Evaluation of a Computable Phenotype for Idiopathic Pulmonary Fibrosis
Open Access
- Author:
- Dimmock, Anne Elizabeth
- Graduate Program:
- Public Health Sciences
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- None
- Committee Members:
- Cynthia H Chuang, Thesis Advisor/Co-Advisor
Wenke Hwang, Thesis Advisor/Co-Advisor
Herbert Y Reynolds, Thesis Advisor/Co-Advisor
Junjia Zhu, Thesis Advisor/Co-Advisor - Keywords:
- idiopathic pulmonary fibrosis
computable phenotype
electronic health record
rare disease - Abstract:
- Introduction: The electronic medical record (EMR) is a common source of data for clinical research. However, the quality of EMR-based research depends on the validity of computable phenotypes, defined as computerized queries to an EMR system or clinical data repository to identify individuals with a clinical condition. Computable phenotyping is especially of interest with rare diseases, such as idiopathic pulmonary fibrosis (IPF). Our objective was to evaluate a computable phenotype for the identification of patients with idiopathic pulmonary fibrosis. Methods: A computable phenotype was adapted from previously published algorithms, which identified patients in our health system if they had at least one ICD-9 code of 516.3 (idiopathic interstitial pneumonia) or 516.31 (idiopathic pulmonary fibrosis) in the past three years associated with inpatient or outpatient visits, excluding emergency department and lab encounters. Patients were excluded if an ICD-9 code for connective tissue disorders was present. Gold standard diagnosis was determined by chart review by at least two reviewers using a standardized procedure. In cases of disagreement, charts were reviewed collectively until agreement was reached. Results: The computable phenotype identified 157 individuals with IPF. After gold standard chart review, 74 patients (47%) were classified as having IPF based on a narrow definition and 89 (57%) were classified as having IPF based on a broad definition. Conclusion: Our computable phenotype yielded a positive predictive value of 47% using a narrow definition of IPF. True positives were older (p=0.0003) than false positives. Studies depending on computable phenotypes for identifying patients with IPF will be limited by false positives.