Artificial intelligence helps take decisions about bone marrow transplant

Bone marrow transplants, fundamental in the treatment of diseases such as leukemia, are a medical milestone with more than six decades of history. Although effective, their challenge lies in the high associated mortality. In this context, detailed analysis of each case is crucial before making the decision to carry out these interventions.

Now, a team of researchers from IDIBELL and the Catalan Institute of Oncology, led by Dr. Alberto Mussetti together with biostatistician Blanca Rius-Sansalvador, has taken a significant step. Using artificial intelligence algorithms, they analyzed data from 33,927 oncohematological patients undergoing transplants between 2010 and 2019 with the aim of developing a new prognostic score.

The study, published today in the journal Bone Marrow Transplantation and which considered both transplant-related mortality and overall survival, achieved effective risk stratification. In this sense, the scoring system can now estimate the probability of death even in cases of transplants with haploidentical donors (match between 50% and 70%), where predictions were not previously possible. According to this new tool, in the most favorable cases, two-year survival reaches a remarkable 87%, contrasting drastically with the least favorable scenario, where unfortunately only 7% survival is recorded. However, although the use of artificial intelligence allows personalized risk stratification, it has not yet been possible to significantly improve the accuracy in predicting pre-transplant toxicity.

For these reasons, Dr. Mussetti emphasizes that “the results of a prognostic score alone should not be a reason to exclude patients from a potentially life-saving procedure. Predicting the future is never safe, especially when the system is complex as in this case.” And he adds that “the patient’s biological and clinical markers must also be considered in these crucial decisions. Therefore, the last decision is in the hands of the clinician, although it may be supported by more modern tools such as these based on artificial intelligence.”