Title

Combining Electronic Health Record Based Automation and Machine Learning to Improve Heart Failure Outcomes, Increase Equity and Retain at Risk Funding in a Safety Net Health System

Introduction

Providing comprehensive and equitable heart failure (HF) care in public, safety net healthcare systems (SNH) is challenging. Patient populations who tend to rely on SNHs often face numerous social determinants of poor health outside of and within the hospital walls, and SNHs often do not possess resources to address these specific needs. As a result, many SNHs experience elevated readmission rates and increased mortality. Beyond the impact to patients, poor outcomes put SNHs at risk of not meeting government pay for performance metrics, compounding their inability to invest in care improvement. Hospitals with the greatest need for funding are disproportionately penalized, and there is a paucity of successful efforts to reduce readmissions and retain funding at SNHs. HF is the greatest contributor to unplanned readmission, yet, targeting this diagnosis comes with additional challenges: low readmission rate often comes at the expense of elevated mortality, and system level interventions can exacerbate existing gaps in demographic equity in outcome. In 2018, a comprehensive HF care improvement program was formed at ZSFGH, a publicly funded SNH, to combat high readmission and mortality rates and inequities in care. This program systematically identified and addressed key barriers to high quality HF care and combined electronic health record (EHR) based automation with machine learning to build a clinical decision support tool called a carepath. The HF carepath improved medication management, and streamlined referral processes, enhancing resource allocation to the neediest patients.

 

Hypothesis/Question

Did the HF improvement program successfully reduce overall risk of unplanned 30-day readmission, without increasing mortality rate or worsening inequities in risk of readmission?

 

Aims

1)     Compare risk of 30-day readmission at our SNH after program initiation to internal historical controls and five external SNHs in California over the same time-period.

2)     Compare survival after program initiation to internal historical controls.

 

Project Design/Methods

At 55 months post program launch, we conducted a retrospective comparative analysis of 30-day all-cause readmission rates and survival with both internal historical controls and external parallel controls from five other urban public healthcare systems in California. All patients ≥18 years old and admitted for a principal diagnosis of heart failure (ICD 10 150.01-150.9) between October, 2015 and January, 2023 at our SNH, Zuckerberg San Francisco General Hospital (ZSFGH), or one of the five additional SNHs were included. All data was collected and reported by a third party, Vizient. We utilized mixed-effects logistic time series analyses to evaluate readmission rates and multivariate Cox proportional hazard regression to evaluate for survival.

All models were adjusted for age, sex, and Charlson comorbidity index.