The model correctly predicted COPD-PH in 93.5% of patients analyzed
Global biopharma algorithm to detect COPD-PH patients
SUMMARY
To enable recruitment for a Phase III trial, we developed an algorithm to detect undiagnosed patients with COPD-PH from generally available clinical data
Scientific Question
The client had launched a clinical trial at >50 sites to test a novel therapy for the treatment of COPD induced Pulmonary Hypertensions (COPD-PH).
Given lack of treatment options for this population, recruitment of qualifying patients proved challenging, as there was no existing way for each center to identify these patients.
The client asked TriAxia to develop an algorithm for detecting COPD-PH patients using generally available clinical data from a hospital EMR.
Given lack of treatment options for this population, recruitment of qualifying patients proved challenging, as there was no existing way for each center to identify these patients.
The client asked TriAxia to develop an algorithm for detecting COPD-PH patients using generally available clinical data from a hospital EMR.
Evidence Triaxia Provided
After developing a protocol with the client, TriAxia extracted a cross-center cohort of both COPD and PH patients as well as patients without any evidence of either disease.
We then used previously published criteria and models as well as machine learning techniques to build an algorithm to screen for COPD-PH patients in a general population, training and testing the model on different sets of hospitals from the TriAxia database.
Results
TriAxia was able to build a model with a combined Area Under Curve (AUC) of up to 0.879, a measure of the classification power the model is in classifying COPD-PH patients using generally available clinical data. The model exceeded the clients’ threshold for classification power to use at sites to support clinical trial recruitment.