An Antibiotic-Based Scoring Model for Predicting Early and Late Hematologic Toxicity Following CD19 CAR-T Therapy in B-ALL Patients

Background: Immune effector cell-associated hematotoxicity (ICAHT), including neutropenia, thrombocytopenia, and anemia, are the most common cytopenia encountered in CAR-T cell therapy. However, traditional CAR-HEMATOTOX model has high specificity but low sensitivity. Disruption of the gut microbiome by antibiotics has been shown to impact both clinical outcomes and CAR-mediated toxicities, including CRS and ICANS. To clarify its impacts on CAR-mediated cytopenia, we develop and validate an antibiotic-based scoring model for predicting early and late hematologic toxicity following CD19 CAR-T therapy in B-ALL patients.

Methods: 93 B-ALL patients receiving CD19 CAR-T cell therapy at Wuhan Union Hospital were enrolled. We collected patients' prior antibiotic exposure, as well as baseline demographic features, prior therapy, tumor burden, inflammatory states, and hematopoietic reserve. The cohort was divided into training and validation cohorts with a 6:4 ratio. A univariate analysis was conducted to find the significant factors with p-value <0.05. To reduce the number of predictors, only those with an area under the curve (AUC) >0.6 and a p-value <0.1 were retained. Multivariate logistic regression with step wise backward elimination was used to identify the most influential predictors. Two models with simple weights (Model 1) or additional weights (Model 2) were developed and compared. Grading of cytopenia was evaluated based on the 2023 EHA/EBMT consensus guideline and CTCAE 5.0.

Results: 64.5% patients had received at least one antibiotic prior to CAR-T cells therapy. High-risk antibiotics (HR ABX) were identified as the most used ones, including meropenem, teicoplanin, piperacillin, and ceftazidime. All other antibiotics were defined as ‘low-risk (LR ABX)‘. Exposure to HR ABX was significantly associated with severe hematotoxicity (P=0.006). After univariate and AUC analysis, prior HR ABX use, baseline bone marrow (BM) tumor burden, platelet, hemoglobin, LDH, CRP, and ferritin were identified as risk factors. Multivariate logistic regression identified baseline LDH levels and BM burden and additional weights were assigned.

Both models showed robust predictive capabilities, with AUCs 0.795 (model 1) vs. 0.779 (model 2) and odds ratios 17.926 (model 1) vs. 15.400 (Model 2). We then validate two models alongside the CAR-HEMATOTOX model using validation cohort. Model 1 showed a high specificity [80% (95% CI:55.7-93.4), vs. model 2, 65% (95% CI: 40.9-83.7), and CAR-HEMATOTOX model, 95% (95% CI: 73.1-99.7)] and sensitivity [76.5% (95% CI: 49.8-92.2), vs. model 2, 82.4% (95% CI: 55.8-95.3), and CAR-HEMATOTOX model, 35.3% (95% CI: 15.3-61.4)]. Consequently, Model 1 was selected as our final scoring model. The cutoff value for Model 1 was set as follows (0 point): baseline platelet >75×10⁹/L, hemoglobin >79.5g/L, CRP <5.54mg/L, ferritin <1065.9μg/L, LDH <248.5U/L, BM burden <71.5%, and no HR ABX usage. Low-risk group was defined as 0-2 points while high-risk group was ≥ 3 points.

We then evaluation of the prolonged severe hematologic toxicities. Within the first six months of follow-up, severe ICAHT was observed in 0-22% of low-risk patients, versus 17%-44% in the high-risk group. Additionally, severe thrombocytopenia occurred in 10%-24% of the low-risk group, compared to 21%-56% in the high-risk group. Moreover, severe anemia was noted in 8%-21% of low-risk patients during the first six-month follow-up, compared to 22%-64% in the high-risk group.

Finally, the survival analysis showed the median OS was 28 months in the low-risk group and 6 months in the high-risk group (P=0.0004). Similarly, the median PFS was significantly shorter at 4 months in the high-risk group compared to 11 months in the low-risk group (P=0.0015).

Conclusion: Through univariate and multivariate analyses, factors including baseline platelet, hemoglobin, LDH, CRP, ferritin levels, BM burden, and prior HR ABX use, are identified to influence hematopoietic recovery. Subsequently, two scoring model were developed, validated, and compared. A high score represents a greater risk of severe cytopenia, either occurring early (before day 28) or prolonged (6-months follow-up). Finally, patients with high scores in the model exhibited notably poorer clinical outcomes.

No relevant conflicts of interest to declare.