HOMERuN
Research Collaboration Proposals

PI NAME + AFFILIATION: Stephanie K. Mueller, MD MPH, Associate Physician, Brigham and Women’s Hospital (BWH), Instructor of Medicine, Harvard Medical School

POTENTIAL CO-INVESTIGATORS:

• Jeffrey Schnipper MD MPH, BWH
• (Potential for collaboration with site-specific co-investigators): Shani Herzig, Greg Ruhnke, Josh Metlay, Jennifer Myers, Ryan Greysen, Sunil Kripalani, Ed Vasilevskis, Neil Sehgal, Grant Fletcher, Kevin O’Leary,  Edmondo Robinson, Scott Flanders, Peter Lindenauer, Sumant Ranji, Mark Williams, Andy Auerbach, David Meltzer

PROGRAM OVERVIEW/INTRODUCTION: Though commonly undertaken to improve care for patients, transfer from one acute care hospital to another (inter-hospital transfer or IHT) can be costly and expose patients to many potential risks, especially given the severity of illness in this patient population and the discontinuity of care that transfers generate. Despite these risks, substantial gaps remain in knowledge about the outcomes for these patients across a variety of diagnoses, as well as various transfer process factors and patient characteristics that may influence these outcomes and that therefore might be targets for safer transfer practices.

KEY CLINICAL QUESTIONS/EVIDENCE GAPS:

• What are patient, hospital, and transfer process predictors of outcomes among transferred patients?

AIMS + HYPOTHESES:

                Specific Aim: To determine the association between transfer process factors, patient and hospital characteristics, and patient outcomes among patients who undergo IHT, using multi-site data.

                Hypothesis: Certain transfer process factors and patient and hospital characteristics are associated with worse patient outcomes among transferred patients.

PRELIMINARY DATA: We are currently in the process of conducting this same proposed study within a single site (BWH), as part of an AHRQ K08 career development award.  We thus hope to utilize the same methodology, expanding this project to multiple sites, to improve statistical power and the generalizability of our findings. At this time, we have gathered all of the relevant data available from computerized data sources on approximately 26,000 patients transferred to BWH from another acute care hospital between January, 2005 and September, 2013. We are developing a data abstraction tool to conduct concurrent medical record review to gather additional variables, and we will complete our analyses evaluating the association of all variables with clinical outcomes in the coming months.

STUDY DESIGN (including study subjects and setting, comparator groups, data sources, outcomes, analysis plan, power and sample size, limitations, and timeline):

1. Study populations: We plan to include patients at least 18 years old who were transferred to participating site hospitals from another acute care facility between January, 2012 and December, 2016 (pending availability of data available encompassing these years). We aim to exclude patients that left against medical advice and patients that were transferred from any closely affiliated acute care hospital (i.e., those that share medical personnel and medical records systems).

2. Variables (Exposures, Outcomes) + Data Sources:

From Computerized Data Sources (PCORnet + New Requests for extra fields): We aim to gather the majority of our variables (exposure and outcome) via computerized data sources, utilizing fields that PCORnet sites are contributing data to, in addition to some extra fields that we will request for the purposes of this study. These variables can be readily adjusted based on data available at the time of the study (i.e., re-defining the variables, removing/adding variables based on data availability). In addition to listing each variable, I have incorporated which PCORnet table the variable can be found (or can be extrapolated from), or if it is a new variable request:

• Patient demographics (Age, sex, race, insurance) [PCORnet Demographics Table]
• Service of admission (i.e., surgery, medicine) [PCORnet Encounter Table or New Request]
• Level of admission upon arrival (ICU vs. not) [PCORnet Encounter Table or New Request]
• Admission diagnosis (ICD-9) [PCORnet Diagnosis Table]
• Primary diagnosis (ICD-9) [PCORnet Diagnosis Table]
• Comorbidity (Elixhauser) [PCORnet Diagnosis Table]
• Severity of Illness (MS DRG-weight based on year admitted) [PCORnet Encounter Table]
• Number of medications at time of transfer [PCORnet Medications Prescribing Table or New Request]
• Number of prior hospitalizations in prior 12 months [New Request]
• Transferred from a hospital that is in-network/affiliated with receiving hospital (Y/N) [New Request]
• Prior relationship to receiving hospital (defined as PCP and/or any clinic visit in 12 months prior to date of transfer) [New Request]
• Distance between receiving hospital and patient’s home (based on zip code of home and hospital address) [PCORnet Demographics Table]
• Distance between receiving hospital and transferring hospital [New Request (extrapolated from transferring hospital name/address]
• Select laboratory values at time of arrival: hemoglobin, sodium, potassium, creatinine, bicarbonate [PCORnet Lab Results Table]
• Day of week of transfer [PCORnet Encounter Table]
• Time of day of arrival [PCORnet Encounter Table]
• Transfer on a pre-holiday/holiday date (Y/N): Dec 24-25, Dec 31, July 3 [PCORnet Encounter Table]
• Time delay between transfer acceptance and arrival of patient [New Request]
• Hospital unit “busyness” (defined as number of admission and discharges to/from the accepting unit on day of patient arrival) [New Request]
• Hospital team “busyness” (defined as the number of admissions and discharges to/from the accepting team on day of patient arrival) [New Request]
• Characteristics of transferring hospital (if able to merge with AHA files based on name of transferring hospital) [New Request-transferring hospital name]

Outcome variables will also be obtained from computerized data sources and will include:

• Cost of hospitalization (at receiving hospital) [New Request]
• Length of Stay (at receiving hospital) [PCORnet Encounter Table]
• Observed-to-Expected Length of Stay (at receiving hospital) [New Request]
• Transfer to a different service (i.e., medicine to surgery) within 48 hours of admission to receiving hospital [PCORnet Encounter Table or New Request]
• Transfer to a higher level of care (i.e., floor to ICU) within 48 hours of admission to receiving hospital [PCORnet Encounter Table or New Request]
• Mortality: 30-day mortality, death within 3 days of transfer, and in-hospital mortality [PCORnet Death Table]
• Observed-to-Expected Mortality [New Request]
• 30-day Readmission [New Request]

From Manual Medical Record Review: In addition to gathering data from electronic sources (above), we aim to engage willing sites to perform manual chart review on randomly selected patients (100 patients/site, evenly distributed over the time course of the study), to gather variables not readily available from electronic sources. We can provide willing sites previously created data-abstraction tools and definitions to gather these data:

• Reason for transfer
• Appropriateness of accepting service
• Information exchange at time of transfer:
  • Availability/Quality of transfer summary
  • Presence of imaging data

3. Analysis Plan:  All analyses will be stratified by primary diagnosis at time of admission to receiving hospital. We will first assess unadjusted associations between each exposure and outcome, using Fisher exact test for discrete outcomes (mortality), and t-test or Wilcoxon rank sum test for continuous outcomes (cost, length of stay), followed by adjusted analyses with addition of all exposures as covariates using multivariate regression analysis (for cost, linear multivariable regression models (with log-transformation of the outcome, if necessary); for length of stay, Poisson regression (or survival analysis); and for mortality outcomes, logistic multivariable regression). Assuming there will be some frequency of missing variables by site, we will address this with use of dummy variables when missing data accounts for > 10% of the n for categorical variables, and multiple imputation when able for other variables. Additionally, time (in years) will be added to the multivariable regression models as a fixed categorical effect to account for time trends. We will also cluster by hospital.

4. Power and Sample Size: Based on limited data that 30-day mortality of transferred patients ranges from 4-11%, and a conservative 15% exposure for any covariate, we will need a sample size of 59,000 patients to detect a 1% difference in 30-day mortality, with 80% power, and an alpha of 0.05 for any particular variable. We know that there are approximately 7,000 patients transferred to BWH annually. Assuming we include up to 10 sites with similar transfer volume, over the 3 years of data we are asking sites to provide, our sample size will be 210,000, thus ample power for this study, allowing for inclusion of sites with lower transfer volume than expected. In analyses of exposures obtained from medical record review, if 7 sites agree to conduct targeted chart review, with 100 patients/per site, that will give us a sample size of approximately 700 patients. With 700 patients, we will be able to detect 0.7 day difference in length of stay with 84% power (assuming a 5-day average length of stay, following a Poisson distribution, and a 5% type 1 error).

5. Limitations: We understand there may be site-specific limitations to the data we are requesting, and that some of the variable data will be incomplete from each site. We plan to account for this in our statistical analyses, utilizing dummy variables and/or imputations for missing data where appropriate.  We understand that there may be other factors that influence the safety of IHT that cannot be captured quantitatively. Lastly, our analysis of factors only available by medical record review may be underpowered for certain rare exposures and for rare outcomes like mortality. 

6. Timeline: With above data collection, additional data requests for each site, we anticipate the timeline of this project to encompass 6 months for IRB approval and finalization of data collection tools, 9 months for data collection and cleaning, and 6 months for analysis and reporting, including manuscript preparation, presentations at meetings, and other dissemination activities.

CHARACTERISTICS OF PARTICIPATING SITES: As we are aiming to examine a cohort of patients who have undergone inter-hospital transfer, ideal participating sites will be those that accept a moderate to large number of IHT patients annually (i.e., tertiary referral centers).  Each current HOMERuN site meets these criteria.

POTENTIAL FUNDERS: To Be Determined, likely will apply for R01 support from AHRQ, or PCORI support for this project within the next 12-18 months. Other potential funders include foundations such as Commonwealth Fund, Moore Foundation, etc.  

1. PI Name And Affiliations

Oanh Kieu Nguyen, MD, MAS
Assistant Professor of Internal Medicine
UT Southwestern Medical Center, Dallas, Texas

2. Potential Co-Investigators

Anil N. Makam, MD, MAS, Assistant Professor of Internal Medicine, UT Southwestern (Co-PI)
Ethan A. Halm, MD, MPH, Professor of Internal Medicine, UT Southwestern
Potential Co-Investigators from collaborating HOMERuN and/or PCORnet CRG partner institutions TBD

3. Program Overview/Introduction

Vital sign instability on discharge could be a simple and clinically objective means of assessing patient readiness and safety for discharge. Our research group has previously shown that the number of vital sign instabilities on discharge is associated with higher rates of 30-day readmission and death in 6 diverse north Texas hospitals. However, several key gaps remain, including 1) unknown national prevalence of vital sign instability among hospitalized patients; 2) reasons why patients are discharged despite the presence of vital sign instability on discharge; and 3) the effectiveness of interventions to reduce vital sign instability on discharge and/or post-discharge adverse outcomes. We propose to: assess the epidemiology of vital sign instability among hospitalized adults at discharge in a more generalizable cohort of hospitals and regions; assess why patients with vital sign instabilities are discharged; and develop and pilot a multifaceted intervention to optimize post-discharge adverse outcomes among patients identified as having vital sign instability on discharge. Our findings will be foundational to developing evidence-based discharge criteria and interventions to optimize patient outcomes, safety, and recovery during the transition from hospital to home and after hospital discharge for patients known to be at higher risk for adverse outcomes.

Our proposed study is aligned with the overall HOMERuN mission to ensure that hospitalized patients receive ‘the best quality, safest, and highest value care from hospitalization through recovery’ and with the specific HOMERuN research priorities of 1) focusing on general medicine patients cared for by hospitalists; 2) increasing collaboration across hospital medicine groups and researchers; 3) optimizing the effectiveness and safety of care transitions; and 4) improving post-discharge patient safety.

4. Key Clinical Questions (PCORI Decisional Dilemma And Gap Analysis)

a) What is the epidemiology of vital sign instability among hospitalized adults at discharge?
b) Why are patients with vital sign instabilities discharged instead of remaining in the hospital?
c) Does a multifaceted intervention improve vital sign stability at discharge and/or post-discharge adverse outcomes?

5. Aims And Hypotheses

Our overall aim is to assess the epidemiology of and reasons why hospitalized adults with vital sign instability are discharged in order to develop and evaluate an intervention to improve management of patients with vital sign instability and reduce subsequent adverse post-discharge outcomes.

SPECIFIC AIM 1: Describe the prevalence of vital sign instability on discharge among hospitalized adults, key clinical characteristics of this population, disposition status, and association of vital sign instability with adverse post-discharge outcomes, including hospital readmission and mortality at 30, 90, and 180 days.

Hypothesis 1: We hypothesize that 1 in 5 adults will have at least 1 vital sign instability on discharge, and that the number (and combination) of vital sign instabilities will be strongly associated with worse post-discharge outcomes even after adjusting for other prognostic factors and post-discharge location.

SPECIFIC AIM 2: Assess reasons for discharging patients with vital sign instability, using a mixed methods approach with both structured chart reviews and interviews of hospitalists.

Hypothesis 2: We hypothesize that there will be three main themes for discharging patients with vital sign instability: 1) known advanced acute or chronic illness not amenable to further in-hospital treatment; 2) known vital sign instability with plan for stabilization at post-acute care facility; and 3) vital sign instability present but not recognized by clinical care team. Furthermore, we anticipate suboptimal rates of goals of care discussion and end-of-life planning prior to hospital discharge.

SPECIFIC AIM 3: Develop and pilot a multifaceted and tailored intervention to reduce vital sign instability and adverse post-discharge outcomes among hospitalized adults.

Hypothesis 3: The intervention will be informed by Aims 1 and 2. We hypothesize that a multi-component intervention may consist of at least: 1) automated best practice alert if vital sign instabilities are present; 2) recommendation to delay discharge until vital signs are stable for at least 24 hours; and/or 3) recommendation for goals of care discussion, palliative care, and/or hospice referral as appropriate if further in-hospital treatment is not warranted.

6. Preliminary Data

Our group conducted a previous AHRQ R24-funded study assessing the association between vital sign stability at hospital discharge and 30-day readmissions and mortality using electronic health record data from a diverse cohort of 6 hospitals in north Texas. (1) In our cohort of 32,835 adults hospitalized on all medicine services, 19% were discharged with at least one vital sign instability and 3% had two or more vital sign instabilities. Overall, 13% of individuals with no instabilities died or were readmitted compared to 17% with 1, 21% with 2, and 26% with 3 or more instabilities. The presence of any vital sign instability was independently associated with higher risk-adjusted odds of the composite outcome of death or readmission within 30 days (AOR 1.36, 95% CI 1.26-1.48), and was more strongly associated with death (AOR 2.31, 95% CI 1.91-2.79) than readmission. Although a smaller proportion of the population had two or more instabilities, the adjusted odds of death more than tripled among this high-risk group compared to those with no instabilities (AOR 3.3, 95% 2.5-4.9).

We also found that the greater the number of vital sign instabilities an individual had on discharge, the more likely they were to be discharged to a post-acute care facility (i.e., nursing homes, skilled nursing facilities, and long-term acute care hospitals). Only 18% of individuals with no instabilities were discharged to post-acute care, compared to 22% of those with one instability, 26.7% of those with two instabilities, and 43% of those with three or more instabilities (p<0.001 for trend). Additionally, individuals with vital sign instabilities discharged to post-acute care facilities had far higher rates of post-discharge adverse events compared those who were sent home. Among individuals with two instabilities, 12% of individuals discharged to a post-acute care facility died within 30 days compared to only 2% of those discharged home (p<0.001 for comparison).

Our study was the first to assess the association between vital sign instabilities and discharge and 30-day mortality and readmission among all adults on medical inpatient services. Although our findings confirm and extend the landmark studies by RAND (2) and Halm et al (3), which assessed the effect of more broadly defined clinical instabilities on discharge in groups with selected conditions in the 1980s and 1990s respectively, several key limitations and knowledge gaps remain. First, the generalizability of our findings beyond the Dallas-Fort Worth metroplex is unknown, especially given the very high post-acute care utilization in this region. Although we included a large sample of diverse patients treated in a variety of settings (one safety net teaching hospital and 5 community non-teaching hospitals), extending our study to HOMERuN CRG institutions (which include a mix of academic/university and community teaching and non-teaching settings) would significantly enhance the external validity or our findings. Second, we were unable to ascertain the reasons and modifiability of vital sign instability on discharge – i.e., patients discharged with unstable vital signs may have had advanced or terminal illness that was otherwise not treatable. Furthermore, among patients known to have advanced or terminal illness, we do not know whether these patients had a goals of care discussion prior to hospital discharge. We were unable to ascertain this information given the limitations of our dataset. Finally, it is unknown whether an intervention strategy targeting individuals with vital sign instabilities would result in stabilization and/or fewer post-discharge adverse events among hospitalized adults. This information is critical to informing the development of objective, evidence-based hospital discharge criteria to optimize post-discharge patient safety.

Use of the HOMERuN CRG data and resources will allow us to validate and extend our prior findings, and develop and pilot an intervention in a more nationally representative cohort of hospitalized adults. Findings from this proposal will form the basis for follow-on multi-site pragmatic cluster randomized controlled trial of our newly developed intervention strategy to reduce vital sign instability on discharge and adverse post-discharge outcomes.

7. Study Design, Including Study Subjects (Patients And/Or Providers) And Setting, Comparator Groups, Data Sources, Outcomes, Analysis Plan, Power And Sample Size, Limitations, And Timeline

SPECIFIC AIM 1:
Study Design, Subjects and Setting: Observational descriptive cohort study of hospitalized adults ≥18 years old cared for on a medicine service

Data Sources: PCORnet Clinical Data Research Network augmented with electronic health record data (EHR) from participating sites, and EHR from 31 hospitals in the Southwestern Health Resources (SWHR) an integrated regional health care network and accountable care organization (ACO) in north Texas that spans a 16-county service area with more than 6 million residents. All SWHR hospitals use the Epic EHR system.

Outcomes: Demographics of patients; frequency and timing of vital sign instabilities on the day of discharge; disposition status (home versus post-acute care), readmissions to the index hospital at 30, 60, and 90 days; mortality at 30, 90, and 180 days

Analytic Plan: Descriptive analyses for prevalence/frequency of vital sign instabilities. To assess the association between vital sign instability and adverse post-discharge outcomes, we will use logistic regression adjusting for sociodemographic characteristics, prior utilization history, principal diagnosis, measures of severity of illness (i.e., selected laboratory abnormalities) comorbidities, hospital complications. We will account for clustering by hospital using generalized estimating equations.

Power and Sample Size: To be determined, depending on number of participating sites.

Anticipated Challenges: We anticipate challenges related to limited availability of data needed to complete this Aim, including: 1) Inconsistent availability of inpatient data on vital signs; 2) incomplete ascertainment of 30-day readmissions, as information will be limited to readmissions to the index hospitalization; and 3) incomplete ascertainment of mortality up to 180 days after hospitalization and in PCORNet. These limitations would be overcome with expanded PCORnet data collection.

One limitation to conducting this study within HOMERuN and/or PCORnet sites is uncertain generalizability to other hospital populations given the overrepresentation of tertiary referral hospitals. The inclusion of community SWHR hospitals in our study cohort will help mitigate this limitation.

Proposed Expansion of PCORnet Data Collection: Expand vital sign collection to extend to all recorded vital sign measurements from inpatient settings on the first and last days of hospitalization; expand ascertainment of mortality to include deaths occurring up to 180 after hospitalization; expand ascertainment of readmissions to include readmissions at any area hospitals (i.e., within catchment area of index hospital)

Timeline: 1-2 years, depending on data availability (years 1-2 of R01)

SPECIFIC AIM 2:
Study Design, Subjects and Setting: Mixed methods study with structured chart review and qualitative interviews of hospitalists

Data Sources: EHR and hospitalist physicians from participating sites

Analytic Plan: We will perform a structured chart review and conduct recall interviews with the treating hospitalists of patients identified with unstable vital signs to elicit reasons for discharge. Among patients identified as having known severe illness, we will assess whether patients had a goals of care discussion, DNR order, and/or palliative care consult. We will conduct descriptive analyses of structured chart review and thematic analysis of interviews and discharge summaries.

Power and Sample Size: 1,000 patients across PCORnet/HOMERuN CRG sites, and SWHR hospitals, or until reaching thematic saturation

Anticipated Challenges: Logistical challenges including 1) reciprocity of IRBs; 2) obtaining data for chart review across sites; 3) conducting interviews across sites; and 4) maintaining fidelity of data abstraction and collection across sites. To overcome these challenges, we would leverage the established expertise of HOMERuN mentors and collaborators in performing multi-site mixed methods studies consisting of chart review and qualitative interviews, given the parallels in our proposed research strategy to methods employed in a previous HOMERuN study assessing the preventability of hospital readmissions. (4)

Additionally, limiting this study to only HOMERuN and/or PCORnet sites may result in uncertain generalizability to other hospital populations given the overrepresentation of tertiary referral hospitals. The inclusion of community SWHR hospitals in our study cohort will help mitigate this limitation.

Timeline: 2 years (years 2-3 of R01)

SPECIFIC AIM 3:
Study Design, Subjects and Setting: Informed by Aims 1 and 2, we will conduct a pilot study of a multifaceted intervention at UT Southwestern Medical Center and one THR community hospital in the DFW metroplex. We hypothesize that this intervention will include 1) automated best practice alerts to notify physicians of the 2 or more presence of vital signs at the time of discharge; 2) automated clinical decision support to recommend tailored interventions including but not limited to a) extending hospitalization; b) structured communication with post-acute care facilities; and/or c) referral for palliative and/or hospice care.

Outcomes and Analytic Plan: We will assess feasibility and acceptability of our intervention using validated constructs as per the RE-AIM framework (Reach, Effectiveness, Adoption, Implementation, Maintenance). These are primarily descriptive analyses to assess the reach, feasibility, and adoption of the intervention. We will conduct a preliminary analysis of the effect of readmissions and deaths at 30, 90, and 180 days with an interrupted time series (ITS) and difference-in-difference (DiD) designs. The ITS design will use the two participating hospitals’ historical data as a control. The DiD analysis will include two other THR hospitals in the DFW metroplex as a reference.

Power and Sample Size: N/A; feasibility study which will inform sample size for multi-site pragmatic RCT

Anticipated Challenges: We anticipate two key challenges: 1) EHR interoperability between UT Southwestern and THR hospitals. All hospitals in the SWHR ACO use the Epic EHR system. By years 3-5 of this R01, we anticipate that SWHR will have a unified data warehouse/repository to allow extraction of EHR data to facilitate our data collection strategy. 2) Ascertainment of hospital readmissions. We will mitigate this challenge by using data on readmissions from a regional all-payer administrative claims database through the Dallas-Fort Worth Hospital Council, which includes data on hospitalizations from 136 hospitals within a 100-mile radius of Dallas, accounting for over 95% of the catchment area.

Timeline: 2 years (years 4-5 of R01)

8. Characteristics Of Sites Who Might Participate

We preferentially seek participation from hospital sites with the following characteristics:

1. Sites with a local champion for hospital medicine research;
2. Sites with experience extracting and using electronic health record (EHR) data for hospital medicine research;
3. Member of either the HOMERuN and/or PCORNet CRGs;
4. (Optional) Sites who have access to regional all-payer claims data to enable ascertainment of utilization and outcomes outside of the site of index hospitalization;
5. Sites that maximize diversity and representativeness of included populations to the general population of hospitalized adults in the U.S.

9. Potential Funders With RFA/RFP And Due Dates (If Known)

We plan to seek R01 funding for this work through the following potential funders:

• AHRQ R01: PA-14-291, AHRQ Health Services Research Projects. This project is aligned with the AHRQ priority areas of patient safety and use of health information technology to improve patient outcomes
• NIA R01: PA-16-160, Investigator-Initiated NIH Research Project Grant. This project is aligned with the NIA priority areas to support research on health services for older adults with multiple chronic conditions, and with the strategic priority to support demonstration/pilot projects for pragmatic clinical trials
• NHLBI R01: PA-16-160, Investigator-Initiated NIH Research Project Grant. This project is aligned with the NHLBI research priority to optimize clinical and implementation research to improve health and reduce disease through leveraging EHRs

Potential non-R01 funders include the following:

• PCORI: Cycle 3, Improving Healthcare Systems
• https://www.pcori.org/funding-opportunities/announcement/improving-healt...) – LOI deadline Oct 31, 2017
• AHRQ R18: PA-17-261, Developing New Clinical Decision Support to Disseminate and Implement Evidence-Based Research Findings
• Moore Foundation: This work is aligned with the Foundation’s priority areas of improving patient safety and improving the experience and outcomes of patients with serious illness
• The Commonwealth Fund: This work is aligned with the Commonwealth Fund’s priority area of health care delivery system reform for individuals with multiple serious chronic conditions

REFERENCES

1. Nguyen OK, Makam AN, Clark C, et al. Vital Signs Are Still Vital: Instability on Discharge and the Risk of Post-Discharge Adverse Outcomes. Journal of general internal medicine. Jan 2017;32(1):42-48.
2. Kosecoff J, Kahn KL, Rogers WH, et al. Prospective payment system and impairment at discharge. The 'quicker-and-sicker' story revisited. JAMA : the journal of the American Medical Association. Oct 17 1990;264(15):1980-1983.
3. Halm EA, Fine MJ, Marrie TJ, et al. Time to clinical stability in patients hospitalized with community-acquired pneumonia: implications for practice guidelines. JAMA : the journal of the American Medical Association. May 13 1998;279(18):1452-1457.
4. Auerbach AD, Kripalani S, Vasilevskis EE, et al. Preventability and Causes of Readmissions in a National Cohort of General Medicine Patients. JAMA Intern Med. Apr 2016;176(4):484-493.

PI name and affiliations: Marisha A. Burden, MD, Denver Health and Hospital Authority, University of Colorado School of Medicine; Angela Keniston, MSPH, Denver Health and Hospital Authority

Site PI name and affiliations: Flora Kisuule, MD, MPH, Johns Hopkins School of Medicine; David Paje, MD, MPH, University of Michigan; Keri Holmes-Maybank, MD, Medical University of South Carolina; Hemali Patel, MD, University of Colorado Hospital, University of Colorado School of Medicine; Jeremy Schwartz, MD, Yale School of Medicine

Potential co-investigators: Jason Stein, MD, 1Unit; Katarzyna Mastalerz, MD, Presbyterian St. Luke's Hospital, University of Colorado School of Medicine; Jon Manheim, MD, Presbyterian St. Luke's Hospital, University of Colorado School of Medicine; Ed Havranek, MD, Denver Health and Hospital Authority, University of Colorado School of Medicine; John Rice, PhD, ACCORDS, University of Colorado School of Medicine

Patient partner: Michelle Archuleta

Program overview/introduction:

The purpose of this proposal is to evaluate a continuum of existing inpatient hospital care models ranging from the traditional care model to the accountable care unit model (ACU) by conducting a pragmatic observational, comparative effectiveness study.

Key clinical questions or evidence gaps:

Fragmented hospital care is common and leads to medical errors, increased utilization of health care and affects many stakeholders including patients, front line staff, and hospital leadership and administration. Currently, the optimal model of care on inpatient medical units is not known. One proposed model of care is the accountable care unit model (ACU) which consists of geographically based teams, structured interdisciplinary bedside rounding, nurse physician leadership dyad, and unit level data (1). Single center observational studies of components of this framework have suggested improved patient-centered outcomes such as patient and family experience of care (2), team communication and collaboration (3), and clinical outcomes (4).  However, there have been no head to head multisite comparisons published on accountable care unit models compared to traditional care.

The proposed study is designed to test the extent to which and for whom the ACU model will be more effective at engaging patients, improving communication across care teams and with patients, and improving clinical outcomes compared to traditional care model. Clinical stakeholders, administrative leaders and healthcare systems need to know which model (and which parts of the models) improves patient centered outcomes and the outcomes of interest to the key stakeholders. We propose a multi-hospital, pragmatic, observational, comparative effectiveness study in which we first assess the fidelity to the ACU care model on a particular unit or by a particular team, followed by observational data collection for the comparative effectiveness study of traditional care models versus accountable care unit models (high versus low fidelity to the components of the ACU).

Aims and Hypotheses:

Specific Aim 1. To assess fidelity to the accountable care unit model with differing levels of ACU model implementation to further refine protocol and prepare for the comparative effectiveness trial.

Aim 1a. Engage patients, families, and caregivers, clinical staff (nursing, providers, other clinical staff), health system administrators and clinical leaders to refine characterization of low and high fidelity ACU units and plan data utilization, collection procedures, and reporting.

Aim 1b. Assess the fidelity of participating sites to the ACU model by measuring through direct observation and at regular intervals, the specific structure and processes and elements present in both the ACU and traditional care model units/teams.

Specific Aim 2. To conduct a pragmatic, observational, comparative effectiveness study on a continuum of existing care models from the traditional care model (with 2 or less of ACU criterion) to the accountable care unit model (with all four criteria)

Aim 2a. Compare patient-centered outcomes such as shared decision-making and patient-reported anxiety and depression levels

Aim 2b. Compare clinical staff outcomes, including interprofessional collaboration amongst care teams and staff assessment of patient safety and quality of care

Aim 2c. Evaluate unit/team level quality outcomes important to patients, clinical care teams, and institutional leadershiputilizing data available through CDRN's and EHRs

Aim 2d. Compare engagement of patients, families, and caregivers, clinical staff, and hospital administrators through qualitative interview process and focus groups about their experiences with traditional care model and ACU model.

Hypotheses (Quantitative): Patients hospitalized in units with high fidelity to ACUs will experience better patient centered outcomes, interprofessional team outcomes, and clinical outcomes compared to traditional care models.

Any preliminary data:

As a part of our current and past work, we have engaged our patient stakeholders, patient partners, clinical staff stakeholders (including nursing, providers, social workers, physical therapists and occupational therapists, and pharmacists), and administrative and clinical leaders. We have conducted multiple focus groups with patients and families, clinical staff, and administrative leaders to better understand what patients want and need in their care and similarly how care teams perceive and experience the way patients are cared for. We have also engaged our local Patient and Family Advisory Council and members of the Colorado Patient Partners in Research network (CoPPiR) along with national experts in ACU model to help us choose the outcome metrics and develop the study methodology. We have also submitted a comparative effectiveness study to PCORI to study mentored implementation of accountable care units compared to traditional care, which is under review and will be complementary to this proposed study.

Study design, including study subjects (patients and/or providers) and setting, comparator groups, data sources, outcomes, analysis plan, power and sample size, limitations, and timeline:

Study design. This is a multi-hospital, within-hospital, pragmatic observational comparative effectiveness trial. A mixed-methods evaluation will include quantitative (patient and frontline clinical staff surveys, data collection via electronic health records, and participation and intervention fidelity tracking data) and qualitative (individual interviews and focus groups of patients, families, and care providers, staff, and leadership).

Study Population and Setting: Adult Spanish and English-speaking patients admitted to a Medicine service and being cared for in participating hospital inpatient units. All CRG sites would be invited to submit applications to participate. Hospitals will be selected to ensure sufficient variability between low and high fidelity units across selected sites.

Comparators: Because of the evolving nature of the care of patients who are hospitalized, our research team and stakeholders felt that in most cases some components of the ACU model are already employed in many hospitals thus we chose as the comparator to have two or less of the ACU components.

1. Traditional care model. For the existing care model to be considered traditional, no more than two ACU characteristics can be in place in the participating hospital inpatient unit.

2. Accountable care unit model. For an existing model to be considered an ACU, the four key characteristics must be in place including geographically located teams, standardized interdisciplinary bedside rounding, nurse-physician dyad leadership model, and unit level reporting.

Outcomes: The primary outcomes, as selected in conjunction with our stakeholders, are patient-reported experience of shared decision-making, clinician-reported perception of interprofessional collaboration, and hospital length of stay. Secondary outcomes include patient-reported symptoms of anxiety and depression, clinician assessment of patient safety and quality, ICU length of stay, rapid response calls, patient falls, inpatient mortality, time of discharge, unexpected returns to the hospital within 30 days of discharge, and HCAHPS survey results (including the nurse and physician communication composites and pain control composite).

Data Sources: Data will come in a variety of formats including patient surveys, care team member surveys. Additional data sources will be from the electronic health record and CDRN databases including PCORnet. Data will include but will not be limited to: hospital length of stay, intensive care unit length of stay, rapid response calls, patient falls, inpatient mortality, time of discharge, and unexpected return to the hospital within 30 days of discharge. All clinical effectiveness outcomes data will be obtained from hospital EHRs and CDRN databases and thus availability of data will not be dependent on patient participation, allowing for estimates of effectiveness among all patients who received care in specific types of care models as well as sub-analyses among those patients who were willing and able to complete patient-centered outcomes surveys with the research assistant. EHR extracts will include patient demographics, patient clinical history data, quality outcomes, and HCAPHS survey results.

In addition to better understanding quantitatively the types of existing care models, we hope to also utilize a qualitative approach as well. By utilizing purposive sampling techniques to recruit perspectives from a variety of stakeholders we hope to hold several focus groups and individual interviews to better understand how a wide variety of stakeholders experience the care model being utilized by their hospital.

Analytic Plan: We will compare our primary outcomes among models of care across all hospital units using linear mixed models, assuming these scores are approximately Gaussian distributed. We will include a parameter to represent the random effect of hospital unit, as patients in the same hospital unit will likely have correlated outcomes. We will include a fixed binary effect for model type and other patient- and hospital-level factors that we presume will be associated with the outcomes, such as age, gender, race, primary language and discharge diagnoses. All analyses will be performed in a statistical software program, such as R or SAS.

Sample Size and Power Calculations. N (total) = 15 hospitals, 7,500 patients (1 unit per hospital and 500 patients per unit), allowing for 30% attrition (leaving approximately 5,250 patients or 350 patients per unit per hospital).

Limitations: We will be studying an array of care models with the foundation for the comparison being whether the model utilized is high versus low fidelity to the ACU model harnessing the power of a natural experiment. Sites that have adopted the ACU model of care may be more progressive or more focused on improvement and thus results may be biased by those inherent traits. If funded, we plan on having a robust process to solicit applications for participation and will prospectively assess both current and past improvement efforts as well as models of care previously utilized. We hope to select a wide array of institutions that are diverse including hospital settings such as community hospitals, academic medical centers and centers that care for underserved populations like safety net institutions. This research is also observational in nature and thus there could be unmeasured and unknown causes of variance because the study does not include randomization. Healthcare systems are dynamic and continually working on improvement efforts, which could also affect findings.

Timeline: 3-5 years

Characteristics of sites who might participate: We have developed a rich network of research partners across the country, including six sites with who we previously submitted an Improving Healthcare Systems PCORI grant together. We anticipate that these sites, along with additional hospitals with whom we have partnered for other projects, will participate in this study. We would also have a larger call to other healthcare systems across the country to recruit sites that are as diverse as possible and also having implemented the ACU model to varying degrees.

Potential funders with RFA/RFP and due dates:

1. Pragmatic Clinical Studies to Evaluate Patient-Centered Outcomes - Cycle 2 2017 – letter of intent due 7/25/2017

2. AHRQ Health Services Research Projects (R01) (PA-14-291) – expiration date July 6, 2018

References:

1. Stein J, Payne C, Methvin A, et al. Reorganizing a hospital ward as an accountable care unit. J Hosp Med 2015;10:36-40.

2. Landry MA, Lafrenaye S, Roy MC, Cyr C. A randomized, controlled trial of bedside versus conference-room case presentation in a pediatric intensive care unit. Pediatrics 2007;120:275-80.

3. Gonzalo JD, Kuperman E, Lehman E, Haidet P. Bedside interprofessional rounds: perceptions of benefits and barriers by internal medicine nursing staff, attending physicians, and housestaff physicians. J Hosp Med 2014;9:646-51.

4. Kara A, Johnson CS, Nicley A, Niemeier MR, Hui SL. Redesigning inpatient care: Testing the effectiveness of an accountable care team model. J Hosp Med 2015;10:773-9.

PIs:  Auerbach, Najafi (UCSF)

Co-Investigators (see below): Raman Khanna UCSF, Arora (Univ Chicago), Gupta (UCLA), Moriates (UT Austin).

Although there is intense scrutiny on costs of care and improving healthcare value, there are few data to describe how trainees and attending physicians’ day-to-day workload, local culture, and training interact to produce variations in the value of care delivered.  As a result, there is a substantial gap in our ability to train physicians towards a value-focused future practice style.

Aim: The overall goal of this study is to develop a prospective and scalable approach to gaining periodic, valid snapshots of trainees’ and attending physicians’ practices in delivering healthcare. 

Hypotheses: That 'ecological momentary assessments' timed to clinical care work will provide important contextual information needed to understand issues such as burnout, specific practice patterns (such as choice of high or low-value tests or medications).  

Methods: We will utilize electronic health record systems (Epic, primarily) linked to call schedules maintained at each site, to carry out prospective momentary ecological assessments – periodic surveys delivered at or near the time specific clinical experiences take place - of how trainees and attending physicians perceptions of the key factors (culture, training, workload) which are highly associated with choices at the point of care. As a key first metric, we will focus on newly validated ‘culture of value’ indicators derived by Drs. Gupta, Arora, and Moriates.

Gaining understanding of front line providers’ perceptions will provide a critical layer of contextual data needed to develop training models, frame approaches to workforce issues (e.g. novel work hours models), as well as enhanced decision support needed to overcome barriers to best practices (or take advantages of opportunities to improve care more rapidly).

Our project will be led by investigators at UCSF (Drs. Auerbach, Najafi, Khanna), University of Chicago (Vineet Arora MD, MAPP), the University of California, Los Angeles (Reshma Gupta MD), as well as the Dell Medical School at The University of Texas at Austin (Christopher Moriates MD);.

We plan to use FHIR AP access to Epic systems at each site to identify general medical patients, certain target care practices (such as obtaining a CT scan for dyspnea, or repeated CBC’s in patients with pneumonia) and the identifier of the team caring for them.  

By linking Epic data to call schedule data (Amion.com or similar) maintained at each site, we can then deliver personalized surveys to provider physicians within a period of time likely to reduce recall bias and increase validity of the results. Examples of surveys to be tested and disseminated include the high-value care culture survey developed by members of our team, surveys examining the perceived utility of specific testing, as well as surveys asking about burnout and job satisfaction. These surveys will be delivered via web-based survey and subsequently linked to clinical and/or administrative data (e.g. total costs of care for pneumonia) from each site.

Our technical approach is built on one successfully piloted by Nader Najafi MD in the Division of Hospital Medicine at UCSF, and will be extended through expertise we are gaining (FHIR APIs) as part of Dr. Auerbach and Dr. Khanna’s work developing the UCSF Digital Diagnostics and Therapeutics Program,  an oversight body charged with the implementation of FHIR integration at UCSF. Finally, and most importantly, our program builds on the leadership and extensive expertise of Drs. Arora, Gupta, and Moriates in defining optimal work hour models, approaches to assessing and improving healthcare value, and GME training,

Study design: We propose a technical feasiblity pilot at sites with Epic health system EHRs, and which utilize AMION for hospitalist and resident call schedules.

Characteristics of sites who might participate: Should have local expertise in API-driven integrations with Epic, use of a standard call schedule program such as Amion. 

Potential funders: We anticipate applying for CTSA supplementary funding in November 2017.