Background: Ciltacabtagene autoleucel (cilta-cel) achieves high rates of durable responses in multiple myeloma (MM) but has been associated with MNTs, including Parkinsonism, cranial nerve palsies, and Guillain-Barré syndrome (GBS). Previous data suggested that the peak absolute lymphocyte count (ALC) – a surrogate for peak in vivo CAR T-cell expansion – may be used to prognosticate the risk of MNTs. ALC thresholds of 3000/µL (Lim et al, EHA 2025) and 5000/µL (Turner et al, Tandem 2025) are tentatively being adopted for pre-emptive treatment with dexamethasone. However, these thresholds have not been validated independently. Using advanced modeling approaches and validation benchmarks, we independently assessed the value of these ALC thresholds, and investigated other modeling strategies to improve MNT prognostication after cilta-cel.

Methods: We included 88 consecutive patients with MM who received standard-of-care cilta-cel at our institution. We applied univariate Firth-penalized logistic regression using restricted cubic splines to enable non-linear relationships between ALC values measured at select timepoints, based on point-of-care availability (D) post-CAR T-cell infusion, and the development of MNTs. We computed the sensitivity (Se; true positive rate), specificity (Sp; true negative rate), and positive (PPV) and negative predictive values (NPV), and net benefit using decision curve analysis. Because the risks of a short course of dexamethasone are typically low relative to the risk of a missed opportunity to prevent MNT, we defined as “optimal” ALC thresholds maximizing Se and NPV (i.e., reducing the risk of false negatives).

Results: Of 88 patients, 17% (n=15) developed MNTs (facial palsy, n=10; GBS, n=1; Parkinsonism, n=3; facial palsy & GBS variant, n=1). Three patients (3%) received pre-emptive dexamethasone at the physician's discretion based on ALC kinetics. Median ALC was significantly higher in the MNT cohort versus the non-MNT cohort (D+10: 3,230 vs 850; p=0.043; D+14, 1,590 vs 750, p= 0.047; D+21 ALC, 845 vs 545; p=0.051). The median time to peak ALC post-infusion was D+11 (IQR = 10–13).

At D+10, an ALC threshold of 3,000/µL yielded a Se, Sp, NPV, and PPV of 0.57, 0.82, 0.9, and 0.4, respectively (60% of patients falsely flagged and 10% missed); an ALC threshold of 5,000/µL yielded a Se, Sp, NPV, and PPV of 0.29, 0.91, 0.86, and 0.4, respectively (60% of patients falsely flagged and 14% missed). At D+21, an ALC threshold of 3,000/µL yielded a Se, Sp, NPV, and PPV of 0.25, 0.96, 0.88, and 0.5, respectively (50% of patients falsely flagged and 12% missed); an ALC threshold of 5,000/µL yielded a Se, Sp, NPV, and PPV of 0.13, 0.98, 0.87, and 0.5, respectively (50% of patients falsely flagged and 13% missed).

Given the relatively high percentage of inaccurate predictions using the above ALC thresholds, we further evaluated multivariable models including the following routine laboratory analytes in addition to the ALC: C-reactive protein (CRP), platelet count, and fibrinogen. D+10 (p=0.015 [non-linear]) ALC and D+21 ALC (p=0.027 [log-linear]; p=0.45 [non-linear]) remained associated with higher odds of MNT. Using a probability threshold maximizing the Se and NPV, our D+10 and D+21 multivariable models yielded a Se and NPV of 100%. Our D+21 model achieved the highest Sp and PPV (84%, 54.5%). Decision curve analysis confirmed a higher net benefit of our multivariable model across all probability thresholds compared to a model including ALC alone and a “treat-all” approach.

Conclusion: We independently evaluated the 3000/µL and 5000/µL ALC thresholds routinely in use in patients with MM undergoing treatment with cilta-cel per standard-of-care. Both thresholds have modest prognostic value to predict MNTs after cilta-cel, specifically, with a 10-14% risk of generating a false negative prediction. In contrast, our prognostic ALC model incorporating CRP, fibrinogen, and platelet count at D+10 and D+21, significantly improved MNT prediction (Se/NPV=100% ). Our findings suggest that ALC alone may be insufficient for accurate MNT prognostication. Further validation of our model in larger datasets is ongoing, as is research to validate ALC-based thresholds for preemptive treatment with dexamethasone.

This content is only available as a PDF.
2025
Sign in via your Institution