Machine Learning Based Prediction of Veno Occlusive Disease Following Allogenic Stem Cell Transplantation

Veno-occlusive disease/sinusoidal obstruction syndrome (VOD/SOS) is a potentially life-threatening complication of stem cell transplantation. Hitherto, patient snap shots have been focussed upon, while predicting the onset and severity of VOD/SOS. We present a novel mechanistic learning of the score of VOD/SOS onset and progression using recorded values of pre-transplant parameters and during-transplant observations on 25 patients, as part of a longitudinal data comprising data stored in patient records at multiple institutes. Subsequently, we learn the relationship between such pre-transplant variables and this learnt score, using Machine Learning (ML) techniques that permit the desired avoidance of imposed forms of this relationship. This allows reliable prediction of the VOD/SOS score of a new patient with observed pre-transplant parameters. We discuss results of such undertaking. Using the learnt relationship between VOD/SOS score and patient pre-transplant attributes, we also identify relative potency of each pre-transplant variable, towards VOD/SOS onset and progression. Our ML-based learning and prediction offer protocols that enable early identification of VOD susceptibility in a new patient at the pre-transplant stage, thus permitting intervention, enabling VOD/SOS mitigation.

No relevant conflicts of interest to declare.