Every year, over $4 billion is spent on hospitalizations for Americans with complications of advanced liver disease: ascites, hepatic encephalopathy (HE), and/or gastrointestinal bleeding.1 Patients with these complications are often frail, medically complex, and experience poor quality of life.2,3 Hospitalization is a major driver of morbidity. The immediate post-discharge period is a vulnerable time for patients, where disease complications can interact with medication errors and poor provider communication to contribute to poor outcomes. More than 25% of discharges result in readmission, some of which are preventable.4–8Outcomes during the post-discharge period could be improved through transitional care (TC) focused on optimizing treatment regimens, improving communication, and facilitating access to community resources.9,10 TC has successfully reduced readmissions (while improving quality of life and patient experience) in multiple complex populations.11–13 However, TC has not been tested for patients with advanced liver disease. We therefore developed a TC program tailored specifically for patients with liver disease—the Transitional Liver Clinic (TLC). The TLC is an advanced practice provider (APP)-led program in which patients discharged from the hospital are contacted by telephone within 2 days, followed by a face-to-face or telehealth APP visit within 14 days. TLC providers manage medications, provide education and linkage to community resources, and facilitate follow-up.14 In our preliminary experience, TLC attendance was associated with a significant reduction in 30-day readmissions (12% vs. 22%, p=0.02).15To test whether the TLC can improve outcomes in this population, we plan to conduct a pragmatic stepped-wedge randomized trial of up to 1000 hospitalized patients with advanced liver disease at 4 academic hospitals to compare the TLC to usual care. Patients are followed from discharge and assessed for outcomes at 30 and 90 days. Our aims are to test the effect of the TLC compared with usual care on readmissions, emergency room visits, mortality, days alive out of the hospital, quality of life, and patient satisfaction. We hypothesize that the TLC will reduce acute healthcare utilization and improve the health and experience of care for patients with advanced liver disease. If the TLC is effective, we plan to expand this work to additional hepatology centers and to assess its impact on other important health and system outcomes.
METHODS
This is a stepped-wedge cluster randomized trial of patients with advanced liver disease enrolling over 45 months at Indiana University, the University of Michigan, Jefferson Einstein Philadelphia, and the University of Chicago. All sites provide usual care (control) during the initial 9-month enrollment interval. At each subsequent 9-month interval, one site crosses over to implement the TLC, which will continue through the final interval, where all sites will have implemented the TLC (Figure 1). The order of crossover is randomized via a computer-generated randomization scheme. All sites enroll for the entire 45 months and thus contribute patients to both the control and TLC groups. Patient follow-up occurs at 30 and 90 days after discharge, assessing hospitalizations, quality of life, and patient satisfaction, and those enrolled during the intervention and control periods are compared. The primary outcome is 30-day readmission. Secondary outcomes include emergency room visits, mortality, days alive out of the hospital, quality of life, and patient satisfaction at 30 and 90 days. We hypothesize that the TLC will provide the necessary resources to reduce readmissions while improving patient quality of life and patient satisfaction. Figure 2 shows the conceptual model. The study is approved by all applicable institutional review boards (IRBs).
Transitional Liver Clinic conceptual model. Abbreviations: ER, emergency room; HRQOL, health-related quality of life.
Hospitalized patients seen by hepatology providers at the enrollment sites are screened for eligibility. Eligibility criteria are listed in Table 1. Most liver disease readmission research has been limited to cirrhosis.4 However, the liver disease complications that lead to readmissions are also common in alcohol-associated hepatitis, and concurrent cirrhosis is often present and hard to identify in this population.6,7 We have therefore included the broader population—advanced liver disease—defined by clinical complications rather than the underlying cause of liver disease. Patients who are listed for liver transplant with Model for End-stage Liver Disease (MELD)-Na ≥35 are excluded because of the high likelihood that they will receive a transplant during short-term follow-up.
Eligibility criteria
Abbreviations: ALT, alanine aminotransferase; AST, aspartate aminotransferase; MELD-Na, Model for End-stage Liver Disease–Sodium; NIAAA, National Institute on Alcohol Abuse and Alcoholism.
Research staff reviews the hepatology census each day, and the medical record is used to screen for eligibility, which is confirmed by the investigators. Those people whose records are reviewed but do not meet eligibility criteria are defined as pre-screen failures. For these individuals, we record the reasons for exclusion (eg, unable to be seen in person or virtually, non-English/Spanish speaking), and these data can be used to explore the limitations of the reach of the TLC. Eligible individuals (those who meet inclusion criteria and do not meet any exclusion criteria) are approached in person for informed consent. Capacity to provide informed consent is assessed by trained research staff under the supervision of study investigators. For patients with HE or other conditions that may impair cognition, additional assessment and investigator input are used to ensure appropriate determination of capacity. Individuals lacking capacity are excluded from participation. Capacity to complete follow-up assessments at 30 and 90 days is also assessed by research staff before data collection at those time points. Participants who provide informed consent but are ultimately unable to proceed with the study (eg, those who are discharged to hospice or who die before discharge) are deemed screen failures.
To ensure adequate retention, we employ follow-up procedures that are flexible and patient-centered. Although the default mode of contact is phone, we also allow alternate modes of contact for data collection based on patient convenience (eg, email, videoconference, text). In addition, the study database (REDCap) has survey functionality in which participants can complete instruments directly using a secure link on their home device. Participant remuneration for follow-up survey completion ($25 each for 30-day and 90-day visits) also enhances retention. The primary outcome, 30-day readmission, is obtained through medical record review and/or direct patient contact.
The TLC is a post-hospital discharge TC management program, meeting criteria established by the Centers for Medicare & Medicaid (CMS) (Table 2).16 At the time of discharge, a face-to-face visit [in-person (preferred) or video telehealth] is scheduled within 14 days of discharge. The specific predefined activities and interventions of the TLC are shown in Table 3. Within 2 business days of discharge, the patient is contacted by phone and is queried as to overall health, new or ongoing problems, medication reconciliation, confirmation of any pending testing or procedures, and confirmation of the TLC visit date and time. Staff also provides necessary education and fields questions. At the clinic visit with the APP, the patient undergoes a full history and physical, with attention to specific complications of advanced liver disease (ascites, HE, and portal hypertension-related bleeding) and their contributing factors. In addition to the direct care provided by the APP (ordering and coordinating tests and procedures, and adjustment of medications), referrals are made for additional services where indicated. These services can include (1) social work for financial assistance and linkage to alcohol and substance use disorder treatment and other community resources, including resources for caregiver support; (2) nutrition services for further dietary counseling; (3) pharmacy for medication review and prescription assistance. The APP also provides further disease education, ensures appropriate follow-up with other healthcare providers, and assists in scheduling follow-up with other community services. Where appropriate, patients will be referred to palliative care or hospice. The TLC is specifically designed to target modifiable drivers of readmission, including early identification and management of ascites and HE, medication reconciliation, and improved care coordination during the high-risk post-discharge period.8 The TLC is a framework for personalized medicine and allows leeway for APPs to make individualized clinical decisions without overly micromanaging all aspects of care (eg, pain management). Before and after the TLC visit, for the full 30-day post-discharge transition period, TLC staff, including the APP, are available to the patient by phone or via a secure electronic medical record-based portal for additional communication. During this time, hepatology care is coordinated through the TLC, while patients may continue to receive care from other healthcare providers without restriction. At the end of this period, patients transition back to their usual hepatology providers for ongoing care. Although the goal is to have all patients scheduled within 14 days, some TLC visits may occur beyond the 14-day period.
Overview of transitional care management requirements
Abbreviations: TC, transitional care; d, days.
TLC activities
Abbreviation: TLC, Transitional Liver Clinic.
APPs participating in the TLC receive standardized orientation to the study protocol, including guidance on core intervention components and clinical management principles. APPs operate within their institutional scope of practice with oversight from site hepatology investigators, while clinical decision-making remains individualized. Regular cross-site meetings are conducted to promote consistency in intervention delivery and address implementation challenges.
In addition to effectiveness outcomes, key implementation metrics will be collected, including reach (proportion of eligible patients enrolled), intervention uptake (attendance at TLC visits), and fidelity to core intervention components. These measures will inform the interpretation of trial findings and future scalability. Intervention fidelity will be assessed based on completion and timing of the 2 core components: (1) telephone contact within 2 business days of discharge and (2) APP visit within 14 days of discharge. Completion rates and timing of each component will be recorded and summarized descriptively. While the target window for the clinic visit is within 14 days, visits occurring within an extended window (eg, up to 30 d) will be considered acceptable, whereas visits beyond this period will be classified as protocol deviations and analyzed accordingly.
Enrolled patients discharged during the usual care period receive post-discharge follow-up care from their usual providers based on recommendations made by inpatient providers at the time of discharge. Usual care, rather than attention control, was chosen for the control because (1) usual care includes follow-up care, so the usual care arm is still an active control that represents current practice in the real world; and (2) the stepped-wedge design reflects cluster randomization, so attention control activities are not needed to incentivize enrollment in the control arm. Usual care includes discharge planning and recommendations for follow-up with hepatology or primary care providers, but does not include a structured transitional care program with standardized timing, proactive outreach, or coordinated multidisciplinary services. While institutions may have general post-discharge follow-up processes, these lack the standardized components of the TLC. The TLC intervention is designed to replace, rather than duplicate, routine post-discharge hepatology follow-up during the intervention period.
Trained research assistants complete assessments according to the schedule of activities (Table 4) with data collection via REDCap.17 MELD and Child–Pugh scores are calculated with standard-of-care laboratory results. The PROMIS-29+2 and PSQ-18 measures are collected via telephone at the 30 and 90-day follow-up time points, though alternative modes of collection are also allowed. AUDIT-C is collected at baseline to characterize alcohol use and is not intended as a longitudinal outcome measure. Healthcare utilization outside of the study, including visits with other providers and hospitalizations, is assessed through patient self-report and medical record review during follow-up.
Schedule of activities
Note: The TLC phone call and visit are only applicable to those in the TLC intervention arm.
The window for the 30-day and 90-day visits is ±14 days.
Abbreviations: AUDIT-C, Alcohol Use Disorders Identification Test–Consumption; ER, emergency room; HIPAA, Health Insurance Portability and Accountability Act; MELD, Model for End-stage Liver Disease; PROMIS-29+2, Patient-Reported Outcomes Measurement Information System 29-Item Profile + 2 Cognitive Function Items; PSQ-18, Patient Satisfaction Questionnaire Short Form (18-item version); TLC, Transitional Liver Clinic; d, days; h, hours.
The primary outcome is 30-day hospital readmission. 30-day readmission is publicly reported by CMS and is the benchmark measure in the Hospital Readmissions Reduction Program, and it is therefore the most appropriate primary outcome.18 Secondary outcomes are preventable 30-day readmission, 90-day readmission, and 30-day and 90-day emergency room (ER) visits, mortality, and days alive out of the hospital. For readmission events, the local site investigator independently adjudicates preventability based on review of admission notes and discharge summaries from both the index hospitalization (the initial hospitalization when patients are enrolled) and the readmission hospitalization. The unblinded determination is anchored by a widely used standard definition for “preventable.”19 For the initial 48 participants with a readmission (12 per site) who have sufficient data to make a determination, preventability is assessed by an investigator from each site independently (4 separate adjudications for each event), and inter-reviewer agreement is examined. Days alive out of the hospital is a novel outcome measure that is patient-centered, accounting for repeated events, event severity, and with appropriate relative weighting of death compared with hospitalization.20 These outcome data are obtained from a combination of self-report and from local institutional medical records, and the mode of data collection is recorded.
Patient quality of life is measured using PROMIS-29+2 v2.1 (PROPr). This measure includes 4 items in each of 7 domains (physical function, anxiety, depression, fatigue, sleep disturbance, ability to participate in social roles and activities, and pain interference), 2 items for cognitive function, and 1 item for pain intensity.21,22 Each domain is scored using a T-score in which 50 corresponds to the average for the US population with a standard deviation of 10, and a higher score represents more of the concept being measured. For example, higher fatigue scores indicate more fatigue (worse quality of life), while higher physical function scores indicate better physical function (better quality of life). The instrument also provides physical and mental health summary scores. PROMIS-29 has excellent internal consistency and reliability. Compared with other measures, PROMIS-29 has improved depth of measurement without floor/ceiling effects, and is completed faster (5.4±3.0 min).23 PROMIS measures have been validated in several samples of patients with advanced liver disease, and have been associated with readmissions in patients with cirrhosis.23–25 Minimally important differences for the domains are 2.5–5.0.26
Patient satisfaction is measured using the short form of the Patient Satisfaction Questionnaire (PSQ-18).27 The PSQ-18 consists of 18 items examining 7 dimensions of satisfaction with medical care: general satisfaction, technical quality, interpersonal manner, communication, financial aspects, time spent with doctor, and accessibility and convenience. Each item is scored on a 1–5 Likert scale, and the individual item scores are averaged, with higher scores reflecting higher satisfaction. Individual subscale scores for the above dimensions are derived by averaging the applicable item scores. The PSQ-18 was derived from the larger Patient Satisfaction Questionnaire, and it has excellent internal consistency and reliability, as well as content, factor, analytic, discriminant, and convergent validity.27 Although not specifically validated in decompensated cirrhosis, it has been employed in numerous populations, including chronic liver disease, and has been used to measure satisfaction with telehealth services.28–30
The distribution of clinical covariates across levels of TLC intervention groups will be assessed using appropriate two-sample tests. Numerical and categorical variables will be assessed using two-sample t tests or Chi-square tests, with consideration given to possible non-normality and small sample sizes via non-parametric and exact tests, respectively. The primary outcome will be analyzed within a generalized linear mixed model (GLMM) framework to account for temporal changes over time and intracluster correlation (ICC) between and within clusters, which are the hallmark analytic factors that need to be accounted for within the stepped-wedge design framework. We will consider the discrete time GLMM for the analysis of stepped-wedge designs as outlined within the seminal work of Hussey and Hughes,31 in which time is a fixed categorical effect, treatment is a time-dependent categorical covariate, and the random effect of cluster is included in the model. Since the Hussey and Hughes model assumes a common secular trend, treatment effect, and ICC between and within subjects, we will also consider the alternative discrete time decay models of Hooper et al32 and Kasza et al,33 which allow for differences in ICC between and within clusters and also allow for the ICC to decay over time between clusters. We will also investigate cluster treatment heterogeneity and the possibility of a time-varying treatment effect via the inclusion of a random cluster × time random effect and treatment × time interaction, respectively, using either discrete time, calendar time, or time on treatment. To address the possibility of a non-linear secular trend, we will consider additional sensitivity analyses using flexible functional forms (eg, restricted cubic splines) and small sample corrections to account for the small number of clusters. In the case where an imbalance of baseline variables is found (eg, cause of liver disease), these variables will be adjusted for in the aforementioned models. Since there are no penalties, with respect to bias, for over-specification of the model by assuming a decay in ICC over time or parametric alternatives for time, the aforementioned models will be fit in order of complexity and collapsing to the Hussey and Hughes model should we encounter issues with convergence or if the assumptions for that seminal model are met.
Secondary outcomes of preventable readmission, 90-day readmission, 30-day and 90-day ER visits, mortality, and days alive out of the hospital will be analyzed as outlined for the primary outcome after modifying the model to match the distributional form of the outcome measures. Quality of life analyses at 30-day and 90-day post-discharge will be performed as outlined in the primary outcome, but reflecting the approximate continuous nature of the PROMIS scores within the GLMM framework for stepped-wedge designs. Patient satisfaction at 30-day and 90-day post-discharge analyses will be performed as outlined in the primary outcome, but reflecting the approximate continuous nature of the PSQ-18 measures within the GLMM framework for stepped-wedge designs. Linear mixed models are used for their ability to handle missing data via maximum likelihood under the missing at random (MAR) data mechanism and thus provide unbiased estimates using complete case analyses. Our main analyses will assume MAR, though we will perform sensitivity analyses using different missing-not-at-random mechanisms.34–36
All analyses will be performed under the intention-to-treat paradigm, though as part of sensitivity analyses we will also perform (1) “as treated” analyses in the event of cross contamination or failure to adhere to the treatment timing and (2) exploratory subgroup analyses to examine co-intervention (home care or primary care), intervention (in-person or telehealth), and liver disease etiology subgroups for possible differential treatment effects. Subgroup analyses will be conducted using interaction terms within the GLMM framework and will be considered exploratory. All analyses will be performed using SAS 9.4.
The power calculations were performed using the seminal work on stepped-wedge designs by Hussey and Hughes31 and implemented in PASS Version 2019. In preliminary data of TLC use and resultant 30-day readmission rates, we observed readmission rates of 22% in those who did not attend the TLC versus 12% in those who did.15 This preliminary data, reflecting a somewhat healthier patient sample with fewer liver disease complications and lower overall readmission rates, postulates an absolute risk reduction of 10%. In papers by Morando and Morales, 19%–27% absolute risk reductions were observed with control group rates between 30% and 40%.37,38 To be as flexible as possible with respect to hypothesized treatment effects, we varied our control proportion of 30-day readmission rates from 0.27 to 0.29 and treatment proportions from 0.10 to 0.12, which results in absolute reductions in hospital readmissions in the range of 15%–20%. These proposed effect sizes fall within the range of both the preliminary data and the previous studies, which showed higher effect sizes, and also reflect a reasonable range of clinically relevant effects. Smaller effect sizes (eg, 5%–8% absolute reductions) are possible in a pragmatic trial; under such scenarios, the trial may be underpowered to detect statistically significant differences but would still provide clinically informative estimates. Since no estimates of between-cluster variability were readily available, we incorporated a large range of possible ICC values (0.01–0.10) that were thought to be clinically reasonable and conservative. Varying the design parameters as outlined above, fixing power at 0.80 and significance level at 0.05, we can expect our effective sample sizes to range from 420 to 900, accounting for 10% attrition (Table 5). Therefore, enrolling 1000 patients with an assumed 10% attrition rate will cover all reasonable design scenarios with average cluster sizes ranging from 105 to 225 and average cluster size per time period ranging from 21 to 45.
Sample size estimates based on projected 30-day readmissions in the TLC versus Usual Care arms
Note: Cell ranges account for ICC estimates ranging from 0.01 to 0.10.
PROMIS measures in patients with chronic liver disease are worse than population norms: physical function 42.9±8.6, anxiety 52.1±9.8, depression 51.2±9.2, fatigue 55.6±11.1, sleep disturbance 53.8±9.1, social roles 48.6±9.5, and pain interference 55.4±10.2.23 In a sample of persons with liver disease, physical and mental health summary scores were 43.0±9.1 and 45.7±9.7, respectively. With effective sample sizes of 800–1000, assuming a worst-case scenario of 10% attrition, we can detect changes in the range of 2.3–2.7 points in the physical and mental health summaries, which is in line with hypothesized clinically meaningful changes of 3–4 points.26
PSQ-18 scores in a previously published hepatology population were 4.0±0.7.30 With effective sample sizes of 800–1000, assuming a worst-case scenario of 10% attrition, we can detect changes in the range of 1–2 points.
The primary outcome and several secondary outcomes are ascertained through medical record review and are therefore not subject to traditional loss to follow-up. Attrition primarily affects patient-reported outcomes, for which we have implemented flexible follow-up procedures. All power scenarios were run with PASS 2019.
Safety and study progress reviews are conducted. The study coordinators and biostatistician generate reports for the PI, safety officer, and Data and Safety Monitoring Board (DSMB) that contain: summary of adverse events and an explanation of how each event was handled; summary of complaints and how each complaint was handled; subject retention, including the number and reasons of participant withdrawals; intervention compliance (attendance); and summary of protocol violations and how each was handled. All reports are submitted to the IRB as per reporting guidelines. The data and safety monitoring plan is monitored by the PI and a three-member DSMB. The safety officer for this trial is a hepatology physician researcher experienced in health services intervention studies with an understanding of the types and severity of complications that can result when providing care to patients with complications of liver disease. The safety officer reviews reports sent by the study manager and uses a checklist to determine whether there is any corrective action, trigger of an ad hoc review, or stopping rule violation that should be communicated to the investigators, DSMB, IRB, and the sponsor. In addition, the safety officer may comment on whether the study investigators need to report any specific out-of-range data to participants and/or their physicians. The safety officer is joined by an expert health services researcher and a biostatistician to constitute the three-member DSMB panel. These persons are not otherwise affiliated with the project. The three-member DSMB panel meets at 12-month intervals.
Deidentified individual participant data and statistical code will be made available in accordance with NIH data sharing policies. Data will be shared through appropriate repositories or via data use agreements, subject to institutional approvals and protection of participant confidentiality. Requests for data access will be reviewed in accordance with institutional requirements, data use agreements, and participant confidentiality protections.
Trials of complex interventions often face challenges in recruiting and retaining participants. Our team has developed an infrastructure to recruit hospitalized patients with liver disease. Regular reports are generated by the research team, and regular meetings occur to ensure adequate recruitment and retention, and to troubleshoot areas of need. Telehealth allows us to provide the TLC and to maintain enrollment for those who live far from the sites. Although CMS authorization for telehealth remains temporary, it has been extended multiple times and continues to evolve within the healthcare system.39 The sites can also offer the TLC on a flexible schedule to accommodate patient schedules. Our recruitment goal is also highly feasible: up to 1000 patients (250/site) over 45 months of planned recruitment requires only <6 patients/site/month. We are able to minimize early withdrawals through quality control procedures, monitoring patient perceptions about the risks and benefits of participation. Follow-up procedures are also flexible and patient-centered. Although the default mode of contact is phone, we also allow alternate modes of contact for data collection based on patient convenience (eg, e-mail, videoconference, text). In addition, REDCap has survey functionality in which participants can complete instruments directly using a secure link on their home device. We monitor retention and troubleshoot study procedures if retention is inadequate. We have instituted alert points for dropout rates with early action plans to correct inadequate retention. For those participants who do not respond to follow-up attempts, the primary outcome, 30-day readmission, and multiple secondary outcomes (90-day readmission, ER visits, mortality, and days alive out of the hospital) can still be obtained from the electronic medical record. Due to the nature of the intervention, it is not possible for participants and investigators to be blinded, and the lack of blinding, including unblinded preventability adjudications, could contribute to bias. Bias is minimized by using objective outcomes, including patient-reported outcomes, which are pre-scripted.40 In addition, exclusion of patients with very high MELD-Na scores may limit generalizability to the sickest patients with advanced liver disease.
It is possible that the trial will not result in a statistically significant improvement in outcomes with the TLC. In this case, the study will still provide valuable information that can be used in practice and in future research. The collection of the proposed rich dataset will allow us to examine other baseline variables and the impact of those variables on the outcomes. In addition, subgroup analyses may help identify sub-populations more likely to benefit from the intervention. Integration of outcomes will also provide insight into the relationship between patient-reported outcomes and acute healthcare utilization. Even if the TLC cannot reduce unavoidable readmissions, it may still facilitate earlier recognition of triggers for hospitalization and thus promote earlier intervention and shorter readmissions, which may be reflected in more days alive out of the hospital. The proposed intervention is complex, with numerous components that are integral to its functioning and are thus inseparable (Figure 2). We will not be able to isolate and assess which of the components is most important overall. However, caring for patients with advanced liver disease is already a multidisciplinary endeavor, and we believe that single components cannot be isolated as more important than others. In addition, variations in APP clinical practice patterns across institutions may contribute to heterogeneity in the intervention delivery, though it may also enhance generalizability.
CONCLUSION
The TLC model is a highly promising approach to reducing the continued high burden of hospital readmissions for those with advanced liver disease. It offers the benefits of multidisciplinary liver-centric personalized care led by experienced hepatology APPs while leveraging TC incentives established by CMS and recent innovations in telehealth care delivery. The results from this multicenter stepped-wedge randomized trial will provide the first evidence that a TC program can improve acute healthcare utilization, quality of life, and the care experience for patients with advanced liver disease. The next steps will be to adapt and expand this intervention to other settings and for other high-risk hepatology populations. We also plan to use the TLC as a template to develop interventions targeting vulnerable patients lacking access to traditional TC (eg, those without internet access who are also unable to be seen in person). Additional work will also be necessary to explore the impact of the TLC on other patient-centered, provider-centered, and caregiver-centered outcomes.
ET
Dr. Elliot Tapper
Gastroenterology · Hepatology & Liver TransplantationUniversity of Michiganelliot
