2017 HSR&D/QUERI National Conference

4067 — Cohort Identification in the VA with the C3PO System

Lead/Presenter: Lewis Frey, COIN - Charleston
All Authors: Frey LJ (Ralph H Johnson VA Medical Center) Lenert LA (Medical University of South Carolina) Duvall SL (University of Utah) Turano A (VA Pittsburgh Healthcare System) Nebeker J (VA Salt Lake City Health Care System)

Objectives:
Clinical Personalized Pragmatic Predictions of Outcomes (C3PO), an open source medical research big data platform, is used for deep phenotyping of tends in patient data obtained from the U.S. Department of Veterans Affairs (VA) Informatics and Computing Infrastructure (VINCI). C3PO is used to identify cohorts for reference patients from sequences of their clinical processes using a novel application of the Smith-Waterman gene alignment algorithm.

Methods:
Cohort identification is conducted for 10 archetypal patient patterns using approximately 9,000 VA type 2 diabetes patients undergoing knee and hip surgery (fiscal years 2006-2009). Specifically, post surgery adverse events were associated with trends in glucose control for a year prior to knee and hip surgery. Our novel application of the Smith-Waterman algorithm does sequence alignment on clinical events by using discrete time-ordered event sequences of glucose measurements.

Results:
The C3PO system dynamically searched for temporal features of mean monthly values of HA1c values over the 12 months prior to surgery using the Hadoop data store of HA1c lab values encode in Operational Medical Outcomes Partnership (OMOP) compatible model. It accomplished this through the use of a feature extraction module. The sequence alignment successfully identified meaningful patterns such as all patients that had HA1c consistently in the range of [6-7) within the 12 months prior to surgery. The C3PO system also has visualization functionality to view the cohorts of interest. The analysis of adverse events associated with sequence-based cohorts is ongoing research.

Implications:
The C3PO system has successfully been deployed and used to identify patient cohorts based on glucose levels over a year prior to knee and hip surgery. The alignment methodology used to match patients with archetypal temporal patterns is an innovation that has the potential to empower clinicians to use the patterns of patients previously treated to inform care for the patient at hand.

Impacts:
The ability to forecast the outcome of a patient based on the collective experience of other patients in electronic health records (EHR) is an open problem in medicine. With the growing global adoption of EHRs, creating systems that tap into these storehouses and identify informative cohorts for improving outcomes could revolutionize healthcare.