Using Administrative Health Data to Identify Patients with NF1 in Ontario, Canada, and to Assess Prevalence, Mortality, and Health Care Utilization Patterns
Abstract
We conducted a free text search in EMRALD using NF-related terms, to identify potential individuals with NF1. Out of 593,408 records in EMRALD, 273,444 were eligible for search. 2,200 records had at least one search term, and we prioritized 837 charts with more than 1 term for initial review. Trained chart abstractors identified 42 definitive NF1 cases, and 228 possible cases. Inter and intra rater agreement for chart abstractors ranged between 96 to 100 percent. After final review, there were 71 confirmed cases and 37 possible cases. A random review of 300 charts with only one search term did not result in any cases. The minimal prevalence of NF1 in this cohort was 1 in 3851. A within EMR algorithm to identify NF1 cases though free text search had sensitivity of 85.9 percent and 100 percent specificity, with 78 percent PPV and 100 percent NPV. All false positives were possible cases. We tested is similar to214 billing algorithms using different combinations of outpatient and/or inpatient billing codes. However, no algorithm had acceptable positive predictive value, although all had 100 percent NPV.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 2020
- Accession Number
- AD1123926
Entities
People
- Carolina Barnett-tapia
Organizations
- University Health Network