Process Evaluation of Mi Puente (My Bridge): A Care Transitions Program for Hispanic/Latino Adults With Multimorbidity

Abstract

OBJECTIVE |

The Mi Puente (My Bridge) randomized controlled trial compared a culturally appropriate, team-based discharge program versus usual discharge procedures (usual care) in reducing readmission rates (inpatient, emergency, and observation visits) among Hispanic/Latino adults with multimorbidity. Primary outcome analyses revealed the Mi Puente group demonstrated increased hospital use relative to usual care. To better understand these unexpected effectiveness findings and inform future research, we conducted a multimethod process evaluation using components of the Reach, Effectiveness, Adoption, Implementation, and Maintenance framework.

RESEARCH DESIGN AND METHODS |

Reach was assessed by examining the proportion of eligible participants enrolled and representativeness of the sample. Adoption was assessed via semistructured interviews with study staff. Implementation was examined using study fidelity forms (e.g., protocol adherence and duration/content of visits) and whether those who received the intervention per protocol differed from those who did not.

RESULTS |

Mi Puente reached the target population, with high acceptance among eligible patients enrolled (75%), although participants tended to be younger, preferred English, and endorsed more behavioral health concerns. Adoption interviews and related Implementation findings revealed gaps in both the intervention content delivered and the availability of resources within San Diego County. Approximately 70% of participants received the intervention per protocol.

CONCLUSIONS |

Multilevel strategies are needed to improve post-discharge outcomes in Hispanic/Latino adults with multimorbidity. These should include improved integration of inpatient and outpatient settings, the use of warm handoffs and home visits, ongoing clinical support, and implementation of structural-level interventions to address social determinants of health.

Graphical Abstract

Hispanic/Latino adults are disproportionately affected by chronic cardiometabolic conditions (e.g., obesity, diabetes, and hypertension) and behavioral health conditions compared with non-Hispanic White adults (1–3). They also experience poorer disease outcomes, including greater risk of complications and recurrent hospitalizations, increasing burden on patients and the health care system (4,5). Reducing preventable hospital readmissions through adherence to discharge instructions and outpatient care connection can alleviate this burden. However, individual, social, and structural barriers (e.g., language barriers, access issues) complicate hospital-to-home transitions for Hispanic/Latino adults, making discharge instructions difficult to follow (6). System-level barriers, including poor care integration and lack of culturally and linguistically concordant care, further exacerbate these challenges, highlighting the need for culturally tailored care transition interventions for this population.

A systematic review of 40 emergency department–based care transitions interventions found they did not reduce emergency department return visits but increased primary care follow-up (7). Improved outpatient follow-up occurred with structured telephone contact/reminders, emergency department–based appointment scheduling, and verbal/written education, emphasizing follow-up importance (7). Another review found that while only 41% of studies successfully reduced hospital readmissions, these programs were more likely to use specific components of The Ideal Transition of Care framework, including monitoring/managing symptoms, community and social support integration, and self-management promotion (8). Another review found personalized discharge planning modestly reduced hospital stays and unscheduled readmissions versus standard discharge procedures (9). Despite a demonstrated need, few studies have focused on improving the care transitions process for low-income, minoritized racial and ethnic populations.

The Mi Puente: My Bridge to Better Cardiometabolic Health and Well-Being (“Mi Puente”) randomized controlled trial compared standard discharge procedures (usual care) with a culturally appropriate, team-based discharge program combining inpatient and telephone support (Mi Puente) among 536 Hispanic/Latino adults with multiple cardiometabolic conditions and at least one behavioral health concern who were hospitalized at a large hospital in South San Diego County (10,11). Usual care participants received standard hospital discharge procedures (written instructions, appointment scheduling, referrals, and other assistance as needed), while Mi Puente participants received standard care plus support from a bilingual, bicultural team comprising a behavioral health nurse and community mentor (intervention agents). The nurse component, delivered by a registered nurse holding a bachelor’s degree, occurred in person prior to discharge, by telephone within 1 week post-discharge, or through a combination of both modalities, depending on patient availability and discharge timing. This component included a needs assessment, creation of the My Personal Health Record (MPHR) (Gallo et al. [10]), goal setting and action planning, medication review, health education, discussion of symptom red flags, and assistance with outpatient follow-up and referrals. When possible, the nurse conducted a warm handoff to the mentor during the inpatient visit. Mentors were volunteers with shared lived experience who provided post-discharge support through weekly telephone calls for 2–4 weeks, reinforcing care plans using motivational discourse, and relying on a resource manual developed with local community partners to address patients’ social needs. Participants with needs beyond the mentor’s scope were referred to community partners and other outside organizations for additional support (see Gallo et al. [10] for more details).

Contrary to predictions, the Mi Puente intervention group experienced significantly higher hospital use at 30 and 180 days versus usual care post-discharge (11). Specifically, the Mi Puente group had a readmission rate of 15.9% vs. 9.4% in the usual care group (odds ratio 1.91, 95% CI 1.09, 3.33). The average number of recurrent hospitalizations was also significantly higher in the Mi Puente group (mean 0.20 [SD 0.49]) compared with usual care (mean 0.12 [SD 0.45]; P = 0.02). Similar trends persisted at 180 days after discharge. Both groups showed improvements in patient-reported outcomes (e.g., quality of life, patient activation), with no significant advantage observed for the Mi Puente intervention (11). Limitations and unique features potentially influencing these findings are described in the primary outcomes article (11).

To better understand these unexpected findings, the current study presents a comprehensive process evaluation of the Mi Puente intervention and study. Specifically, we applied components of the Reach, Effectiveness, Adoption, Implementation, and Maintenance (RE-AIM) model to identify potential mechanisms and implementation factors that may explain these paradoxical results. The RE-AIM framework is widely used for evaluating multifaceted components of health interventions, as well as dissemination and translation potential (12,13). This framework defines reach as the proportion of enrolled among eligible patients and representativeness of enrollees relative to the target population; effectiveness as intervention impact on key outcomes; adoption as the proportion of settings and/or intervention agents willing to implement the program and reasons for participation; implementation as fidelity to intervention protocols; and maintenance (at the setting level) as the extent to which a program becomes part of routine organizational practices. While the primary outcomes article described effectiveness results (11), the current study provides important context to the unexpected effectiveness findings to guide the development of future care transitions interventions for this at-risk population.

Research Design and Methods

Quantitative and qualitative methods were used to evaluate Mi Puente’s success across multiple RE-AIM domains, to contextualize the unexpected effectiveness findings, identify areas for improvement, and provide suggestions for future research based on lessons learned. The following section provides details about data collection methods and analyses for each of these domains.

Reach

Demographics (age, sex), clinical data (e.g., number of cardiometabolic conditions), LACE index (predictor of readmission and mortality comprising length of stay, acuity of admission, comorbidities, and emergency department use), Charlson comorbidity index (CCI) (an index used to predict mortality using information about comorbidities) (14), and behavioral health data (e.g., number of behavioral health concerns) were extracted from the electronic health record for all eligible patients (i.e., those identified as clinically eligible via electronic health record reports and study screener).

To assess reach, we calculated the proportion of enrolled eligible patients among screened-eligible patients. Representativeness was examined by comparing demographic and clinical characteristics of enrollees versus the overall eligible pool using t tests and χ² tests. These statistical tests also compared enrollees with those who declined participation to identify characteristics associated with enrollment decisions.

Effectiveness

The effectiveness of Mi Puente in reducing readmissions and improving patient-reported outcomes, as well as the approach used for these evaluations, have been previously reported (11) and are described briefly in this report in the context of the other RE-AIM domains.

Adoption

In this study, adoption refers to staff-level uptake rather than at the setting level. Trained research staff conducted semistructured interviews with intervention agents (nurse, mentors) to examine staff-level adoptability. Interviews were conducted with the sole nurse employed throughout the study and with 10 of 15 community mentors within 1 year of study conclusion. The five noninterviewed mentors were employed early in the study and could not be reached or declined participation. The interview script explored interventionists’ experiences, barriers, and facilitators.

Given only one nurse participated, responses were narratively summarized. Mentor interviews were analyzed using thematic analysis focusing on warm handoff/nurse collaboration, resources, engagement, call dosage, and overall experience. Two independent reviewers coded interviews using a developed coding scheme; discrepancies were resolved by a third party until consensus was reached. Highly endorsed responses across interviews were identified.

Implementation

Study fidelity forms documented completion, timing, duration, and topics covered for both the nurse and mentor intervention components. Receiving the intervention “per protocol” was defined as completing the 1) initial nurse visit within 1 week of enrollment (ideally inpatient or by telephone if discharged early) and 2) at least two mentor calls (but up to a possible four) within 30 days of discharge.

Protocol adherence was assessed by calculating the percentage of participants receiving both components of the intervention as intended (“per protocol”). We used t tests and χ² tests to examine whether those who did and did not receive the intervention per protocol differed by age, sex, LACE index, CCI, number of cardiometabolic conditions, or behavioral health concerns. Intervention visit duration and content are also described.

Maintenance

Since Mi Puente did not reduce hospital readmissions as expected, program sustainability in routine organizational practices was not examined. However, lessons learned and recommendations for addressing future patient needs are integrated in the Conclusions.

Ethics Approval and Consent to Participate

This study was reviewed and approved by the Scripps Health Institutional Review Board, San Diego, CA (IRB-15–6585).

Data and Resource Availability

The data and materials that support the findings of this process evaluation are available from the corresponding author upon reasonable request. This includes relevant data sets, interview transcripts, and other materials used in the study.

Results

Participant Characteristics

Baseline characteristics of the total sample and each randomization group are provided in Table 1. The average age was 62.3 (SD 13.2) years, and 48% were female. The household income for 77% was <$30,000/year, 77% preferred Spanish, and 48% had less than a high school education. Most were born in Mexico (73%), had health insurance (93%), and were unemployed (80%). Mi Puente participants had significantly higher educational attainment than usual care participants (P < 0.05); no other between-group differences in sample characteristics were observed.

TABLE 1.

Sample sociodemographic and health characteristics at baseline, overall, and by group

	Total sample (N = 536)	Usual care (n = 265)	Mi Puente (n = 271)
Sociodemographic characteristics
Age, mean (SD), years	62.3 (13.2)	62.7 (12.2)	61.9 (14.1)
Sex
Male	279 (52.1)	141 (53.2)	138 (50.9)
Female	257 (47.9)	124 (46.8)	133 (49.1)
Education
Less than high school education	247 (48.1)	134 (53.2)	113 (43.3)
High school education or higher	266 (51.9)	118 (46.8)	148 (56.7)
Income
<$30,000/year	414 (77.2)	207 (78.1)	207 (76.4)
≥$30,000/year	80 (14.9)	38 (14.3)	42 (15.5)
Employment
Employed	106 (20.5)	50 (19.7)	56 (21.3)
Not employed	411 (79.5)	204 (80.3)	207 (78.7)
Place of birth
U.S.	137 (25.6)	58 (21.9)	79 (29.2)
Mexico	390 (72.8)	204 (77.0)	186 (68.6)
Other	9 (1.7)	3 (1.1)	6 (2.2)
Language
Spanish	412 (76.9)	212 (80.0)	200 (73.8)
English	124 (23.1)	53 (20.0)	71 (26.2)
Insurance
Uninsured	39 (7.3)	17 (6.4)	22 (8.2)
Insured (any)	495 (92.7)	248 (93.6)	247 (91.8)
Health characteristics
Total qualifying cardiometabolic conditions, mean (SD), n	3.33 (1.4)	3.35 (1.4)	3.31 (1.3)
Total qualifying behavioral health concerns, mean (SD), n	2.98 (1.6)	3.03 (1.7)	2.93 (1.6)

Data are presented as n (%) unless indicated otherwise. Sample sizes differ due to missing data for individual variables.

Reach

Study eligibility was determined using daily electronic health record lists and a staff-administered screener to assess the presence of behavioral health concerns and telephone access. Between July 2016 and March 2020, 5,064 patients met initial study criteria based on electronic health record lists, of which 45% (2,279) were screened, and 55% (2,785) were not screened due to patient factors (e.g., patient unavailable, patient was sleeping or in pain during recruitment, discharged early, declined screening) or staff factors (e.g., staff unavailable to screen patient). Of those screened, 68% (1,560) were ineligible, and 32% (719) were eligible. Of those eligible, 16% (116) were discharged before randomization, and 9% (67) declined participation. Common reasons for declining participation included not feeling well enough to continue with enrollment, not wanting “extra help,” and feeling that participation would be burdensome. The remaining 536 of potential participants were enrolled, representing a reach of 75% (536 of 719) among screened and eligible patients. Figure 1 presents the Consolidated Standards of Reporting Trials (CONSORT) diagram for the trial.

FIGURE 1.

Consolidated Standards of Reporting Trials (CONSORT) diagram. Reproduced from Gallo et al. (11).

Table 2 presents descriptive information for eligible participants, those who were enrolled in each of the study groups, did not enroll, and declined participation. Compared with those who did not enroll (due to declining or other reasons), those enrolled were significantly younger (mean 62.4 [SD 13.2] age vs. 67.2 [SD 13.2] years), were less likely to prefer Spanish (76.9% vs. 86.3%), and had more behavioral health concerns (mean 3.0 [SD 1.6] vs. 2.5 [SD 1.5]; P < 0.01 for all). Patterns were similar for those who declined and those who enrolled (Table 2).

TABLE 2.

Demographic and clinical variables for eligible participants who are enrolled, are not enrolled (includes declined), and declined participation

Variables	Eligible participants (N = 719)	Enrolled (n = 536)	Not enrolled (n = 183)	Declined participation (n = 67)
Age, mean (SD), years	63.6 (13.4)	62.3 (13.2)*	67.2 (13.2)*‡	66.3 (13.9)‡
Female sex, n (%)	351 (48.8)	258 (48.1)	93 (50.8)	37 (55.2)
Spanish preference, n (%)	570 (79.3)	412 (76.9)*	158 (86.3)*	57 (85.1)
LACE index, mean (SD)	7.4 (2.4)	7.4 (2.3)	7.3 (2.5)	7.4 (2.3)
CCI, mean (SD)	3.6 (2.5)	3.4 (2.5)	3.8 (2.6)	4.0 (2.9)
Cardiometabolic conditions, mean (SD), n	3.4 (1.4)	3.3 (1.4)	3.5 (1.4)	3.6 (1.5)
Behavioral health concerns, mean (SD), n	2.8 (1.6)	3.0 (1.6)*	2.5 (1.5)*‡	2.7 (1.5)‡

Those who declined participation are a subset of those in the not enrolled category. Eligible participants include those enrolled and not enrolled. LACE index is used to predict readmissions and mortality. CCI categorizes and weighs comorbidities to predict mortality and resource use.

^*Statistically significant difference between enrolled and not enrolled (P < 0.01).

^‡Statistically significant difference between enrolled and declined participation (P < 0.05).

Adoption

The nurse interview revealed that the role was well accepted, with the nurse noting that she enjoyed connecting with patients to help them reach their goals. She reported there were some instances where she would have liked to spend more time with participants with complex needs. When asked which intervention component participants engaged with most, she said making a plan to schedule a follow-up visit with the provider. Contrarily, she expressed patients struggled most with medication reconciliation, mentioning that “Some patients did not have proper reading glasses, some did not have proper caregiver support, some did not know what they were taking medications for … for patients with any of these concerns, I would recommend a home health visit to conduct medication reconciliation in-person.” Overall, however, she reported the intervention content was comprehensive and relevant to the patient population. While it was difficult to engage participants who were in a lot of pain or drowsy from medications, most patients were highly engaged and thankful for the additional support.

Mentor interview results are summarized in Supplementary File 1. Of the 10 interviews conducted, 90% of mentors said the in-person, nurse-mentor warm handoff was helpful for participants, improved engagement, and/or helped establish the mentor as part of the team, although only 40% found handoffs feasible. Despite 60% reporting “great” communication with the nurse, mentors reported logistical issues (e.g., discordant schedules, locations) that hindered successful handoffs. The resource manual, comprising several resources in San Diego County (e.g., food banks, low-cost transportation services, family resource centers, etc.), was described as very helpful (100%) and easy to use (90%), with 70% perceiving resources as helpful to participants. However, participants faced numerous access-related challenges, including ineligibility, language barriers, unreturned calls, resource unavailability, and difficulty completing required paperwork due to literacy issues or lack of access to technology. Mentors reported housing and transportation were most challenging due to insufficient San Diego County options.

Among the mentors, 90% rated participant engagement at 7 or higher (0–10 scale). Good mentor-participant rapport enhanced engagement, while confusion about call purpose, patient-level factors (financial hardship, resource access difficulties), and scheduling challenges hindered rapport. Regarding dosage, 80% of mentors said two to four calls were appropriate, although 60% felt severe or complex cases needed additional time. Overall, 90% described their experience as “good” or “great,” with most enjoying building relationships with and supporting participants.

Implementation

Study fidelity forms revealed that 69.7% (189 of 271) received the intervention per protocol. Only 8.1% of participants did not receive the nurse component as intended. Nearly 25.5% did not receive the mentor component as intended, largely due to the mentor’s inability to reach participants by telephone. There were no statistically significant differences between those who did and did not receive the intervention per protocol according to age, sex, LACE index, CCI, or number of cardiometabolic or behavioral health conditions (P > 0.05 for all; data not shown).

The average duration of the initial nurse visit was 59 (SD 25) min. The nurse covered seven major topics (Table 3), and within these were multiple subtopics (Supplementary File 2). A topic was considered to have been formally discussed if the nurse covered one or more subtopics within that category. The most commonly covered topics were follow-up with the primary care physician (96.3%), medications (95.9%), action planning (95.9%), the MPHR (95.6%), health education (94.1%), referrals (91.5%), and symptom red flags (88.2%). Additional information on commonly discussed subtopics is presented in Supplementary File 2. Of the participants who agreed to create an action plan (249 [91.9%]), 54.6% focused on attending routine medical visits, 48.6% on managing chronic health condition(s), and 47% on managing chronic stress, with many participants creating action plans targeting multiple of these goals.

TABLE 3.

Topics covered by nurse and mentor

Topic	Description	Discussed with nurse	Discussed with mentor
Referrals	Made necessary referrals (e.g., to wellness center and case management)	248 (91.5)	173 (63.8)
Symptom red flags*	Discussed when to call PCP vs. what constitutes an emergency situation	239 (88.2)	—
Action plan	Collaborated with patient to complete My Action Plan	260 (95.9)	217 (80.1)
Follow-up with PCP	Discussed follow-up appointment, which included encouragement to follow through and role-play	261 (96.3)	228 (84.1)
MPHR	Collaborated with patient to complete MPHR	259 (95.6)	216 (79.7)
Medications*	Answered questions, encouraged logging, and conducted medication reconciliation	260 (95.9)	—
Health education*	Provided health education, which included answering questions	255 (94.1)	—
Transition home‡	Discussed transition home and any barriers met/expected	—	226 (83.4)

Data are n (%). PCP, primary care provider.

^*Topic meant to be covered by nurse only.

^‡Topic meant to be covered by mentor only.

On average, mentor follow-up calls were 12.6 (SD 7.2) min. The first call was within 15 (SD 11.3) days of discharge, and the second call was within 24 (SD 9.4) days of discharge, which was in line with the intervention design. Common topics covered by mentors included encouraging follow-up with primary care physicians (84.1%), the hospital-to-home transition (83.4%), and following up on action plan progress (80.1%). While addressed with most participants, referrals to outside resources (e.g., wellness center, case management) were less frequently discussed (63.8%) (Table 3).

Conclusions

The current study conducted a multimethod process evaluation of the Mi Puente intervention, which sought to improve the hospital-to-home care transition among low-income, Hispanic/Latino adults with multiple cardiometabolic conditions and behavioral health concerns. To better understand the unexpected finding of increased readmissions among Mi Puente versus usual care participants (12) and to inform future research, we examined the reach, adoption, and implementation of the Mi Puente program. This study addresses an important literature gap by examining effective care transitions for populations affected by adverse social determinants of health.

Among screened patients in the current study, 75% of eligible patients were enrolled in the trial, indicating a high level of “reach.” Two systematic reviews of care transitions and discharge planning interventions across various medical settings and populations (7,9) found that among nurse-led, U.S.-based studies published since 2000, the average “reach rate” among eligible patients was 74.7% (SD 17.3; range 58.7−98.8). While direct comparisons are cautioned due to variability in intervention approaches, settings, and participant characteristics, Mi Puente’s enrollment rate was similar to that of prior studies, indicating the program was well received by the target population. However, significant differences emerged in key health and demographic factors between those who enrolled versus those who were eligible and did not enroll. Specifically, enrolled participants were younger, less likely to prefer Spanish, and reported more behavioral concerns. Several factors may explain these patterns. First, patients with greater behavioral concerns may have been more motivated to participate due to higher perceived need for support in managing anxiety, depression, chronic stress, or chronic health problems. Second, nonenrollees who preferred Spanish may have identified more closely with Hispanic/Latino cultural values and faced cultural stigma around behavioral health (15), which could have deterred participation. Third, older patients may have been less likely to enroll due to more compromised health or hesitancy about participating in research. Conversely, some eligible patients who declined may have preferred or already had other support systems in place, such as family networks, or perceived themselves as needing less support given fewer behavioral health issues.

Adoption and implementation results provide some insight into the primary trial’s unexpected effectiveness findings and suggest important considerations for future studies. While the proportion of enrolled patients among those initially identified by the EHR may appear low (11%), it reflects the practical challenges inherent in implementing care transitions interventions in hospital settings, particularly when secondary screening is needed to better identify at-risk patients. Our study operated with limited, part-time staffing, rather than dedicated full-time personnel available 24/7 to recruit eligible patients. Consequently, recruitment staff could not identify and approach potentially eligible patients during off-hours, weekends, or periods of high patient volume. Patient-level factors further constrained enrollment opportunities. This included patient unavailability due in part to procedures or tests during hospitalization, patients asleep or in pain during recruitment windows, and unpredictable discharge time that left insufficient time for enrollment. Other exclusion criteria, such as lack of telephone access, may have prevented some patients from participating even when they would have benefited from the intervention. Future interventions should consider dedicated full-time recruitment staff with extended coverage hours, flexible enrollment procedures that accommodate patient circumstances (e.g., providing cell phones during study duration), and alternative communication methods to maximize reach and ensure that all eligible patients who could benefit have the opportunity to be screened and potentially participate.

Implementation data also confirmed the nurse discussed primary care follow-up with nearly all participants and successfully helped them create post-discharge plans. However, participants’ actual engagement in outpatient care was not objectively captured (i.e., from medical records), and no differences in self-reported outpatient visits emerged between Mi Puente and usual care groups during the follow-up period (11). Several factors likely impeded follow-through on outpatient care plans.

First, the outpatient clinics were not directly connected to our hospital system. We collaborated with a large Federally Qualified Health Center system serving South San Diego, but many patients could not be connected to this system due to insurance differences, existing primary care relationships, or patient preferences. For patients without established providers, we provided community referrals based on insurance and geography, but these represented external handoffs rather than seamless transitions within an integrated system.

Second, although the nurse prioritized discussion of outpatient care, several social and structural barriers may have impeded follow-through, such as a lack of transportation, caregiving/work responsibilities, and language barriers (6). While mentors were instructed to provide resources to address these barriers during support calls, many resources were often difficult to access and/or insufficient to meet participants’ needs. This combination of fragmented care coordination and unresolved access barriers likely undermined the intervention’s potential effectiveness despite discharge planning discussions.

To ensure outpatient care is obtained as intended, future studies might consider scheduling primary care follow-up appointments before discharge as recommended by a prior review (7), although patient follow-through is likely to remain limited by persistent social and structural obstacles. Other reviews found strong relationships and “community contracts” between hospitals and primary care clinics can reduce readmissions, particularly when facilities have the infrastructure to seamlessly exchange patient health information, communicate, and collaborate during and after hospitalizations (16–18). Primary care represents the “new frontier” for efforts to reduce readmissions, yet financing regulations, misaligned incentives, and fragmented health care systems fundamentally limit primary care’s capacity to coordinate this level of complex, post-discharge care (16). For primary care to adopt this process, policies must allocate resources to enhance infrastructure and support primary care staff, which may require a large-scale primary-care transformation, similar to the patient-centered medical home framework (16). Although this approach is ambitious, it may be integral in bridging the inpatient-outpatient care gap and reducing hospital readmissions.

Implementation analysis revealed the nurse addressed symptom red flags with most patients. However, for participants with complex needs or presentations, longer inpatient visits and additional, ongoing clinical support may have been necessary to cover more than the basic needs and discuss nuances. Discussion of symptom red flags aimed to equip patients with knowledge for managing symptoms. However, these discussions may have inadvertently increased participants’ symptom awareness and sensitivity, leading to greater emergency service use among Mi Puente participants, particularly given prominent barriers to outpatient care in this population (11). Among The Ideal Transition of Care domains, post-discharge symptom monitoring and management have been most consistently linked with reducing readmissions, although successful studies typically used in-home visits by health care professionals rather than telemonitoring (19). Additional clinical support after discharge, including home visits, may be important for future interventions.

Although the nurse discussed certain medication-related subtopics with most participants, medication reconciliation was less common due to factors including time constraints, participants’ confusion about their medications, and lack of caregiver support. Patients with multimorbidity are often prescribed multiple medications from different health care providers and specialists (referred to as polypharmacy). When excessive or unnecessary, polypharmacy has been linked to poor health outcomes and increased risk for hospital readmission, particularly in older adults (20–22). Since the study sample was older and had multiple cardiometabolic and behavioral health conditions (and likely multiple prescribing providers), thorough medication reconciliation warranted more time and attention but was not feasible in our model, given priorities to increase sustainability by keeping costs low. Home visits to conduct medication reconciliation could address this challenge and have been effective in prior studies (8,19,23). Future studies should also consider using pharmacist support, which has been shown to reduce readmission rates significantly (24–26). Health information exchange, or the ability for providers across systems to access patient health information such as medications, could further enhance these efforts moving forward.

Implementation findings from the mentor interviews revealed important considerations for future care transitions interventions. Approximately one-fourth of our study sample did not receive the mentor component as intended, and calls were generally conducted in the expected time frame but were often shorter than expected (<15 min). Mentors felt dosage was sufficient but additional calls would have been beneficial in some cases, consistent with the nurse report. Future interventions might consider a more intensive approach for patients with more comorbidities and/or more adverse social factors. Furthermore, the mentors were mostly unpaid volunteer undergraduate students seeking research/health care experience. Professionals, such as licensed clinical social workers or community health workers, may have been better suited for the mentor role given their expertise in helping patients problem-solve through social barriers and their direct experience with serving their communities in a professional role. However, the higher costs associated with employing professionals would have limited the program’s ability to remain cost-effective, reach a broader group of at-risk patients, and be sustainable long-term. Future studies should explore alternative payment models that incentivize and support these types of roles within the health care system to enhance both program effectiveness and sustainability (27). Furthermore, mentors found the warm handoff with the nurse to be helpful, but logistical barriers often complicated this process. While limited, there is evidence that warm handoffs improve client follow-up with referrals to various services (28,29). Although the impact of warm handoff completion on implementation outcomes was not examined, nearly 25% of participants did not complete the mentor portion as expected, and challenges with warm handoffs may have contributed to this.

These implementation findings have important implications for interpreting intervention mechanisms and outcomes. Approximately 30% of participants who did not receive the intervention per protocol experienced barriers at multiple levels. First, the fast-paced hospital environment may have created variable conditions that affected our ability to deliver the intervention as planned, with competing clinical priorities and unpredictable discharge timing limiting intervention opportunities. Second, trial staffing constraints, including part-time research staff, volunteer mentors, and lack of 24/7 hospital presence, reduced our ability to reach all participants consistently. Third, patient-level factors, such as disconnected telephones, failure to answer calls, and acute health conditions and/or psychosocial stressors, prevented successful contact and intervention delivery. While we found no significant differences in measured baseline characteristics between participants who did and did not receive the protocol as intended, factors that were not measured, such as housing stability, social support networks, and social vulnerability, may have influenced participant’s ability to engage with the intervention and could have independently affected readmission outcomes. Furthermore, short call durations, variable handoff completion, and limited professional experience may have constrained the support provided by the mentor. The intended mechanism likely relied on adequate time for rapport building, needs assessment, and resource connection, which are processes that may have been compromised given the aforementioned challenges. Future studies should measure social determinants of health that may predict successful intervention delivery, explore dose-response relationships between call duration/frequency and patient outcomes, and consider study designs that account for multilevel implementation barriers in real-world hospital settings.

A primary goal of the mentor role was to problem-solve multilevel barriers by connecting participants with community resources to address adverse social determinants. Social determinants (e.g., housing, transportation, economic stability) contribute significantly to health disparities (30,31), and several adverse social factors were present in our study sample (e.g., unemployment, low income, less than a high school education). Furthermore, mentors found community resources were often insufficient, inaccessible, or inadequate to address this population’s needs, which may be partly due to language barriers and ineligibility (11). This might partially explain why mentors did not discuss outside referrals with nearly 40% of participants. These findings highlight the critical need for structural-level interventions addressing fundamental social needs, such as housing, transportation, food security, and other basic necessities, before individuals can effectively prioritize their health. Several programs targeting these issues show positive outcomes (32), including permanent supportive housing for patients with chronic conditions that has been shown to reduce hospital and long-term care use and improved health outcomes (33–35). Other interventions include nonemergency medical transportation, which has increased outpatient and preventative care use (34), and the Supplemental Nutrition Assistance Program, which demonstrated 1% lower odds of hospitalization for every $10 increase in monthly benefit in Medicare/Medicaid-eligible individuals (32,36,37). Implementing and scaling such programs through public policy is challenging but essential for addressing readmission rates among individuals experiencing adverse social determinants.

Strengths and Limitations

The current process evaluation has several important strengths. While prior systematic reviews of care transitions programs document “reach rates” (7,8), research on other process evaluation metrics in this area remains limited. In addition, care transitions studies have not examined process metrics (reach, adoption, and implementation) to elucidate unexpected effectiveness findings, which is an important gap given the implications for understanding the dissemination and sustainability potential of care transitions programs. Moreover, our process evaluation findings add to the literature by demonstrating that high reach rates and adequate implementation fidelity do not guarantee intervention effectiveness when intervention design is insufficient to address the complexity of patient needs and contextual barriers. Our findings illustrate how barriers across multiple levels may have interacted to undermine effectiveness despite strong process metrics. Specifically, our intervention model may have been insufficient for this population with high medical, social, and behavioral needs. Patients may have required more intensive support, such as home visits, home-based monitoring, social workers with greater expertise in directly addressing social adversities, more comprehensive medication reconciliation, and greater continuity (e.g., less health care fragmentation) between inpatient and outpatient care. These findings align with a systematic review of hospital-based intervention, which identified multilevel factors that influence implementation success at the system level (organizational structures, resources, workflows), staff level (capacity, training, buy-in), and intervention level (design, complexity, fit with existing practices) (38). Future work with at-risk populations should consider whether intervention intensity and design match population complexity, including the need for specialized personnel, intensive support modalities, and preimplementation assessment of system-context and intervention-context fit.

It is important to consider several limitations when interpreting the results of this process evaluation. First, fidelity data relied on interventionist-documented checklists rather than objective measures; however, interventionists received ongoing training and support, and fidelity checks were conducted throughout to prevent protocol violations. Second, only 10 of 15 mentors were interviewed, and only one nurse was employed during the study, limiting perspectives from these roles. Additionally, these interviews occurred up to 1 year post-study, potentially affecting recall. Third, the coronavirus disease 2019 pandemic coincided with recruitment’s end, potentially affecting some participants’ experiences and community resource access, although likely affecting relatively few participants. Fourth, while Mi Puente was rigorously informed by evidence-based theoretical frameworks and interventions and adapted using community-based participatory research principles (10), our findings underscore that intervention components effective in controlled settings or implemented with alternative patient populations may function differently among at-risk populations that face greater socioeconomic adversity. Additional limitations of the broader trial are presented in detail in our primary outcomes article (11).

Conclusion

To minimize burden on patients and the health care system, culturally tailored, innovative care transitions programs are needed to reduce hospital readmission rates among vulnerable groups. While Mi Puente effectively reached its target population (reach) among those screened and eligible, was well accepted by interventionists (adoption), and was adequately delivered as intended (implementation), it unexpectedly increased hospital use. The current study incorporated many evidence-based components from prior studies, but our findings underscore how some of these approaches may need to be further adapted to achieve the desired outcomes (11). Future efforts must place strong emphasis on mitigating adverse social circumstances through policy initiatives that address fundamental needs, such as stable housing, reliable transportation, and food security. Without addressing these upstream factors, clinical interventions may inadvertently increase health care use among vulnerable populations who face persistent structural barriers to health.

Supplementary Material

This article contains supplementary material online at https://doi.org/10.2337/figshare.31424426.

ARTICLE HIGHLIGHTS.

• Why did we undertake this study?

  To better understand the unexpected effectiveness findings from the Mi Puente randomized controlled trial using components from the Reach, Effectiveness, Adoption, Implementation, and Maintenance framework.

• What is the specific question we wanted to answer?

  What intervention factors contributed to the unexpected increase in readmission rates among participants in the Mi Puente group relative to usual care?

• What did we find?

  Mi Puente successfully reached the target population; however, gaps related to intervention content delivered and the availability of resources within San Diego affected both adoption and implementation.

• What are the implications of our findings?

  To enhance outcomes post-discharge, it is essential to implement strategies that address multilevel barriers to care, particularly during the transition from inpatient to outpatient settings.