Effectiveness of a Care Transitions Intervention for Older Adults Discharged Home from the Emergency Department: A Randomized Controlled Trial

Abstract

Background:

Improving care transitions following emergency department (ED) visits may reduce post-ED adverse events among older adults (e.g., ED revisits, decreased function). The Care Transitions Intervention (CTI) improves hospital-to-home transitions; however, its effectiveness at improving post-ED outcomes is unknown. We tested the effectiveness of the CTI with community-dwelling older adult ED patients, hypothesizing it would reduce revisits and increase performance of self-management behaviors during the 30 days following discharge.

Methods:

We conducted a randomized controlled trial among patients age≥60 discharged home from one of three EDs in two states. Intervention participants received a minimally-modified CTI, with a home visit 24-72 hours post-discharge and 1-3 phone calls over 28 days. We collected demographic, health status, and psychosocial data at the initial ED visit. Medication adherence and knowledge of red flag symptoms were assessed via phone survey. Care use and comorbidities were abstracted from medical records. We performed multivariate regressions for intention-to-treat and per-protocol analyses.

Results:

Participant characteristics (N=1756) were similar across groups: mean age 72.4 ± 8.6 years, 53% female. Of those randomized to the intervention, 84% completed the home visit. Overall, 12.4% of participants returned to the ED within 30 days. The CTI did not significantly affect odds of 30-day ED revisits (adjusted odds ratio [AOR] = 0.97, 95% confidence interval [CI] = 0.72 to 1.30) or medication adherence (AOR = 0.89, 95% CI = 0.60 to 1.32). Participants receiving the CTI (PP) had increased odds of in-person follow-up with outpatient clinicians during the week following discharge (AOR = 1.24, 95% CI = 1.01 to 1.51) and recalling at least one red flag from ED discharge instructions (AOR = 1.34 95% CI = 1.05 to 1.71).

Conclusions:

The CTI did not reduce 30-day ED revisits but did significantly increase key care transition behaviors (outpatient follow-up, red flag knowledge). Additional research is needed to explore if patients with different conditions benefit more from the CTI and whether decreasing ED revisits is the most appropriate outcome for all older adults.

Introduction

Older adults are a growing and vulnerable population with healthcare needs requiring distinctly different emergency care practices.^1,2 The CDC reports that in 2017, 22 million ED visits were made by adults age≥65.³ These ED visits take longer, represent higher acuity illnesses, and require more intensive services, resulting in increased costs.^1,3,4 ED visits for older adults are frequently followed by adverse events such as functional and physical decline, delirium, and death,^5-8 with risk increasing for those with advanced age, frailty, multiple chronic conditions, and other conditions prevalent among older adults (e.g., dementia).^3,7,9-11 Although the majority of older ED patients are discharged, they are more likely to revisit the ED than younger adults and be subsequently admitted to the hospital.^3,10 Recent studies have shown approximately 20-25% of older ED patients return within 30 days of discharge, often due to non-adherence with medications and treatment plans, not understanding discharge instructions, lack of disease knowledge, increases in symptoms, or lack of social support.^6,10,12-15

Improving ED-to-home care transitions has been promoted as a strategy to prevent revisits and improve health and cost outcomes for older ED patients.^12,16-19 Suboptimal transitions result in poor continuity of care, fragmentation of services, and increased costs.^17,19,20 Facilitating the ED-to-home transition may improve use of outpatient services and adherence to treatment plans. Interventions have taken many forms, including improved information provision during discharge, post-discharge telephone follow-up, outpatient appointment scheduling, processes to improve communication with primary care providers, and providing care coordination/management services.^16,20 Research testing the effectiveness of these interventions has frequently focused on ED revisits as the primary outcome, yielding mixed results with limited generalizability.^{12,14,16,21,22}

The Care Transitions Intervention (CTI), developed by Coleman and colleagues, employs a well-established coaching model for post-discharge self-management support demonstrated to decrease readmission, improve follow-up care, and decrease costs in hospital-to-home transitions.^23,24 Hospital-to-home transitions, however, differ significantly from ED-to-home transitions, with less time available for care management activities prior to discharge. No studies have examined the effect of the CTI specifically on ED-to-home transitions in the general older adult population. We adapted delivery of the CTI for use following ED discharge among community-dwelling older adults, demonstrating high fidelity in both training and administration.^25-27 We hypothesized that older adults randomized to receive the CTI would have fewer ED revisits within 30 days, as well as perform self-management behaviors targeted by the intervention (outpatient follow-up, red flag identification, medication adherence), at higher rates than those receiving usual care.

Methods

Study Design

Setting:

We conducted a single-blind randomized controlled trial to examine the effectiveness of an ED-to-home CTI among community-dwelling older adults (clinicaltrials.gov registration NCT02520661) at three university-affiliated hospital EDs: one in Madison, WI and two in Rochester, NY. The study was approved by Institutional Review Boards at the University of Wisconsin and University of Rochester with written informed consent. Enrollment and data collection occurred between January 2016 and July 2019.

Participants:

Eligible subjects were at least 60 years of age, resided in either Dane County, WI or Monroe County, NY, had a primary care provider (PCP) affiliated with either health system, had a working telephone, and were discharged from the ED to a community residence within 24 hours of arrival. Subjects were excluded if they did not speak English, were visually or hearing impaired, did not have a permanent residence, were actively enrolled in either hospice, a transitions program, or a care management program, presented with a primary behavioral or psychiatric problem, or had an Emergency Severity Index (ESI) category of 1. Previous participants were also excluded. A priori calculations determined that data from 860 participants per study arm (1720 total) would be needed to detect a 5% absolute decrease in 30-day ED revisits with 80% power using a two-sided significance level of 5%, as well as odds ratios of 1.6-1.7 for dichotomous predictor variables (details described in the published study protocol).²⁵

Study Procedures:

Study protocol details have been previously published,^25,26 and are depicted in Figure 1. In brief, research coordinators identified and obtained consent from eligible ED patients. Legally-authorized representatives could provide consent for patients without decisional capacity. Once consented, participants were randomly assigned to either the control (usual care) or treatment (intervention) group based upon within-site block randomization strategies conducted by a data analyst not involved with enrollment. Research enrollers were blinded until after consent, when a sealed envelope was opened containing the group assignment associated with the participant identification number.

Figure 1:

Study process diagram (Adapted from Mi et al., BMC Geriatrics, 2018)²⁵

Baseline survey instruments were verbally administered to all participants prior to ED discharge, assessing self-reported health status, demographic characteristics, cognitive status, and psychosocial measures. Legally-authorized representatives could assist in completing most, but not all (e.g., cognition assessments), of these measures. Follow-up phone calls occurred approximately 4 and 30 days after ED discharge to collect patient-reported outcomes and collect longitudinal data using some scales from the baseline survey (as previously reported).²⁵ We abstracted data from medical records including healthcare use and comorbid conditions using best practices for unbiased chart review.²⁸ A formal audit and consensus process was conducted at regular intervals throughout the study to ensure adherence to data abstraction protocols and continuing reliability at each site.

Intervention:

Treatment group participants received the CTI, adapted for use following ED visits,²⁶ delivered by community paramedic coaches certified to deliver the CTI by intervention developers.²⁷ The intervention consisted of a home visit 24-72 hours post-discharge and up to 3 coaching phone calls within 28 days.²⁵ Research coordinators provided paramedic coaches with the participant’s After Visit Summary (AVS) and a short narrative description immediately following ED discharge. Prior to the home visit, coaches reviewed this information and called the participant to verify time and location. During the home visits, paramedics used coaching strategies (e.g., motivational interviewing) to address CTI content, focusing on outpatient follow-up, medication reconciliation and self-management, understanding “red flag” symptoms necessitating further care, creation or review of a personal health record for use in future healthcare appointments, and health-related goal setting. Coaches used standard CTI assessment forms to document content covered, as well as descriptive information to capture context, details, and information useful for planning follow-up phone calls. Each home visit lasted approximately one hour. In follow-up phone calls, coaches reinforced content and behaviors previously addressed, answered questions, and employed additional motivational interviewing strategies as needed. All participants completing the home visit were scheduled for at least one follow-up phone call, with up to two additional calls scheduled at the coach’s discretion based on perception of need related to targeted self-management behaviors and progress towards self-identified goals.

Measures and Key Outcomes

Primary Outcome—ED Revisits:

Examined as a quality indicator, ED revisits are commonly measured at 30 days after discharge with additional timeframes used to obtain a more nuanced understanding of processes influencing revisits.^20,29 We included all unplanned ED use, regardless of reason, during the 30 days after discharge as abstracted from electronic medical records (with out-of-system ED use identified during participant phone surveys). We constructed dichotomous variables measuring whether or not any ED visits occurred within 14 and 30 days after discharge, with the 30-day interval being the primary pre-specified outcome.

Secondary Outcomes—Self-Management Behaviors:

The main components of the CTI (referred to as “pillars”) target specific self-management behaviors associated with effective transitions. We measured three of these as secondary outcomes³⁰ to assess whether the intervention influenced care transition behaviors following ED visits:

1. Outpatient follow-up: Follow-up visits, abstracted from participant medical records, included office visits with primary or specialty providers, telephone calls, and online patient portal messaging (excluding automated reminder messages, electronic messages that did not receive a patient response, laboratory testing, and previously-scheduled outpatient procedures). Outpatient follow-up was dichotomized by whether or not any contact with outpatient providers occurred either within 7 or 30 days of discharge. This approach allowed us to differentiate follow-up occurring soon after discharge (consistent with most ED discharge instructions) from less-timely contact with outpatient clinics. In order to conduct a pre-planned sub-analysis, we also categorized all follow-up as either “in-person” or “electronic”. Dichotomous variables were created for each modality at each time point, as well as a combined variable representing all forms of contact. Telemedicine was not available during this time.

2. Red flag knowledge: Red flags are clinical signs and symptoms that should lead a patient to seek immediate medical care. During the telephone survey four days after discharge, participants reported red flags provided at discharge. We determined their ability to correctly recall at least one specific red flag of those included on the individualized AVS given to them at discharge. Comparisons of each patient’s self-reported red flags to those contained on their printed AVS instructions were made by medically-trained researchers. Red flags were considered correct if they were similar or described a broader category of symptoms. Participants did not have to specify severity or particular part of the body. Only patients with specific red flags listed on their discharge instructions were included in this analysis (n=1197).

3. Medication adherence: We also asked participants to self-report any medication changes (starts, stops, or modifications) they had made since discharge during the 4-day survey. They had to provide the name, classification, or purpose of each medication. We did not ask participants to report details of dosage, frequency, or timing. Medically-trained researchers compared self-reported medication changes to those listed on participants’ AVS discharge instructions, excluding those with “as needed” instructions. The dichotomized variable indicates whether or not the participant reported making all recommended medication changes. Only participants with medication-related changes on their AVS instructions were included in this analysis (n=457).

Other Measures:

The complete list of variables to be evaluated as potential covariates was determined a priori based upon theoretical and established relationships to the primary independent and dependent variables.²⁵ These included the following:

1. Demographic characteristics: We included self-identified educational attainment, gender, race, ethnicity, living arrangements, and marital status as dichotomous variables (referent categories presented in Table 1). Participant age was included as a continuous variable.

2. Health characteristics: Number of comorbid conditions was measured according to the Charlson Comorbidity Index,³¹ abstracted from the participant medical record. Health status was measured using the general health item of the Short Form-12 general health item.³² Limitations in activities of daily living (ADLs)³³, health literacy³⁴, anxiety (Generalized Anxiety Disorder-2)³⁵, and depressive symptoms (Patient Health Questionnaire-9)³⁶ were assessed during the initial survey. Each characteristic was treated as a binary variable using established thresholds.

3. Cognitive status: We considered participants cognitively impaired if they met any of the following conditions: (1) score >10 on the Blessed Orientation Memory Concentration Test (BOMC)³⁷; (2) reported a diagnosis of dementia or cognitive impairment; or (3) had medical record documentation of a memory-related condition (e.g., dementia).

4. Health-related self-efficacy: The Perceived Health Competence Scale (PHCS)³⁸ was administered in all three surveys to assess self-efficacy regarding one’s ability to influence health outcomes through health-related behaviors. This construct was included as a continuous variable, with possible scores from 8 (low) to 40 (high).

Table 1:

Descriptive Characteristics, Overall and by Treatment Condition

Sample Characteristics	Overall	Control	Treatment (ITT)	Treatment (PP)
N	1756	893	863	726
Age (mean [σ])	72.39 (8.58)	72.10 (8.50)	72.69 (8.65)	72.82 (8.65)*
Sex = Male (%)	818 (46.6)	413 (46.2)	405 (46.9)	336 (46.3)
Race = Non-White (%)	126 ( 7.2)	69 ( 7.8)	57 ( 6.7)	43 ( 5.9)
Ethnicity = Hispanic (%)	27 ( 1.6)	15 ( 1.7)	12 ( 1.4)	9 ( 1.2)
Education = Some College or Less (%)	689 (39.4)	352 (39.6)	337 (39.2)	271 (37.4)
Marital Status = Not Married (%)	713 (40.7)	352 (39.6)	361 (41.9)	300 (41.3)
Number of Charlson Comorbidities (mean [σ])	2.72 (1.67)	2.65 (1.69)	2.79 (1.65) **	2.75 (1.63)
Deficiencies in 1+ ADLs (%)	631 (36.3)	299 (33.9)	332 (38.7) **	280 (38.7) **
Cognitive Impairment (%)	127 ( 7.3)	60 ( 6.8)	67 ( 7.8)	48 ( 6.6)
Health Literacy = Inadequate (%)	214 (12.3)	98 (11.1)	116 (13.6)	91 (12.7)
GAD-2 = Anxiety (%)	293 (16.8)	155 (17.5)	138 (16.1)	115 (15.9)
PHQ-9 = Depression, Moderate to Severe (%)	189 (10.9)	95 (10.7)	94 (11.0)	79 (11.0)
PHCS Score (mean [σ])	30.09 (5.29)	29.87 (5.28)	30.31 (5.29)*	30.43 (5.25) **
SF-12: Fair or Poor Overall Health (%)	355 (20.5)	181 (20.5)	174 (20.4)	138 (19.2)
Had Any Hospitalization in 30 Days Prior to Index ED Visit (%)	90 ( 5.1)	43 (4.8)	47 (5.4)	39 (5.4)
ED Visits in 30 Days Prior to Index ED Visit (mean [σ])	0.11 (0.38)	0.12 (0.40)	0.10 (0.34)	0.10 (0.32)

Significance indicators for each analysis, comparing control to intervention group:

^*p<0.10

^**p<0.05

Age, Number of Comorbidities, PHCS Score, and Number of ED Visits in the prior 30 days are represented as continuous variables. All other covariates are dichotomous.

Abbreviations: ADL, activities of daily living; GAD-2, Generalized Anxiety Disorder-2 (Anxiety); ITT, intention-to-treat analysis; PHQ-9, Patient Health Questionnaire-9 (Depression); PHCS, Perceived Health Competence Scale (health-related self-efficacy); PP, per-protocol analysis; SF-12, Short Form-12 general health item.

Data Analysis

We conducted analyses first using intention-to-treat (ITT) comparisons, followed by a per-protocol (PP) approach. Participants who were randomized to receive the CTI but did not receive the intervention home visit were included in the intervention group for ITT analyses, but not the PP analyses.

Intention-to-treat:

ITT analyses estimated the average treatment effect of assignment to the CTI group on outcomes using multivariate logistic regression. Pearson’s chi-squared and two-sample t-tests were used to compare and identify systematic differences in baseline characteristics between participants randomized to the intervention and control groups. We included variables that differed at p≤0.10 as covariates for the regression models [total number of comorbidities, having one or more functional limitation (ADL), and PHCS score at baseline]. For conceptual consistency, all ITT analyses included the same set of covariates.

Per-protocol:

To evaluate potential treatment effects related to receipt of the intervention itself, rather than assignment to treatment condition, we also conducted PP analyses for each outcome variable. This targeted analysis excluded intervention participants who did not receive the home visit, enabling us to further investigate the effects of the adapted intervention on those who received it. As with the ITT analyses, we compared characteristics of participants who received the intervention with those in the control group, using Pearson’s chi-squared and two-sample t-tests to account for systematic differences, We also compared characteristics of participants randomized to the intervention group who completed the home visit with those who did not (results not shown). Participant characteristics differing by p≤0.10 for any outcome variable were selected as covariates for all regression models. Based upon this process, race, educational attainment, cognitive impairment, and health literacy were included as PP covariates, in addition to the three already included in ITT analyses.

All independent variables were entered simultaneously into ITT and PP regression models. We define statistical significance as p<0.05, reporting all multivariate logistic regression results as adjusted odds ratios (AORs) with 95% confidence intervals. We also report Likelihood Ratio tests, reported as X², for each multivariate regression as a measure of model fit. Prior to running regression models, we conducted several diagnostic assessments of our data including assessments of multicollinearity and influential outliers. One potential covariate, living arrangements (i.e., living alone or with others), was found to be mutlicollinear with marital status and was therefore not included. All analyses were conducted using the R statistical software environment.

Results

Recruitment and Sample Characteristics

As depicted in Figure 2, 6102 (11.3%) of the 53,801 patients age≥60 presenting during recruitment hours were approached for the study. Primary reasons for exclusion were inpatient admission from ED, not having a primary care provider in the health system, and living in a long-term care/assisted-living facility. After additional eligibility checks and refusals, 1979 patients (32.4%) were consented and randomized into the study, 200 of whom were found to be ineligible post-randomization (e.g., admission to the hospital when not originally planned, change of address).

Figure 2:

CONSORT diagram depicting flow of participants through the study. CTI, Care Transitions Intervention; ITT, intention-to-treat; PP, per-protocol

At the conclusion of the study, 1756 subjects were included in the overall data set (893 control, 863 intervention), 1746 of whom had outcomes data abstracted during chart review for use in the primary ITT analysis. The other 10 withdrew prior to chart review, but allowed us to use survey data already collected. 1611 participants had primary outcomes data for use in the PP analysis. We completed 4-day phone surveys with 91.0% of control and 88.8% of intervention group participants. 30-day phone surveys were completed by 88.4% of control and 85.7% of intervention group participants. Participants receiving intervention home visits had higher rates of phone survey completion than either the intervention or control groups overall (95.0% for 4-day and 91.9% for 30-day surveys). Sample characteristics were mostly comparable across groups, significantly differing in functional limitations (having one or more ADL deficiencies), number of comorbid conditions, and PHCS score (Table 1).

Intervention Completion Rates

Of the 863 participants scheduled to receive the intervention, 726 (84%) completed the home visit. Of the 137 participants not receiving the home visit, 24 canceled due to a medical appointment (18%), 17 were at home but too ill to receive the visit (12%), 13 had a work or scheduling conflict (9%), and 43 (31%) did not share a reason. Only 7 (5%) were cancelled by the coach. Participants randomized to the intervention arm who did not receive a home visit were significantly more likely to have cognitive impairment (p<0.001) and fewer years of formal education (p=0.016) than those who did receive the home visit. Of the follow-up calls scheduled (based on coach assessment of need), 95% of first calls (n=684), 92% of second calls (n=536), and 88% of third calls (n=213) were successfully completed.

Primary Outcome: ED Revisits

Overall, 12.4% of participants revisited the ED at least once during the 30-day study period and 2.3% returned multiple times. Unadjusted logistic regression analysis demonstrated no significant differences in ED revisits between the control and treatment groups at either time point (Table 2).

Table 2:

Bivariate Comparisons of Healthcare Use During the 30-day Follow-up Period [n (%)]

Healthcare Use	Control	Treatment (ITT)	Treatment (PP)
Total N	893	863	726
ED Revisits, 14 days	84 (9.5)	69 (8.0)	51 (7.0)
ED Revisits, 30 days	113 (12.8)	103 (12.0)	81 (11.2)
All-contact Outpatient Follow-Up, 7 days	657 (73.6)	664 (76.9)	561 (77.3)
All-contact Outpatient Follow-Up, 30 days	773 (86.6)	775 (89.8) **	653 (89.9) **
In-Person Outpatient Follow-Up, 7 days	428 (47.9)	456 (52.8) **	385 (53.0) **
In-Person Outpatient Follow-Up, 30 days	698 (78.2)	702 (81.3)	592 (81.5)
Electronic Outpatient Follow-Up, 7 days	510 (57.1)	524 (60.7)	445 (61.3)*
Electronic Outpatient Follow-Up, 30 days	659 (73.8)	646 (74.9)	548 (75.5)
Urgent Care Visit, 30 days	24 (2.7)	24 (2.8)	23 (3.2)
Unplanned Inpatient Hospitalization, 30 days	46 (5.2)	45 (5.2)	30 (4.1)

Significance indicators for each analysis, comparing control to intervention group:

^*p<0.10

^**p<0.05

ITT: Intention-to-treat analysis; PP: Per protocol analysis

As depicted in Table 3, we found no statistically significant treatment effects for reducing 30-day ED revisits relative to the control group, either in the ITT (AOR 0.97, 95%CI: 0.72-1.30) or PP analyses (AOR 0.90, 95%CI: 0.66-1.24). This was also true for 14-day revisits. Of interest, participants with cognitive impairment (included as a covariate based upon aforementioned criteria) had approximately twice the odds of revisiting the ED within 14 days (AOR 1.96, 95%CI: 1.05-3.68) and 30 days (AOR 2.12, 95%CI: 1.25-3.61), compared to those without impairment. Greater number of comorbidities (also a covariate) predicted ED revisits at both time points as well.

Table 3:

Adjusted Odds Ratios [95% Confidence Interval] for ED Revisits at 14 and 30 Days, Intention-to-Treat and Per-Protocol Analyses

	Intention-To-Treat		Per-Protocol
Variables	ED Visits (14d)	ED Visits (30d)	ED Visits (14d)	ED Visits (30d)
Treatment	0.87 (0.62, 1.22)	0.97 (0.72, 1.30)	0.76 (0.52, 1.10)	0.90 (0.66, 1.24)
Number of Comorbidities†	1.18*** (1.060, 1.301)	1.20*** (1.10, 1.31)	1.13** (1.01, 1.26)	1.16*** (1.05, 1.27)
1+ ADL Deficiencies† (Present)	1.04 (0.72, 1.51)	1.05 (0.76, 1.44)	1.13 (0.75, 1.69)	1.09 (0.77, 1.53)
Health-related Self-efficacy Score†	0.97* (0.94, 1.01)	0.97** (0.94, 0.99)	0.97* (0.93, 1.00)	0.96** (0.93, 0.99)
Race◊ (Non-white)			1.01 (0.51, 2.01)	0.84 (0.46, 1.55)
Education◊ (Some College or Less)			0.93 (0.63, 1.36)	1.01 (0.73, 1.41)
Cognitive Impairment◊ (Present)			1.96** (1.05, 3.68)	2.12*** (1.25, 3.61)
Health Literacy◊ (Inadequate)			0.70 (0.38, 1.32)	0.91 (0.56, 1.49)
Observations	1,713	1,713	1,572	1,572
Likelihood Ratio Test (X2)	31.18***	44.13***	47.12***	61.76***

^*p<0.10

^**p<0.05

^***p<0.01

Abbreviations: ADL, activities of daily living; ITT, intention-to-treat; PHCS, Perceived Health Competence Scale; PP, per-protocol.

Number of Comorbidities and Health-related Self-Efficacy are represented as continuous variables. All other covariates are dichotomous, with the category assigned a score of 1 listed in parentheses.

^†Covariates for the Intention-To-Treat analyses were chosen based on differences of p<0.10 between control and intervention groups on that characteristic (number of comorbidities, having one or more deficiencies in activities of daily living, and PHCS score). These covariates were also included in PP analyses.

^◊Covariates for the Per-Protocol analyses also included variables with differences of p<0.10 between intervention group participants who did and did not complete the intervention (race, education, cognitive impairment, and health literacy).

Secondary Outcomes: Self-Management Behaviors

Outpatient Follow-Up:

In bivariate analyses (Table 2), we found significant unadjusted treatment effects for 30-day outpatient follow-up in both ITT and PP analyses, as well as in-person visits at 7-days (as part of our pre-planned sub-analysis). ITT multivariate analysis (Table 4) showed no significant treatment effects, with all-contact outpatient follow-up at 30 days (AOR 1.29, 95%CI: 0.95-1.74) and in-person follow-up at 7-days (AOR 1.20, 95%CI: 0.99-1.45) only approaching significance at the p<0.10 level.

Table 4:

Intention-to-Treat Multivariate Analyses for Self-management Outcomes, Adjusted Odds Ratios [95% Confidence Interval]

Variables	All-contact Follow-up (7d)	All-contact Follow-up (30d)	In-Person Follow-up (7d)	In-Person Follow-up (30d)	Electronic Follow-up (7d)	Electronic Follow-up (30d)	Red Flags	Med Adherence
Treatment	1.17 (0.94,1.46)	1.29* (0.95,1.74)	1.20* (0.99,1.45)	1.17 (0.92,1.48)	1.14 (0.94,1.38)	1.02 (0.82,1.27)	1.21 (0.95,1.53)	0.89 (0.60,1.32)
Number of Comorbidities	1.14*** (1.07,1.23)	1.32*** (1.19,1.47)	1.08** (1.02,1.15)	1.22*** (1.12,1.32)	1.11*** (1.05,1.18)	1.23*** (1.14,1.32)	0.96 (0.89,1.03)	0.97 (0.85, 1.10)
1+ ADL Deficiencies	0.91 (0.71,1.17)	0.96 (0.68,1.35)	0.95 (0.77,1.17)	0.99 (0.76,1.30)	0.91 (0.73,1.13)	1.06 (0.83,1.37)	0.75** (0.57,0.97)	0.83 (0.53,1.30)
Health-related Self-efficacy Score	1.01 (0.99,1.04)	1.00 (0.97,1.03)	1.00 (0.98,1.02)	1.01 (0.99,1.04)	1.39*** (1.20,1.63)	0.99 (0.97,1.01)	1.02 (0.99,1.04)	1.01 (0.97,1.05)
Observations	1,723	1,723	1,723	1,723	1,723	1,723	1,197	457
Likelihood Ratio Test (X2)	13.77***	32.30***	6.52*	25.25***	11.63***	40.86***	12.77***	1.82

^*p<0.10

^**p<0.05

^***p<0.01

Number of Comorbidities and Health-related Self-Efficacy are represented as continuous variables. ADL (Activities of Daily Living) is a dichotomous variable, indicating the presence of 1 or more deficiencies.

Abbreviations: ADL, activities of daily living; ITT, intention-to-treat.

PP multivariate analyses (Table 5) demonstrated that participants who received the intervention had significantly greater odds of an in-person visit in the week following ED discharge than control participants (AOR 1.24, 95%CI: 1.01-1.51). Although not statistically significant at p<0.05, intervention participants also demonstrated 1.31 times the odds of all-contact follow-up at 30 days (AOR 1.31, 95%CI: 0.95-1.80, p<0.10). Greater number of comorbidities also significantly increased odds of outpatient follow-up across all ITT and PP models.

Table 5:

Per-Protocol Multivariate Analyses for Self-management Outcomes, Adjusted Odds Ratios [95% Confidence Interval]

Variables	All-contact Follow-up (7d)	All-contact Follow-up (30d)	In-Person Follow-up (7d)	In-Person Follow-up (30d)	Electronic Follow-up (7d)	Electronic Follow-up (30d)	Red Flags	Med Adherence
Treatment	1.20 (0.95,1.51)	1.31* (0.95,1.80)	1.24** (1.01,1.51)	1.20 (0.93,1.54)	1.17 (0.95,1.43)	1.05 (0.83,1.33)	1.34** (1.05,1.71)	0.97 (0.64,1.48)
Number of Comorbidities	1.18*** (1.09,1.27)	1.33*** (1.19,1.49)	1.08** (1.02,1.15)	1.25*** (1.14,1.36)	1.15*** (1.08,1.23)	1.25*** (1.16,1.35)	0.99 (0.91,1.07)	0.99 (0.86,1.13)
1+ ADL Deficiencies	0.93 (0.71,1.20)	0.98 (0.68,1.41)	0.91 (0.73,1.14)	0.95 (0.72,1.27)	0.92 (0.73,1.16)	1.15 (0.88,1.50)	0.84 (0.63,1.10)	0.80 (0.50,1.29)
Health-related Self-efficacy Score	1.01 (0.99,1.03)	0.99 (0.96,1.02)	1.00 (0.98,1.02)	1.01 (0.98,1.03)	1.00 (0.98,1.02)	0.98* (0.96,1.00)	1.01 (0.99,1.04)	0.99 (0.95,1.04)
Race (Non-white)_	0.61** (0.40,0.93)	0.57** (0.33,0.97)	1.14 (0.77,1.69)	0.68* (0.43,1.07)	0.71* (0.48,1.05)	0.62** (0.41,0.96)	0.70 (0.42,1.17)	0.46 (0.16,1.28)
Education (Some College or Less)	0.77** (0.61,0.98)	0.72* (0.52,1.00)	1.05 (0.86,1.30)	0.87 (0.67,1.13)	0.76** (0.61,0.94)	0.74** (0.58,0.94)	0.77* (0.59,1.01)	1.19 (0.77,1.84)
Cognitive Impairment (Present)	0.73 (0.45,1.18)	1.42 (0.66,3.07)	0.71 (0.46,1.10)	1.25 (0.69,2.25)	0.83 (0.54,1.28)	0.87 (0.52,1.43)	0.42** (0.21,0.83)	0.79 (0.29,2.14)
Health Literacy (Inadequate)	0.79 (0.55,1.13)	0.66* (0.42,1.05)	0.88 (0.64,1.22)	0.85 (0.57,1.27)	0.94 (0.67,1.30)	0.83 (0.57,1.20)	0.73*** (0.59,0.90)	0.72 (0.33,1.56)
Observations	1,580	1,580	1,580	1,580	1,580	1,580	1,131	407
Likelihood Ratio Test (X2)	29.63***	41.26***	8.93	32.16***	26.29***	53.60***	39.16***	5.63

^*p<0.10

^**p<0.05

^***p<0.01

Number of Comorbidities and Health-related Self-Efficacy are represented as continuous variables. All other covariates are dichotomous, with the category assigned the number 1 listed in parentheses.

Abbreviations: ADL, activities of daily living; ITT, intention-to-treat.

Red Flag Identification and Medication Self-Management:

We found no significant treatment effects for red flag knowledge or medication adherence in the multivariate ITT analysis (Table 4). However, we did see a significant treatment effect for red flags in the PP analysis (Table 5), as participants receiving the intervention had higher odds of recalling at least 1 red flag from their discharge instructions compared to control (AOR 1.34, 95%CI: 1.05-1.71). We found no significant differences in medication adherence.

Discussion

Although we did not find evidence that the intervention significantly decreased odds of ED revisits within 30 days, per protocol analyses demonstrated positive intervention effects on post-discharge self-management behaviors consistent with hospital-to-home delivery. Our results are similar to other studies examining interventions to facilitate improvements in post-ED discharge outcomes for older patients. A 2019 review by Hughes et al.³⁹ synthesized the results of randomized studies of ED care transition interventions for older adults finding no significant effect on ED revisits. Of the non-randomized studies reviewed, only two showed significant effects (with one demonstrating decreased odds and one increased odds). This parallels other reviews, which found mostly non-significant or increased associations between these ED-based interventions and 30-day revists.^14,20

Across delivery methods and formats, interventions to improve care transitions have been much more successful at improving outpatient clinical follow-up or adherence behaviors/outcomes than ED re-visitation, regardless of the timeframe measured.^15,20,21,40 Aghajafari et al.’s 2020 meta-analysis, involving 20 RCTs of ED-based interventions for improving care transitions,²⁰ found 79% increased odds of outpatient follow-up overall, but no significant treatment effects on ED utilization or hospitalization. Interestingly, studies of hospital-to-home care transitions interventions also have not found a significant impact on ED use,^41,42 even when finding positive effects on re-hospitalization.

Our findings contribute to the care transitions literature in ways that allow us to better explicate the mechanisms through which ED-to-home CTIs can improve continuity of care between acute and primary/outpatient settings. Continuity of care is a key element of successful post-ED care transitions, inherently requiring some form of coordinated follow-up with outpatient providers.² As such, recommended guidelines for geriatric EDs⁴³ include having processes in place to facilitate outpatient follow-up either through direct or technologically-mediated channels.

In this study, participants receiving the CTI had significantly greater odds of in-person follow-up visits during the 7 days following discharge than those receiving usual care. This finding is particularly relevant within the context of emergency care, where it is best practice to recommend "prompt" post-ED follow-up with an outpatient provider.^{16 & 17} Given the relatively-short timeframes specified by ED providers, and traditionally low rates of post-discharge followup by older adults,²¹ the fact that the intervention increased in-person visits with outpatient providers during the week immediately following discharge indicates the usefulness of the CTI in facilitating recommended post-ED clinical care.

Interestingly, we did not find significant differences for 7-day electronic follow-up or in-person follow-up over the longer 30-day period. Participants receiving the CTI did have 31% greater odds of outpatient follow-up within 30 days, yet this result failed to meet the threshold for statistical significance. It does, however, support the overall pattern of positive effects of CTI delivery. Lack of statistical significance may be due in part to very high rates of follow-up in the study population overall (86.6%). It could also be compounded by the variance accounted for by two covariates predicting significantly lower likelihood of electronic visits (race and education), but not in-person visits. Understanding possible disparities in the use of electronic health communication systems is a question in need of further exploration.

Participants receiving the intervention also had significantly greater knowledge of red flags related to their index ED visits—information provided to all patients at discharge and in their written after visit instructions. This illustrates the potential of the CTI to improve upon the inability of many patients to recall information provided about treatments and self-management following ED visits.^44,45 Although the CTI home visit led to improvements in red flag recall, it did not have similar effects on medication adherence. This may have been due to the smaller number of participants who were prescribed medication changes (necessary for inclusion in the analysis), or that medication adherence required patients to engage in behavioral change in addition to recalling information, making it a more complex process to improve.

One key piece of missing evidence is the extent to which improved performance of these three self-management behaviors contribute to use of emergency care services by older adults, and how that differs based upon patient characteristics. We do not know whether increased in-person outpatient follow-up and/or knowledge of red flags had any downstream effects on acute care utilization for our study participants, or if these affected any other patient-centered outcomes that might be related to future care-seeking behaviors (e.g., satisfaction, information sharing). Currently, there is a paucity of evidence supporting the assumption that care transitions behaviors directly translate to meaningful health outcomes related to future ED use—an issue that needs to be remedied in future studies.

The central question coming from this study, however, is why we did not find any effect on ED revisits within the 30 days following discharge. CTI delivery assessment monitoring²⁶ eliminates a lack of intervention fidelity as a possible explanation. This is further supported by the demonstrated treatment effects for targeted self-management behaviors. One possibility is that the characteristics of our recruitment sites limited CTI effectiveness. Both healthcare systems are strong accountable care organizations (ACO), which, due to their integrated structure and allocation of staff to care coordination activities, could reduce the added value of the CTI.⁴⁶ Our overall 30-day ED revisit rate (12.8%) was below those reported elsewhere, possibly due to ACO policies and/or care-coordination programs.

The second possibility is that our target population was too broad. Some researchers^12,18,29 contend that care transition interventions may only prove effective when targeted to those at highest risk of return or experiencing care fragmentation, such as those with advanced age, medical complexity, cognitive impairment, mental health issues, low health literacy, low social support.^{2,6,10,19,42,47,48} Others argue that interventions targeting patients with modifiable risk factors would be more effective than using an undifferentiated approach.^11,49,50 Thus, if our goal is to decrease ED revisits, programs should identify the specific populations for which interventions may have that desired effect and target them.

The third possibility is that the underlying assumption that successful care transitions programs should always result in decreased ED revisits for all older adults, without regard to population heterogeneity, is false. A differentiated patient-centric approach would allow us to embrace evidence that care transition programs may result in increased revisits for some patients once care is optimized (e.g., those with greater medical complexity), without branding that as a failure or null finding.^14,22,51 The results of multiple recent studies strongly indicate that older patients who seek prompt outpatient follow-up may have acute symptoms identified earlier than those who did not, resulting in ED revisits driven by recommendations/referrals from primary care providers.^11,12,15,52 Study authors suggest that although ED revisits were deemed necessary by outpatient providers due to worsening or new onset acute symptoms, these may have been caught early enough to prevent the need for more advanced (and costly) inpatient care.^12,52 This could likely be similar to what occurs when patients better understand their red flags, allowing them to easily self-identify potential problems and seek ED care¹⁴ before symptom severity increases. More research needs to be conducted to better understand the directionality of utilization patterns so interventions can be targeted and appropriately measured.

Extending that point leads us to question whether 30-day ED revisits is really an appropriate or meaningful outcome measure for assessing quality care transitions. Unplanned revisits frequently result from disease progression or symptom exacerbation,⁴⁹ or involve non-modifiable risk factors related to chronic conditions or particular types of diagnoses/chief complaints, making them difficult to prevent. We may need to expand our definition and measurement of successful care transitions outcomes, potentially including more comprehensive, patient-centered, and/or longitudinal utilization approaches, and evaluate programs against those measures.^2,18 For instance, recent studies discovered that while ED revisits did not decrease, patients who did return tended to have lower-severity symptoms and/or decreased post-ED hospitalization rates than those receiving usual care.^4,13,52 Studies examining reasons for return, in addition to rates of return, also allow us to better determine the most appropriate outcome measures for different types of patients.⁵³ For example, other outcomes could include functional status, quality of life, patient experience, or use of other health-related services in the community.^{4,14,15,39,42,54}

We have no way of distinguishing how each of these possible factors may have contributed to our results. For now, they collectively suggest that we should focus on determining which interventions work best in which populations, as well as the most appropriate outcome measures, rather than focusing on average decreases in 30-day ED revisits.

Future Directions

There are a number of additional questions to answer and considerations to address prior to implementing the CTI for use following ED discharge. The first would be identifying the types of older adult patients (e.g., health conditions, demographic characteristics, care utilization patterns) who might benefit the most from this type of program. Although our current results may generally suggest some characteristics of potential targets (e.g., those with cognitive impairment), actual subgroup identification will require more nuanced analytic strategies (e.g., latent class analysis). Once those have been identified, implementation strategies would be tailored and rigorously tested to ensure they meet the needs of these targeted groups.

Patients are not the only stakeholders impacted by the design and implementation of care transitions interventions. Organizational-level factors also need to be considered in evaluating potential intervention adoption and sustainability moving forward. For instance, even if providing home visits is feasible for an organization, it can be a very resource intensive strategy both in terms of time and cost. It also constrains the reach of the intervention based upon travel time, limiting access to patients living in hard-to-reach or distant locations. Future research should examine the potential benefits and costs of the CTI compared to other interventions offering the similar content using different modes of delivery (e.g., telehealth approaches).

Limitations

In addition to aforementioned possible limitations, two other limitations are worth mentioning. First, as we only included participants with primary care providers within these two ACO healthcare systems, excluding those without established primary providers or who lacked sufficient insurance to cover primary care in these organizations, it is likely that our rates do not reflect ED utilization patterns typical for older adults in these EDs. Second, this study took place in two mid-sized urban environments with surrounding rural areas, and therefore may not be as generalizable to other communities with different composition and diverse populations. Although there were no statistical differences in the racial/ethnic composition of the study arms, there was an overall lack of diversity in these characteristics. This, along with the exclusion of non-English speakers, limited our ability to meaningfully assess trends related to non-English speaking populations, which could reflect an area of concern for future CTI implementation.

Finally, it is important to note that results of per-protocol analyses should be interpreted with caution, as the exclusion of randomized participants not receiving the intervention allows for sampling bias otherwise mitigated using intention-to-treat strategies. We felt that these results were important to include within the context of this particular study, however, as they help demonstrate the potential effectiveness of the CTI for use following ED discharge outside the controlled design of a clinical trial, and in particular the extent to which CTI delivery by community paramedics in this clinical setting can influence the self-management behaviors targeted by the intervention.

Conclusion

The adapted CTI, delivered to community-dwelling older adults following ED discharge, did not significantly reduce ED revisits within 30 days. The intervention did significantly increase outpatient clinical follow-up in the week following discharge and improve red flag identification, two key care transition behaviors specifically targeted during CTI delivery. Additional research is needed to explore whether patients with different conditions or characteristics benefit more from the CTI, whether decreased ED revisits should always be the desired outcome, and whether other outcome measures are potentially more meaningful to patients and clinicians.