Towards a Learning Health System to Reduce Emergency Department Visits at a Population Level.

High utilizers of the Emergency Department (ED) often have complex needs that require coordination of care between multiple organizations. We describe a Learning Health Systems (LHS) approach to reducing ED visits, in which an intervention is delivered to a cohort of high utilizers identified using population-level data and predictive modeling. We focus on the development and validation of a random forest model that utilizes electronic health record data from three health systems across two counties in Michigan to predict the number of ED visits each resident will incur in the next six months. Using 5-fold cross-validation, the model achieves a root-mean-squared-error of 0.51 visits and a mean absolute error of 0.24 visits. Using time-based validation, the model achieves a root-mean-squared error of 0.74 visits and a mean absolute error of 0.29 visits. Patients projected to have high ED utilization are being enrolled in a community-wide care coordination intervention using twelve sites across two counties. We believe that the repeated cycles of modeling and intervention demonstrate an LHS in action.

Clinical redesign using all patient refined diagnosis related groups.

OBJECTIVE: Clinical redesign of processes in hospitals that care for children has been limited by a paucity of severity-adjusted indicators that are sensitive enough to identify areas of concern. This is especially true of hospitals that analyze pediatric patient care using standard Centers for Medicare and Medicaid Services (CMS) diagnosis-related groups (DRGs). The objectives of this study were to determine whether 1) utilization of all-patient refined (APR)-DRG severity-adjusted indicators (length of stay, cost per case, readmission rate) from the National Association of Children's Hospitals and Related Institutions (NACHRI) database could identify areas for improvement at University of Michigan Mott Children's Hospital (UMMCH) and 2) hospital staff could use the information to implement successful clinical redesign. METHODS: The APR-DRG Classification System (version 20) was used with the NACHRI Case Mix Comparative Database by severity level comparison from 1999 to 2002. Indicators include average length of stay (ALOS), case mix index, cost per case, and readmission rate for low acuity asthma (APR-DRG 141.1). UMMCH cases of 141.1 (n = 511) were compared with NACHRI 141.1 (n = 64,312). Although not part of the standard report, mortality rates were calculated by NACHRI for UMMCH and an aggregate of NACHRI member children's hospitals. RESULTS: Data from 1999 revealed that in noncomplicated asthma cases (level 1 severity), the UMMCH ALOS versus NACHRI ALOS was slightly longer (UMMCH 2.16 days vs NACHRI 2.14 days), and the cost per case was higher (UMMCH $2824 vs NACHRI 2738 dollars), whereas levels 2, 3, and 4 cases (moderate, major, and extreme severity) indicated the ALOS and cost per case were lower than the national aggregate. This showed that the APR-DRG system was sensitive enough to distinguish variances of care within a diagnosis according to severity level. After analysis of internal data and meeting with clinicians to review the indicators, 3 separate clinical processes were targeted: 1) correct documentation of comorbidities and complications, 2) standardized preprinted orders were created with the involvement of the pediatric pulmonologists, and 3) standardized automatic education for parents was started on the first day of admission. Yearly data were reviewed and appropriate adjustments made in the education of both residents and staff. In 2002, the UMMCH ALOS dropped to 1.75 +/- .08 days from 2.16 +/- .09. In 2002, the NACHRI ALOS was 2.00 days +/- 0.01 versus the UMMCH ALOS of 1.75 days +/- 0.0845, indicating that the UMMCH ALOS dropped significantly lower than the NACHRI aggregate database over the 3-year period. Cost per case of UMMCH compared with NACHRI after the 3 years indicated that UMMCH increased 12%, whereas the NACHRI aggregate increased 18%. These data show that length of stay and cost per case relative to the national database improved after clinical redesign. Improvements have been sustained throughout the 3-year period. Readmission rates ranged from 2.97% to 0.80% and were less than the national cohort by the third year. There were no mortalities in the UMMCH inpatient asthma program. This demonstrates that clinicians believed that the data from the APR-DRG acuity-adjusted system was useful and that they were then able to apply classical clinical redesign strategies to improve cost-effectiveness and quality that was sustained over 3 years. CONCLUSIONS: Severity-adjusted indicators were useful for identifying areas appropriate for clinical redesign and contributed to the improvement in cost-effective patient care without a detriment in quality indicators. This method of using a large comparative database, having measures of severity, and using internal analysis is generalizable for pediatric hospitals and can contribute to ongoing attempts to improve cost-effectiveness and quality in medical care.