A registry analysis of 3,302 adults receiving CD19 CAR-T reported that machine-learning models using information available by Day 14 can identify a low-risk subgroup with a 3.1% subsequent infection rate. The approach relies on early post-infusion signals to stratify infection risk rather than waiting for clinical deterioration. The dataset includes calibration and limitations that the authors flagged for deployment readiness and cross-disease generalization. Still, the Day-14 framing aligns with operational hospital workflows that need actionable infection-control decisions early. For CAR-T programs, the finding supports the growing push toward risk-adapted monitoring and prophylaxis as more centers scale cell therapy delivery.