A Regional Payer/Provider Partnership to Reduce Readmissions The Bronx Collaborative Care Transitions Program: Outcomes and Lessons Learned Stephen Rosenthal, MBA President and COO, Montefiore Care Management Henry Chung, MD Chief Medical Officer, Montefiore Care Management Anne Meara, RN, MBA Associate Vice President, Montefiore Care Management Herb Fillmore Vice President, Strategic Innovation, Treo Solutions
Presentation Outline The Bronx Collaborative Description of the Care Transitions Program Predictive Model Evaluation Findings Lessons Learned Discussion Please note that the views expressed by the conference speakers do not necessarily reflect the views of the American Hospital Association and Health Forum. 2
The Bronx 1.4 million residents in the poorest urban county in the nation 30% live at or below the poverty line 12.5% unemployment rate 89% non-white minority 43% of children are at or below poverty 54% Hispanic, 37% African-American High burden of chronic disease 12% have diabetes 30% are obese 16% have asthma Per capita health expenditures 22% higher than national average 80% of health care costs paid for by government payers 3
What is the Bronx Collaborative? Incorporated as a New York State not-for-profit in 2009 with goal of improving health care delivery in the Bronx 3 major delivery systems Bronx Lebanon Hospital Center 593 beds, 29,000 admits/yr. Montefiore Medical Center 1,491 beds, 90,000 admits/yr. St. Barnabas Hospital 461 beds, 21,000 admits/yr. 2 major payers EmblemHealth Healthfirst Each of the 5 participating entities has equal representation on the Board of Directors Covers 16% of the county s population or 220,000 Bronx residents with Medicaid, Medicare and commercial insurance Accounts for 27% of Medicare managed care and 22% of Medicaid managed care markets in the Bronx 4
Why focus on care transitions? Potential win-win for providers and payers Program targets health plan members that hospitals are at risk for Medicare and Medicaid readmission penalties Research suggests readmissions can be reduced by addressing problems during transitions of care 5
Unique Features of the Collaborative s program Active participation of payers Program conducted across multiple hospitals RHIO used to establish electronic care transition record, facilitate data exchange/reporting and support program uniformity Using predictive model to target high-risk cases Focus on 60-day readmissions vs. 30-day More clinically diverse and socially disadvantaged population than most other programs 6
Care Transitions Program Financing Hospitals Facility-based staff Supervision, office space, equipment Payers One-time fee for cases with interventions Care transition home visits New York Community Trust Collaborative formation New York State Health Foundation Program design Electronic Care Transitions Record and RHIO costs Staff training Montefiore Care Management Project management Evaluation 7
Care Transitions Program Design Built on evidence-based models including: Project Red, BOOST, Coleman and Naylor Developed by team from all 5 partners IRB-approved design Standardized training Centralized project oversight and coordination Shared data collection system in RHIO Centralized evaluation resources 8
Eligible Population Bronx residents aged 50 and older Medicare, Medicaid and Commercial members of the 2 health plans English- or Spanish-speaking (patient or caregiver) with a telephone Admitted to Medicine Service and expected to be discharged home (not SNF) Excludes: HIV/AIDS, transplant, dialysis, psych, substance abuse, homeless, elective admits, and cognitively-impaired patients without an active caregiver 9
Staffing Operational Components 1 RN Care Transition Manager and.5-time Care Transition Analyst in 4 hospitals Shared Pharmacist Centralized Program Coordinator Reimbursement by payers Personal Discharge Record Predictive model Program and RHIO consents for all patients 10
Program Overview STEP 1 STEP 2 STEP 3 STEP 4 STEP 5 STEP 6 Care Transition Manager (CTM) provides education on red flags, Rx review, postdischarge MD visit, contact info 48-72 hours CTM follows up on red flags, Rx reconciliation, referral to Pharmacist if needed, reinforce need for MD visit, address questions Care Transition Pharmacist does Rx reconciliation, resolves RX issues (only for those referred by the CTM) 7-14 days CTM checks on MD visit w/in 14 days and checks for problems, assists in resolving open issues 15-60 days CTM follows up on any remaining issues, transitions to care management if needed Home care RN does transition home visit (only for those readmitted within 60 days) = Qualifying Interventions 11
12 Here s the problem
Treo Readmission Prediction Tool (RPT) Problem : Limited resources for care management Need to target patients most in need at a time of greatest vulnerability Solution: Prioritize new admissions for intensive care management interventions by Predicting probability of readmission on the morning after an admission using a small set of readily available administrative variables 13
RPT Applications Each morning a list of new admissions with key historical variables is automatically transmitted to Treo electronically RPT scores are applied, the list is ranked by priority and placed on a secure site for case manager retrieval Case managers retrieve list and begin working with those having highest priority of readmission 14
Readmission Prediction Tool (RPT) Raw data elements, before enrichment Discharges in last 12 months, Length of Stay of last previous admission, Readmissions in last 12 months, Health status and disease severity, Age, Gender, Race, Zip Code, Payer All of the above variables predate the current admission and are facility-specific Original training data set based on census of New York State, followed by census of Bronx discharges in 2008. Validation data set from census of Bronx discharges in 2009. 15
Why These Variables & Not Others These are the variables that are reliably available the day after admission. Other variables like diagnosis of current admission, source of admission, length of stay of current admission, EMR data, and patient social characteristics might improve the model. Even without additional variables, the current model s statistical performance is about 40% better than flipping a coin. (Validation C=.67) This performance is better than many disease-specific models with many times more variables requiring extensive data collection. 16
Other Predictive Models Most focus on specific diseases. Many do not report validation statistics Some mix readmission with mortality Some have no exceptions for trauma or metastatic malignancies or other conditions that would be expected to have readmissions Many rely on expensive non-administrative data 17
The C The C-statistic (area under a ROC curve) measures the test s ability to discriminate between those who will be admitted from those who will not. A useless test (no better than flipping a coin) has a C 0.5. A perfect test (zero false positives and zero false negatives) has a C of 1.00. 18
Other Models Model Year DepVar Admin C statistic data Only? LACE (Walraven) 2010 30 day death or unplanned No.684 (V) readmission Felker et al 2004 60 day death or all cause No.69 (NV) readmission among HF patients Yamokoski et al 2007 180 day all cause No.60 (NV) readmission among HF patients Kroch et al 2010 All cause 30 day No.682 CMS All cause 30 day readmission HF, heart attack, pneumonia PARR (Billings) 2006 All cause 365 day readmission Kansagara (Meta Review) JAMA October 19 2011 26 models. Most common depvar was 30 day.63 Yes *.685(V) Mixed.56-.83 19 * Inpt & Otpt
20 Here s the problem
21 Next Generation
Program Population 775 Selected using predictive model 190 Comparison Group 585 Intervention Group 500 2 Qualifying Interventions 85 1 Qualifying Intervention 22
Population Profile--1 Descriptor Intervention (n=585) Comparison (n=190) Patients with previous admission (at index CTP hospital) 44% 51% Mean predictive score 21.5 22.5 Mean age 67.8 67.4 % Males 33.7% 30.0% Race - Black 32.9% 31.6% - Hispanic 46.5% 57.4% - White 6.8% 6.3% - Other/Missing 13.8% 4.7% Spanish speaking 33.5% 43.7% 23
Population Profile--2 Descriptor Intervention (n=585) Comparison (n=190) Line of business - Medicare 64.4% 59.5% - Medicaid 23.8% 32.6% - Commercial 11.8% 7.9% Index hospital - Bronx Lebanon 10.6% 24.2% - Moses (Montefiore) 34.4% 24.2% - St. Barnabas 11.8% 26.8% - Weiler (Montefiore) 43.2% 24.7% Average index LOS 5.0 4.7 24
Evaluation Methods Multivariate logistic regression to evaluate impact of the program on 60- day readmissions for all program participants (n = 585) Evaluated 500 individuals who received minimum of two qualifying interventions Performed further review of the 85 patients who received only one intervention 25
Data for Program Evaluation Care Transitions Record in Bronx RHIO Claims data from payers Hospital EMRs Program staff interviews 26
Levels of Intervention (n=585) 0% 20% 40% 60% 80% 100% At least 1 qualifying intervention Post-discharge booklet At least one completed post-discharge call 100% 99% 93% Two completed calls 62% Pre-discharge education session with CTM Call from the pharmacist 38% 46% 3 or more post-discharge calls Care transitions home visit 11% 11% 27
Outcome Metrics Intervention Total (n=585) Intervention Subgroup with at least 2 interventions (n=500) Intervention Subgroup with 1 intervention (n=85) Comparison (n=190) 60-day readmissions 22.8% 17.6% 52.9% 26.3% 30-day readmissions 14.9% 9.4% 47.1% 17.9% % of patients with ED visit within 60-days % of patients with MD visit within 14 days 18.4% 19.1% 22.4% 18.4% 70.4% 74.4% 47.1% 61.1% Note: Comparison Group received usual care. 28
Key metric: 60-day readmit rate 22.8% readmit rate for total Intervention group is 21.4% lower than the baseline rate of 29% 13.3% lower than the Comparison group rate of 26.3% 17.6% readmit rate for those with 2 or more qualifying interventions is 39.3% lower than the baseline rate 33.0% lower than the Comparison group rate 52.9% readmit rate for those with only 1 qualifying intervention is 200% higher than the rate of those with 2 or more qualifying interventions 100% higher than the Comparison group rate 29
Variation in Program Groups 30 1 Intervention (n=85) Intervention (n=585) 2 or more Interventions (n=500) Comparison (n=190) Usual Care 60-day Readmissions 52.9% 17.6% 26.3% Mean (Median) # days to readmit 11.5 (5) 28.4 (28) 23.8 (19) Mean Index LOS 6.9 4.7 4.7 Index AMA Discharge 5.9% 0.8% 2.6% Office Visit Within 14 Days of Admission Patients with Previous Admission (at same hospital) 47% 74% 61% 51.7% 43.2% 50.5% Charlson Score 2.25 2.11 2.09 Mean Predictive Score 21.9% 21.4% 22.5% Pre-Discharge Contact 34.1% 51.2% Not applicable Post-Discharge Contact 45.9% 99.8% Not applicable
5 key factors that contribute to readmissions Medicare patient No follow-up physician office visit w/in 14 days < 2 qualifying interventions Charlson score > 2 Readmission Longer lengths of stay 31
Lessons Learned: Collaboration All Collaborative members consider the program a success Collaboration and competition can co-exist if incentives are appropriately aligned Value in obtaining input/perspective from multiple organizations in program design Opportunities exist to: Leverage resources among hospitals, e.g. training Leverage resources of health plan, e.g. streamline hand offs for longer term case management 32
Lessons Learned: RHIO Value in utilizing RHIO for information exchange Problems using the RHIO as the primary platform for the care transition program Duplicate data entry required into hospital system and RHIO Need a care planning system designed to: take data from multiple sources without separate data entry support workflows to alert appropriate staff as needed 33
Lessons Learned: Predictive Model Patients with a prior admission more likely to be readmitted Model limited by data available at point of admission from all hospitals Need to Refine Predictive Model with additional data elements Use predictive results to stratify population Consider using 1 model at admission; 1 at discharge 34
Lessons Learned: Care Transitions Program Effectiveness--1 Key program activities: MD office visit within 14 days of discharge Minimum of 2 qualifying interventions (contacts) Dedicated staff with comprehensive, standardized training and oversight is effective, but difficult to scale Need to enhance usual care in hospital and post-discharge to address program scalability Need to develop an equitable reimbursement methodology 35
Lessons Learned: Care Transitions Program Effectiveness--2 Current evidence-based practices fit most, but not all, patients Very high readmit rates for those with only 1 qualifying intervention Staff feedback suggests psychosocial differences account for higher readmission rates Neither claims data nor medical record reviews explain high readmit rate for those with only 1 intervention Need to continue to explore causes of readmit rate variation to develop strategy for proactive identification of higher risk group and then identify appropriate interventions 36
Factors Affecting Readmission Clinical/ Health System Lack of timely follow up care (primary, specialty, testing, home care) Insufficient patient education on medication, signs and symptoms, Patient co-morbidities and severity of illness Stage of patient s primary disease Polypharmacy, medication adherence/ compliance Readmission Psychosocial Patient s insurance Homeless, unstably housed Transportation problems Lack of involved care giver Health literacy Financial barriers Language / cultural barriers Care giving responsibilities Self-perception of health Stress Depression Cognitive impairment Substance abuse 37
Accountability for Preventing Readmissions Accountable Care Community Patient/ Caregiver Hospital PCP/ Specialist Health System Health Plan Prevent Readmissions 38