Penn Medicine Heart Failure Solution

Heart Failure solution team

Lee R. Goldberg, MD, MPH

Medical Director, University of Pennsylvania Heart Failure and Transplantation program

Jo Anne Fante-Gallagher, MHA, BSN, RN

Director of Quality & Safety for Heart & Vascular Service Line at the University of Pennsylvania Health System

Yevgeniy Gitelman, MD

Clinical informatics manager at the center for health care innovation.

Susan Harkness Regli, Ph.D

Human Factors Dcientist for the University of Pennsylvania Health System.

Asaf Hanish

Asaf Hanish is a Data Scientist on the Penn Medicine Predictive Healthcare team.

Heart failure continuity project

The goals of this work is to set the framework for delivering reliable care and a reliable patient experience to the hf patient population at penn medicine by the elimination of unidentified heart failure inpatients, the connection and provisioning of cardiology specific care and the standardization of care to a post-acute setting. The long-term care management of patients with a history of heart failure (hf) is a major challenge for the healthcare industry as a whole. Patients with a history of hf require coordination of care to manage their complex comorbidities, arrest disease progression and insure the highest possible quality of life. The proper coordination of care includes both the identification of at risk patients and application of a standard treatment protocol.

High risk of readmission for heart failure patients

The hf patient readmission risk stratification application performs a daily audit of the electronic medical record (emr). The hf risk stratification algorithm was developed in close consultation with dr. Goldberg and jo anne fante-gallagher to assess each hf patient’s risk of returning to an emergency room (regardless of admission status) within 30 days of discharge. The algorithm is based on the patient’s clinical, diagnostic and previously health care resources utilization pattern.

Coded diagnostic history

Clinical and demographic characteristics

Health system resource utilization

Patient population

Heart Failure Alert Inclusion Criteria:

Diagnosis of Heart Failure at a previous Inpatient or Outpatient encounter.

Readmission label

We trained the algorithm to detect patients with a history of Heart Failure, who return to a Penn Medicine Emergency room within 30 days of discharge. To do this, we searched all patient encounters between January 2013 and January 2015 who met the inclusion criteria (n = 9,989).

Algorithm performance

With over 500 patient features, a machine learning algorithm was trained using 10 fold cross validation and resulted in an AUC of 0.68 (+/- .03).

Operational metrics

The high risk for readmission alert identifies about 9 at risk patients every day across the 3 downtown hospitals. The identified patients have a readmission rate of 35%, which is about 3x higher then the non-alerted population.