evidence of publication bias across RCTs included in the meta-analysis at the 1 month timepoint (Kendall’s tau=−0.24, p=.075) or the 2 or more months timepoint (Kendall’s tau=−0.13, p=.341).

Quantitative Synthesis

A total of 66 rehospitalizations effects sizes were abstracted across studies (some interventions reported rehospitalizations at more than one timepoint). Among these, 29 pertained to all-cause rehospitalizations at 1 month, and 27 to all-cause rehospitalizations at 2 or more months.

All-cause rehospitalizations at 1 month.

Under a fixed effects model, the overall effect of TCIs on all-cause rehospitalizations at 1 month was non-significant (p=.107) (Table 4; Figure A, Appendix). Moderation analysis was not indicated due to the fixed effects model (i.e., non-significant between study variation).

Table 4.

Series of Meta-analyses

Analysis (Model)	All-cause Readmissions (Moderator)	Studies (n)	Effect Sizes (k)	Participantsa	LRR	SE	z	95% CI LB UB		RR	95% CI LB UB		p
Overall (FE)	At 1 month	27	29	22,221	−0.05	0.03	−1.61	−0.12	0.01	0.95	0.89	1.01	.107
Overall (RE)	At 2 or more months	24	27	14,741	−0.13	0.05	−2.79	−0.22	−0.04	0.88	0.81	0.96	.005
Moderation (ME)	At 2 more months (Caregiver engagement)	24	27	Studies with engagement:	−0.19	0.06	−3.54	−0.30	−0.09	0.82	0.74	0.92	< .001
				Studies without engagement:	−0.04	0.06	−0.63	−0.15	0.08	0.96	0.86	1.09	.530

^aRepresents the total number of participants across the study samples included in the analysis; Abbreviations: FE = fixed effects model, LB = lower bound, ME = mixed effects model, RE = random effects model, UB = upper bound

All-cause rehospitalizations at 2 or more months.

Under a random effects model, the overall effect of TCIs on all-cause rehospitalizations at 2 or more months was significant (RR=0.88, 95% CI: 0.81, 0.96, p=.005) (Table 4; Figure B, Appendix), which indicated that patients who received TCIs experienced a 12% reduction in the risk of all-cause rehospitalizations on average compared to participants across control groups. The estimated amount of between study variation (Tau²=.03, SE=.02) was medium in magnitude I² = 56%, indicating moderator analysis. Caregiver engagement was a significant moderator of this effect (p=.05), explaining 38% of between study variation (R²=38.2); interventions with caregiver engagement produced more robust effects (RR=0.82, 95% CI: 0.74, 0.92, p<.001) than interventions without caregiver engagement (RR=0.96, 95% CI: 0.86, 1.09, p=.530).

As studies often reported rehospitalizations outcomes at several timepoints following 2 months post-discharge, a sensitivity analysis was conducted to confirm stability of both the overall intervention effects at 2 or more months and the moderation (Appendix). The above results represent the analysis of timepoints most proximal to the index hospitalization after the 2-month timepoint. The sensitivity analysis utilized timepoints most distal to the index hospitalization after the 2-month time point. The sensitivity analysis revealed stable findings, where intervention effects significantly reduced all-cause rehospitalizations at 2 or more months (RR=0.89, 95% CI: 0.82, 0.96, p=.003), and caregiver engagement continued to be a significant moderator (p=.041).

Discussion

This study represents a synthesis of nearly thirty years of evidence from RCTs of TCIs. While TCIs were not found to produce significant effects on all-cause rehospitalizations in the short-term (i.e., at 1 month), the interventions produced significant effects on all-cause rehospitalizations in the medium- to long-term (i.e., 2 to 12 months). Additionally, the findings uncovered caregiver engagement as a critical feature of TCIs, along with other key components, that influence TCI outcomes. Further, gaps in outcome monitoring were highlighted—fewer patient outcomes than healthcare system outcomes, and virtually non-existent caregiver outcomes.

Our findings add empirical support for the value in providing wraparound care transitions services to patients when transitioning from one healthcare setting to another. The significant overall intervention effect on all-cause rehospitalizations at 2 or more months identified in this study are similar in magnitude to that of a meta-analysis of TCIs for a disease-specific population (e.g., heart failure), where the risk for all-cause rehospitalizations at 3–6 months was reduced by 8–25%,²⁶ compared to 12% herein. However, this study provides some contrasting evidence with a prior meta-analysis of care coordination interventions for adults living with chronic disease, where significant short-term effects, not long-term effects, were identified.⁴⁷ It is possible that the narrower focus on care coordination in this prior meta-analysis explains the more significant effects on rehospitalizations in the short-term, while our review may have captured a broader set of TCI components that are significant in a reducing rehospitalizations in the medium- to long-term. Additionally, our findings may have produced more precise estimates of effects than these prior reviews given the greater number of effects sizes included.

Advancing the literature, our moderation analysis revealed that not all interventions were created equal. Interventions with caregiver engagement produced significant reductions in the risk of all-cause rehospitalizations at 2 or more months, whereas those lacking engagement did not. This provides empirical support of the contribution that caregivers make on care transitions and the critical need to include them to improve patient health. Further examination of the features of interventions with and without caregiver engagement revealed key differences. While the average number of components in interventions with and without caregiver engagement did not significantly differ, the proportion of interventions using certain types of intervention components did significantly differ across caregiver engagement groups. Interventions with caregiver engagement more commonly included baseline needs assessments, discharge planning, and service coordination—components which have been previously identified by caregivers as critical to achieving desired outcomes across care transitions.²⁵ These components also reflect varying stages of transitions – components delivered upon admission, during admission, and post-discharge. Prior reviews have also suggested that it is not the number of intervention components, but the complexity management^16,28 and the types of intervention components that span the transition that influence effective care transition outcomes.^10,48 Care transition interventions that bundle care coordination and caregiver engagement components in the pediatric population have resulted in superior outcomes,⁴⁹ and an emerging conceptual framework in the adult population supports connecting caregivers to healthcare providers as well as coaching them in healthcare system navigation as key mechanisms in TCI outcomes.⁵⁰ Taken together, this suggests that caregiver engagement is an essential component of TCIs that enhances effectiveness particularly when delivered in conjunction with a broader package of key intervention components across the care transition.

This study also highlights persistent gaps in outcome monitoring during care transitions. More diverse outcomes, like caregiver burden and caregiver emotional health,³¹ were lacking across studies. Such gaps in outcome monitoring are striking given the well-documented burden caregivers experience when patients have episodic health crises and the subsequent potential for worsening patient outcomes—challenges which have been further underscored by the COVID-19 pandemic.^17,51,52 The Patient Reported Outcomes Measurement Information System (PROMIS) initiative has created a bank of patient-centered tools to assess the physical, mental, and social health of individuals.⁵³ Existing PROMIS tools for families of children with chronic disease⁵⁴ provide a foundation for adapting these measures for caregivers of adults with chronic illness. Improved measurement of caregivers’ experiences during TCIs is an important aspect of developing TCIs that are more patient- and family-centered.³¹

These findings emphasize the important fact that care transitions are not occurring in a vacuum, but in the larger context of patients’ families. Thus, by ignoring important contextual factors, like caregiver engagement, it should come as no surprise that interventions are less effective. These findings provide empirical support for the central role of caregivers in the quality of care transitions as patients age with chronic illness. Such evidence also holds policy implications. While the CARES Act made the important step in ensuring healthcare systems are documenting a caregiver in the electronic health record, it does not incentivize more active partnerships with caregivers in care delivery. Increasing support for caregiver engagement during care transitions may require the development of value-based reimbursement models that reward healthcare systems for the adoption of caregiver-engaged principles, creating an environment for systemic change in the way healthcare systems conduct business,²⁴ which may also increase the capability of caregivers to provide care in the home over time, thus, delaying institutionalization.

Limitations

This study had some limitations. The inconsistent operational definitions of the patient-level outcomes and limited reporting of caregiver-level outcomes prevented quantitative synthesis of intervention effects beyond systems-level outcomes. Further, the caregiver engagement moderator explained 38% of the between study variation—leaving room for other potential moderators. However, inconsistent reporting of race, ethnicity, and caregiver-level data prevented further testing of these other sample characteristics as moderators. Future studies could also consider analyzing interaction effects (race/ethnicity with caregiver-level variables or age characteristics with caregiver-level variables) to further understand the differential impacts of TCI interventions among the sub-group of studies that did have caregiver engagement. A post-hoc exploration of a study characteristic as a moderator (disease-centered vs. risk factor-centered sample inclusion criteria) was also conducted but did not prove significant (Appendix). However, a prior meta-analysis indicated that TCIs were not as effective among patients with certain diagnoses (e.g., depression)—suggesting that differential effects with certain study characteristics may also remain.²⁷

Conclusions

These meta-analytic findings provide compelling evidence that caregiver engagement is an essential component of any TCI intervention. The findings underscore that it is not the number of components used in the TCI, but the component type and caregiver engagement in those components that optimize transitional care outcomes. Whether in research or clinical practice, transitional care should not be conducted without careful consideration of where and how caregivers will be incorporated and supported as active partners in optimizing patient care across healthcare transitions.