PAH diagnostic algorithms vary widely
New data show that pulmonary artery hypertension diagnostic algorithms that have been developed using administrative health data differ significantly in design and predictive value.
In a search for articles that applied an algorithm to an administrative or electronic health record database to identify PAH, researchers identified four studies that validated the algorithm against a reference standard — medical chart review by a pulmonary hypertension specialist. Chart review included ICD diagnosis codes and procedure codes, as opposed to granular data, in two studies.
Three of the four studies used a single visit-linked ICD diagnosis code for case definitions and one study developed a series of eight algorithms, including both claims-based and EHR-based algorithms. Claims-based and EHR-based algorithms included combinations of ICD diagnosis codes with visit, procedure and pharmacy codes.
The positive predictive values of algorithms that used only visit-linked ICD diagnosis codes were low and varied significantly, ranging from 3.3% to 66.7%. Similarly, in the study that used multiple algorithms, the algorithms using visit-linked ICD diagnosis codes only had a positive predictive value of 9.3% in the development cohort and 15.8% in the validation cohort.
Claims-based algorithms, which paired ICD codes with prescriptions for two or more classes of PAH-specific therapy, performed better, with a positive predictive value of 66.9% and specificity of 98.6%. The EHR-based algorithm, which combined visit-linked ICD codes, EHR encounter diagnoses, the performance of both an echocardiogram and a right heart catheterization, and a prescription for a PAH-specific therapy, had the highest positive predictive value (69.4%) and specificity (96.9%). However, sensitivity was lower (67.4%).
In an accompanying editorial, C. Gregory Elliott, MD, MACP, FCCP, medical director of the Pulmonary Hypertension Center at Intermountain Medical Center, highlighted the complexities surrounding an accurate diagnosis of PAH, such as its nonspecific symptoms and the presence of related comorbidities. Further, he wrote, the ICD-10 is inconsistent with WHO diagnostic classification of PAH, which is a challenge to the use of real-world data in creating diagnostic algorithms.
“In 2019, investigators who would use real-world data to study PAH are at a crossroads. They have electronic health records and large administrative databases, with attendant opportunities for population-level studies of PAH. However, thanks to Gillmeyer et al, they also have clear evidence that most previous studies using administrative databases failed to diagnose PAH accurately,” Elliott wrote.
“The path forward is clear: Algorithms for an accurate diagnosis of PAH need to be derived and validated in administrative databases. Validated algorithms for the diagnosis of PAH must become the standard for future studies of PAH that use real-world data. Finally, ICD codes must be revised to support accurate coding of [pulmonary hypertension] diagnostic groups based on broad-based consensus diagnostic definitions. Only then can real-world data provide new insights into PAH.” – by Melissa Foster
Disclosures: One author reports she receives research grant support from Actelion, Arena, Bayer, Incyte and Reata and is a member of the acute chest syndrome adjudication committee for the phase 3 study of rivipansel for treatment of vaso-occlusive pain in sickle cell disease (Pfizer). All other authors report no relevant financial disclosures. Elliott reports he has been a consultant for Actelion, Bayer and Bellerophon Therapeutics and has received grant or research support from Actelion, Gilead Sciences, NIH/NHLBI, Intermountain Research and Medical Foundation United Therapeutics Corp