Patient characteristics
Overall, 58 patients with MM were included in this retrospective study and were divided into three groups according to the recovery of platelet: stable (recovery time < 0 days, n = 14), non-delayed platelet recovery (recovery time ≤ 28 days, n = 23) and delayed platelet recovery (recovery time > 28 days, n = 21), as detailed in Table 1.
Next, we analyzed the change in thrombocyte levels during the first 8 weeks post-infusion (Fig. 2). The aggregated time course of CAR T-associated thrombocytopenia followed a curve with decrease after lymphodepletion (day − 5) and a recovery around month 2. There was no crossover between the curves of these three groups. The majority in delayed platelet recovery group had prolonged thrombocytopenia and barely recovered. The stable group never decreased to 50×109/L after CAR-T infusion and recovered in week 2. The platelet level of non-delayed platelet recovery group was characterized by an initial drop after lymphodepletion and a nadir in week 2, and the most of them achieved normalization of platelet counts (100×109 /L) by the second month.
The Differences Between Groups
The differences in the clinical factors between the delayed platelet recovery group, non-delayed platelet recovery group, and stable group were analyzed (Fig. 3A-H). Inflammation-related factors (such as CRS grade, IL-6, and ferritin) and hematopoietic reserve (baseline platelet count and baseline hemoglobin) were linked to delayed platelet recovery. Notably, although there was a significant difference in both the logarithm of peak and foldchange values of IL-6 (Fig. 3A-B) and ferritin (Fig. 3C-D) among those groups, the difference in the logarithm of baseline IL-6 and ferritin between groups were not strong (Additional file 4: Figure S1). And the proportion of patients who had autologous stem cell transplantation (ASCT) before CAR-T were different between stable group and delayed platelet recovery group (Fig. 3G). Other factors that did not differ significantly in the distribution among groups could be found in the Additional file 4: Figure S1.
Factors Influencing Platelet Levels After Infusion
To explore the factors that might affect platelet recovery after CAR-T infusion, recovery time was used as an indicator and then analyzed the association between recovery time and patient characteristics (Fig. 4A-G) (Additional file 5: Figure S2). A strong association between hematopoietic reserve before lymphodepletion and the prolongation of recovery time could be seen in platelet counts (Fig. 4A, p < 0.0001, r = -0.59) and hemoglobin (Fig. 4B, p = 0.0005, r = -0.42). However, there was no significant correlation between white blood cells (WBC) or neutrophil count (NEU) and recovery time (Additional file 5: Figure S2 A-B). Notably, there was also a correlation between inflammation-related factors and recovery time, such as the logarithm of ferritin before CAR-T infusion (Fig. 4C, p = 0.0004, r = 0.45), the logarithm of peak ferritin (Fig. 4D, p = 0.0001, r = 0.49), as well as the logarithm of peak interleukin-6 (IL-6) (Fig. 4E, p = 0.0004, r = 0.45) after CAR-T infusion. CRS grade was also a related factor (Fig. 4F, p = 0.0040, r = 0.36). In addition, ASCT before CAR-T exhibited a positive correlation with recovery time (Fig. 4G, p = 0.0046, r = 0.38). Interestingly, age, gender, β2-microglobulin (β2m) and lactate dehydrogenase (LDH) were not associated with recovery time. (Additional file 5: Figure S2).
Development Of A Predictive Model Based On Platelet Levels After Car-t Infusion
We next aimed to develop a clinical score that discriminates between a high and a low risk for delayed platelet recovery. The patients were divided into two groups by recovery time ≤ 28 days vs recovery time > 28 days and ROC analysis was performed on them using the baseline indicators mentioned above (Fig. 5) (Additional file 6: Figure S3). Ultimately, the indicators included in the model were baseline platelet count, baseline hemoglobin, the logarithm of baseline ferritin, and cytokine release syndrome grade. Two test Recovery-Model were built and compared with each other.
In model 1 (Additional file 2: Table S2), we included the four variables in the model directly and assigned simple weights to them (either 0 or 1 point for each factor). The cut-off value was determined by optimizing the Youden Index (sensitivity + specificity − 1) for the outcome of recovery time ≤ 28 days vs recovery time > 28 days for each variable while specificity ≥ 0.75. Multivariate logistic regression analysis with stepwise backward elimination was performed to find out the most influential variable (Additional file 1: Table S1). This yielded baseline platelet count. In model 2 (Table 2), additional weights were assigned to baseline platelet count based on the results of multivariate logistic regression (0 vs. 1 vs. 2 points). Two cut-off values of baseline platelet count were determined by sensitivity ≥ 0.8 and specificity ≥ 0.85, respectively (Table 2). Finally, by comparing the AUC and odds ratio (OR) of the two models, we concluded that model 2 had a stronger association with platelet recovery after CAR-T infusion (Additional file 3: Table S3).
Model 2 (Recovery-Model) was next used to score the MM patients (n = 58), and we tested multiple thresholds for this model to differentiate between low and high risk for delayed platelet recovery after CAR-T infusion (Additional file 7: Table S4). The threshold was set at 1 based on the optimal Youden-Index.
Survival Analysis
In these patients, delayed platelet recovery and a high Recovery-Model score was associated with worse clinical outcomes. We first plotted the corresponding survival curves according to the recovery time of patients (Fig. 6A-B) and the overall survival differ significantly by delayed platelet recovery, non-delayed platelet recovery, and stable group (P = 0.0457). However, the impact of platelet recovery on progression-free survival did not differ between those groups (P = 0.1753). In addition, we plotted the corresponding survival curves according to the Recovery-Model score (Fig. 6C-D). The results were similar to those of the delayed platelet recovery grouping, with the Recovery-Model low-risk group differing from the high-risk group only in OS (P = 0.0204) but not in PFS (P = 0.7379).