A retrospective registry analysis of 3,302 adults with large B-cell lymphoma identified a machine-learning infection risk subgroup using information available by Day 14 after CD19 CAR-T. The model flagged a low-score group with a 3.1% subsequent infection rate, though authors emphasized calibration and cross-disease generalization challenges for real-world deployment. The work centers on improving post–CAR-T infection surveillance by enabling earlier risk stratification using routine, early timepoint clinical inputs rather than waiting for events. For CAR-T programs, the contribution is a step toward operationalizing predictive analytics—paired with the caveat that external validation and implementation constraints remain key before clinical use.