Abstract
Rationale Chronic obstructive pulmonary disease (COPD) is a leading cause of mortality in the United States. Electronic health records provide large-scale healthcare data for clinical research, but have been underutilized in COPD research due to challenges identifying these individuals, especially in the absence of pulmonary function testing data.
Objectives To develop an algorithm to electronically phenotype individuals with COPD at a large tertiary care center.
Methods We identified individuals over 45 years of age at last clinic visit within Vanderbilt University Medical Center electronic health records. We tested phenotyping algorithms using combinations of both structured and unstructured text and examined the clinical characteristics of the resulting case sets.
Measurement and Main Results A simple algorithm consisting of 3 International Classification of Disease codes for COPD achieved a sensitivity of 97.6%, a specificity of 76.0%, a positive predictive value of 57.1%, and a negative predictive value of 99.0%. A more complex algorithm consisting of both billing codes and a mention of oxygen on the problem list that achieved a positive predictive value of 86.5%. However, the association of known risk factors with chronic obstructive pulmonary disease was consistent in both algorithm sets, suggesting a simple code-only algorithm may suffice for many research applications.
Conclusions Simple code-only phenotyping algorithms for chronic obstructive pulmonary disease can identify case populations with epidemiologic and genetic profiles consistent with published literature. Implementation of this phenotyping algorithm will expand opportunities for clinical research and pragmatic trials for COPD.
