Identification of pulmonary arterial hypertension patients in coded healthcare databases: a clinical evaluation study

Abstract

Coded&nbsp;healthcare databases can be used for studies in rare diseases like Pulmonary Arterial Hypertension (PAH), a form of Pulmonary Hypertension (PH). However, identifying&nbsp;adequate&nbsp;patient cohorts based on administrative codes is challenging as unique codes and clinical outcomes are not widely available. <b>Aim:</b> To evaluate the accuracy of different combinations of administrative codes (algorithms) for PAH patient identification in coded US Electronic Health Records (EHRs) linked with a clinical PH database <b>Method:</b> Over a dozen published PAH algorithms and clinically meaningful&nbsp;extensions were applied to identify PAH patients in the coded EHR data. Each algorithm’s accuracy was assessed against the patients’ true clinical diagnoses in the PH database. <b>Results:</b> Out of 828 PH patients with linkable EHRs and clinical records, 646 had a confirmed PAH diagnosis. Algorithms with only one PAH-related diagnosis code classified nearly all PAH and non-PAH patients as PAH patients (sensitivity: &gt;97%; positive predictive value (PPV): &gt;79%; specificity: &lt;13%).&nbsp;One of the most complex algorithms consisting of PAH diagnosis and drug codes&nbsp;and exclusionary codes classified only 1 non-PAH patient as PAH patient but missed 606 true PAH patients (sensitivity: 6%; PPV: 98%; specificity: 99%). Most algorithms with a complexity between these extremes were balanced in terms of sensitivity, PPV, and specificity. <b>Conclusion:</b> Algorithm accuracy varied widely. Algorithm choice should thus be aligned with the study objective. E.g., to analyze diagnostic journeys, a complex PAH algorithm excluding other forms of PH should be used. Performance results will likely differ in other databases&nbsp;or regions.