Identifying the optimal conditioning intensity for stem cell transplantation in patients with myelodysplastic syndrome: a machine learning analysis

A conditioning regimen is an essential prerequisite of allogeneic hematopoietic stem cell transplantation for patients with myelodysplastic syndrome (MDS). However, the optimal conditioning intensity for a patient may be difficult to establish. This study aimed to identify optimal conditioning intensity (reduced-intensity conditioning regimen [RIC] or myeloablative conditioning regimen [MAC]) for patients with MDS. Overall, 2567 patients with MDS who received their first HCT between 2009 and 2019 were retrospectively analyzed. They were divided into a training cohort and a validation cohort. Using a machine learning-based model, we developed a benefit score for RIC in the training cohort. The validation cohort was divided into a high-score and a low-score group, based on the median benefit score. The endpoint was progression-free survival (PFS). The benefit score for RIC was developed from nine baseline variables in the training cohort. In the validation cohort, the hazard ratios of the PFS in the RIC group compared to the MAC group were 0.65 (95% confidence interval [CI]: 0.48–0.90, P = 0.009) in the high-score group and 1.36 (95% CI: 1.06–1.75, P = 0.017) in the low-score group (P for interaction < 0.001). Machine-learning-based scoring can be useful for the identification of optimal conditioning regimens for patients with MDS.