Development and Verification of a “Virtual” Cohort Using the National VA Health Information System

    loading  Checking for direct PDF access through Ovid

Abstract

Background:

The VA's integrated electronic medical record makes it possible to create a “virtual” cohort of veterans with and without HIV infection to monitor trends in utilization, toxicity, and outcomes.

Objectives:

We sought to develop a virtual cohort of HIV-infected veterans by adapting an existing algorithm, verifying this algorithm against independent clinical data, and finally identifying demographically-similar HIV-uninfected comparators.

Research Design:

Subjects were identified from VA administrative data in fiscal years 1998–2003 using a modified existing algorithm, then linked with Immunology Case Registry (ICR, the VA's HIV registry) and Pharmacy Benefits Management (centralized database of outpatient prescriptions) to verify accuracy of identification. The algorithm was modified to maximize positive predictive value (PPV) against ICR. Finally, 2 HIV-uninfected comparators were matched to each HIV-infected subject.

Results:

Using a single HIV code, 30,564 subjects were identified (positive predictive value 69%). Modification to require >1 outpatient or 1 inpatient code improved the positive predictive value to 88%. The lack of confirmatory laboratory and pharmacy data for the majority of subjects with a single outpatient code also supported this change. Of subjects identified with the modified algorithm, 89% had confirmatory evidence. When the modified algorithm was applied to fiscal years 1997–2004, 33,420 HIV-infected subjects were identified. Two HIV-uninfected comparators were matched to each subject for an overall cohort sample of 100,260.

Conclusions:

In the HAART era, HIV-related codes are sufficient for identifying HIV-infected subjects from administrative data when patients with a single outpatient code are excluded. A large cohort of HIV-infected subjects and matched comparators can be identified from existing VA administrative datasets.

Related Topics

    loading  Loading Related Articles