Predicting infection risk following CAR-T using the bedside CAR-HEMATOTOX score: performance in an Australian cohort

• CAR HEMATOTOX performed best a predictor of late, severe infections in an Australian cohort. • Post-infusion immunomodulatory therapies and neutropenia were associated with increased infection risk in multivariable models. The CAR-HEMATOTOX (HT) predicts prolonged neutropenia after CAR-T therapy. Aggressive lymphoma patients with HT scores ≥ 2 (HT high ) have experienced higher rates of severe infections compared to patients with lower scores (HT low ). 1 We aimed to validate HT’s prediction capability in an independent aggressive lymphoma cohort, treated with standard-of-care CAR-T, between January 2019 and September 2023. HT scores were calculated according to original methods. Infections were defined as microbiologically-confirmedor clinically-diagnosed. Severe infections were classified as CTCAE grade ≥ 3. Logistic regression evaluated HT’s association with infections within 90 days , and separately for early (Day 0-30) and late (D30-90) infections. Model fit was evaluated by AUC. Among 109 patients infections occurred in 31 patients (54% grade ≥ 3). HT high status was not associated with any-grade infections across D0-90 (OR = 1.91, 95%CI: 0.82-4.42, p = 0.13), but was significantly associated with severe infections (OR = 8.67, 95%CI: 2.30-32.68, p = 0.001), with very good discriminative validity (AUC = 0.74). HT high status performed as a better predictor for late severe infections (AUC = 0.85), compared with early severe infections (AUC = 0.62). Multivariable logistic regression also identified tocilizumab (OR = 1.52, 95%CI: 1.02 – 2.38, p = 0.041), prolonged neutropenia (OR = 11.95, 95%CI: 1.64 – 210.75, p = 0.001), and prolonged steroid (OR = 38.58, 95%CI: 2.05 – 725.91, p < 0.01) as significantly associated with infection. The relationship between HT high and risk of severe infection was validated in our L-BCL cohort. HT high was an excellent discriminative predictor of late severe infections.