Prediction of resistance to hydroxyurea therapy in patients with polycythemia vera: a machine learning study (PV-AIM) validated in a prospective interventional phase IV trial (HU-F-AIM)

Abstract

Polycythemia vera (PV) is a myeloproliferative neoplasm associated with increased thromboembolic (TE) risk and hematologic complications. Hydroxyurea (HU) serves as the most frequently used first-line cytoreductive therapy worldwide; however, resistance to HU (HU-RES) develops in a significant subset of patients, leading to increased morbidity and necessitating alternative treatments. This study, part of the PV-AIM project, employed machine learning techniques on real-world evidence (RWE) from the Optum® EHR database containing 82.960 PV patients to identify baseline predictors of HU-RES within the first 6–9 months of therapy. Using a Random Forest model, the study analyzed data from 1850 patients, focusing on laboratory parameters and clinical characteristics. Key predictive markers included red cell distribution width (RDW) and hemoglobin (HGB), showing the strongest association with HU-RES. A synergistic interaction between RDW and HGB was identified, enabling TE risk stratification. This study provides a robust framework for early detection of HU-RES using readily available clinical data, facilitating timely intervention. These findings underscore the importance of personalized treatment approaches in managing PV and highlight the utility of machine learning in enhancing predictive accuracy and clinical outcomes. Based on the results of PV-AIM we initiated an open-label, prospective, single-arm, interventional, phase IV study (HU-F-AIM) evaluating HU-resistance/intolerance. Validation of predictive biomarkers may facilitate identification of patients at risk of HU resistance who may benefit from alternative treatment options, possibly preventing ongoing phlebotomy during HU treatment, a frequent therapeutic choice in high-risk PV associated with early disease progression and increased thromboembolic complications. We propose an updated terminology that differentiates between true molecular resistance and clinical resistance, that may indicate the requirement for alternative therapeutic strategies.