Author: Anthony Lin, DIHI 2018 Impact Report
The Opportunity in Health Data
Since its implementation in 2013, Epic has enabled the creation of a comprehensive health record for all patient encounters at Duke Health to aid clinicians in healthcare delivery. However, given the 125,000 different data fields used to store information and the inherent complexity of Epic’s data backend, the health system’s ability to derive meaningful insights from clinical care data has become bottlenecked. Our ability to use data-driven insights to inform academic, operational, and learning health system strategy becomes limited by the technical barrier to access, clean, and validate health system data.
Our Experience
The Duke Institute for Health Innovation (DIHI) has curated large datasets for more than a dozen data science projects over the past five years. These datasets have spanned both the outpatient and inpatient setting, and have been used to derive a variety of clinical and operational insights. They have enabled predictive modeling of chronic kidney disease, first hospital admissions, and sepsis, as well as informed quality improvement of new care delivery models and innovation pilots. Over the course of gathering and validating these datasets with clinical experts, DIHI has developed a suite of extensible and reproducible tools to rapidly curate datasets.
A Culmination of Effort
DELPHI (Duke Environment for Learning and Promoting Health Innovation) is the culmination of years of work in understanding how to leverage Duke clinical operations data to support investigators and health system leaders hoping to drive change in healthcare delivery. This data asset facilitates the rapid exploration and analysis of validated clinical data in a large, diverse, and comprehensive inpatient population at Duke University Hospital. DELPHI extracts clinical care data from our EHR relational reporting database and cleans, normalizes, and standardizes the data elements. We’ve worked with clinical domain experts to validate the data elements and populate them with meaningful metadata to enable grouping of high-level features and facilitate custom disease phenotyping. To-date, DELPHI contains millions of data points ranging from encounter characteristics, patient demographics, and transfer times to laboratory results, vital signs, and medication administrations.
DELPHI’s breadth of curated features, speed of access, and complete transparency of data curation processes enable our healthcare community to leverage data-driven insights in a matter previously unrealized. It greatly reduces the time to procure meaningfully curated health data and empowers investigators and health system leaders with a clearer understanding of the clinical care they seek to improve.
