Machine learning helps predict risks during myeloma stem cell treatment

Multiple myeloma is a cancer in which plasma cells, which normally produce antibodies, grow uncontrollably in the bone marrow. There is currently no cure. However, various treatments can stabilize the disease and reduce symptoms. One such treatment is using the patient’s own stem cells. In many cases, hospitalization for several weeks is required. The research team used machine learning techniques to assess the conditions under which some treatments could be safely performed as an outpatient. The study was carried out by researchers from the University of Göttingen’s Institute for Biological Network Dynamics (CIDBN), the Göttingen University Medical Center (UMG) and the University Medical Center Bielefeld (OWL). Published in a magazine npj digital medicine.

The treatment, known as “autologous stem cell transplant,” involves harvesting stem cells from the patient’s blood. To prepare, patients receive chemotherapy and then go through a step in which the stem cells are flushed from the bone marrow into the blood so they can be harvested. During the mobilization phase, patients are often hospitalized for two to three weeks so that any serious side effects, such as kidney failure or infection, can be treated immediately. However, in some patients, serious side effects may develop very late or not at all. “We asked ourselves whether long-term hospital stays were really necessary for all patients,” explains Dr. Enver Aydilek, who works at UMG and OWL.

Researchers evaluated treatment data for 109 patients with multiple myeloma who underwent stem cell mobilization at UMG. Using machine learning techniques, we were able to identify a time window in which most patients are least likely to experience serious side effects. The next step was to use the data to develop a model that predicted as accurately as possible which side effects would occur in which individuals and when.

For certain types of side effects, the model made very accurate predictions, allowing for more accurate risk assessments. “With our data-driven treatment roadmap, we can now better assess who requires inpatient monitoring and who can safely undergo chemotherapy and the mobilization phase with close monitoring and care in the outpatient setting,” explains lead author Friedrich Schwarz, a researcher in medicine and data science at UMG and the University of Göttingen and a researcher at CIDBN and the UMG Hematology and Medical Oncology Clinic. This means that this treatment is more patient-friendly and modern.

The research team said simulations of different treatment pathways showed that outpatient treatment can have a dual effect. On the other hand, patients have a better quality of life in the familiar home environment, making treatment more personalized and less stressful. At the same time, clinics can plan in a resource-efficient manner. “Of course, for effective interaction between outpatient and inpatient treatment, the right conditions need to be created,” emphasizes Schwartz. This study places the first building blocks for this.