Longitudinal Clinical Data Improves Survival Prediction after Hematopoietic Cell Transplantation Using Machine Learning.

Serial prognostic evaluation of patients after allogeneic hematopoietic cell transplantation (alloHCT) might help identify patients at high risk of developing potentially lethal organ dysfunction. Current prediction algorithms are based on models that do not incorporate changes to the patients' clinical condition that occur after alloHCT in the model development, which limits their predictive ability. We developed and validated a robust risk-prediction algorithm to predict short-term and long-term survival after alloHCT in pediatric patients that includes baseline biological variables, as well as changes in the patients' clinical status after alloHCT. The model was developed using clinical data from children and young adults treated at a single academic quaternary-care referral center. The model was created using a randomly split training dataset (70% of the cohort), internally validated (remaining 30% of the cohort from the same center), and then externally validated on patient data from another tertiary-care referral center. Repeated clinical measurements performed from 30 days before alloHCT to 30 days afterwards were extracted from the electronic medical record and incorporated into the model to predict survival at 100 days, 1-year, and 2-years after alloHCT. Of the 738 patients who underwent their first alloHCT at our institution between 2000 and 2020, 517 (70%) were randomly included in the training dataset and 221 (30%) constituted the validation dataset. When compared with models constructed from baseline variables alone, the naïve-Bayes machine learning models incorporating longitudinal data were significantly better at predicting whether patients would be alive or deceased at the given timepoints. This proof-of-concept study demonstrates that unlike traditional prognostic tools that use fixed variables for risk assessment, incorporating dynamic variability using clinical and laboratory data improves the prediction of mortality in patients undergoing alloHCT.