Dr Orgueira on the Use of Machine Learning to Predict Survival Outcomes After Allo-HSCT in Myelofibrosis

"We used the advantage of a quantitative score like this machine learning tool to optimally find a threshold that can actually identify high-risk patients. We identified the optimal threshold [as the] 25% upper percentile of risk…that more than doubles the [number] of high-risk patients [identified using] other tools, and it is better aligned with what we see in the clinics."

Adrián Mosquera Orgueira, MD, PhD, a specialist in Hematology and Hemotherapy and lead researcher of the Computational and Genomic Hematology Group at the Health Research Institute of Santiago de Compostela, discussed results from a study investigating the accuracy of a prognostic model for overall survival (OS) that uses machine learning techniques to identify high-risk patients with myelofibrosis with poor outcomes following allogeneic hematopoetic stem cell transplant (allo-HSCT).

The data showed that a machine learning–based approach for predicting OS outcomes after allo-HSCT using random survival forest (RSF) demonstrated notable advantages over traditional methods, Orgueira reported. Specifically, the RSF model achieved slightly superior predictive performance compared with other tools, including Cox regression, he clarified. One key advantage of RSF was its generalizability, attributable to the use of cross-validation during training, which mitigates the risk of overfitting—a common limitation in machine learning models, he noted.

Although the improvement in accuracy and concordance index with the RSF model over conventional strategies was modest, the RSF model outperformed the previously established CIBMTR risk score, Orgueira detailed. The major differentiator, however, was the integration of clinical expertise into the interpretation and application of the machine learning outputs, he said. This enabled the identification of a clinically actionable threshold to classify high-risk patients—those for whom allogeneic stem cell transplantation may carry a significantly higher risk of poor outcomes, he explained.

Whereas the CIBMTR score and the Cox-derived score identified 8.2% and 10.1% of patients as high risk, respectively, the RSF model—through optimal threshold selection—classified 25% of patients as high risk, Orgueira continued. This revised threshold more accurately reflects the proportion of high-risk cases encountered in clinical practice and thus offers a more practical tool for patient stratification, he expanded. By expanding the identification of high-risk patients, this approach could inform transplant eligibility decisions and the need for intensified supportive care or alternative therapeutic strategies in this vulnerable subgroup, Orgueira concluded.