A retrospective registry analysis of 3,302 adults with large B-cell lymphoma reported that machine-learning models using information available by Day 14 after CD19 CAR-T therapy can identify a subgroup at low risk for subsequent infection. The study described a low-score category with a 3.1% subsequent infection rate. The authors emphasized limitations, including calibration concerns and reduced generalizability across disease contexts, signaling that clinical deployment would require additional validation beyond the retrospective dataset. The modeling approach reflects the broader move toward operational risk stratification in post–CAR-T monitoring. For clinicians and developers, the pivotal question is whether Day 14 risk prediction improves allocation of infection surveillance, prophylaxis intensity, and early intervention without increasing false reassurance.