Introduction: Allogeneic stem-cell transplantation (HSCT) is regarded as a potentially curative approach for treating acute lymphoblastic leukemia (ALL) in adults. Unfortunately, part of the anti-leukemic benefit of HSCT is offset by its toxicity. Prior research from developing countries has highlighted that early mortality (EM) plays a significant role in their comparatively worse outcomes. Therefore, it is crucial to identify the factors involved in EM after HSCT for ALL in adults, aiming to better select eligible cases and allocate resources effectively in this setting. Considering this, this study aims to develop a multivariable prediction model for EM.

Methods: Data were extracted from a retrospective multi-site registry study in Brazil. Our study conforms to the TRIPOD reporting guideline. Patients with ALL over 16 years old who underwent a first HSCT at five Brazilian centers between January 2007 and December 2017 were included. Consecutive patients allografted at a reference center in Monterey, Mexico, were used as an external validation cohort. EM was defined as death within 100 days of HSCT. A multivariable logistic regression analysis was performed, with Akaike's Information Criteria (AIC) used for model selection. A p-value <0.05 was considered statistically significant. All numerical variables were dichotomized using the function “cutpointr.” At each step, the model's discrimination was assessed by the C statistic. Internal validation was performed using bootstrapping, and external validation was conducted with the cohort from Mexico. To obtain a robust estimate of the AUC, we applied a bootstrapping procedure with 1,000 resamples.

Results: Overall, 275 patients were included in this study. Baseline characteristics of this registry were previously reported (Silva W et al., 2022). Most patients were in the first complete remission (CR1) (66%). Donor types included matched sibling donor (MSD) (53%), matched unrelated donor (MUD) (19%), mismatched unrelated donor (MMUD) (9%), haploidentical donor (19%), and umbilical cord donor (5%). The EM rate in the derivation cohort was 24.4%. Several baseline factors related to patients, disease, and HSCT were included in the original model, which was refined using a backward selection approach based on AIC. The final model included gender, Ph-status, age, and donor age as predictors for EM, with odds ratios for mortality ranging from 1.9 to 3.1. By assigning one point for each positive factor (male sex, Philadelphia-positive leukemia, age >20 years, and donor age ≥38 years), an EM score was created. This score divided the cohort into three categories: low [0-1] (n=46), intermediate [2] (n=85), and high [3-4] (n=65). EM rates for these groups were 6.5%, 14.1%, and 35.4%, respectively. The bootstrapped mean AUC of the logistic regression model was 0.7 (95% confidence interval 0.612-0.775). The validation cohort (Mexican cohort, n=100) comprised patients who underwent allogeneic HSCT for ALL between 2015 and 2023, with a median age of 23 years (range, 15-62). This cohort had some distinct characteristics: haploidentical donors were the main source (78%), and myeloablative total body irradiation was used in a small subset of patients (7%), and most patients were transplanted in second or later remission (84%). The EM rate in the Mexican cohort was 9%. In this cohort, EM rates were 7.1%, 19%, and 5% in low (n=42), intermediate (n=21), and high-risk (n=20) groups, respectively. The Brazilian and Mexican registries are being extended, and a newer analysis with more subjects is still pending.

Conclusions: Reducing EM rates is pivotal to improving HSCT outcomes in patients with ALL, especially in less resourced settings where newer therapies are not widely available. Notably, disease- and patient-related factors such as sex, age, donor age, and Philadelphia-positive status were found to be associated with EM in our predictive model. This model demonstrated acceptable discrimination ability, although it did not successfully distinguish EM in the validation cohort. Marked differences between the two cohorts may have influenced our findings. We believe that expanding the database and reinforcing cooperative strategies in HSCT will enhance the model's ability and make it more generalizable. This EM score might be helpful in improving donor selection and conditioning strategies for high-risk patients.

Disclosures

Silva:Amgen: Consultancy, Speakers Bureau; Libbs: Research Funding; Pfizer: Speakers Bureau; Abbvie: Speakers Bureau. Gomez-De Leon:Abbvie: Honoraria; Amgen: Honoraria; bms: Honoraria; Novartis: Honoraria; Pfizer: Honoraria; Janssen: Honoraria; Sanofi: Honoraria, Other: Advisory board; Janssen: Other: Advisory board. Rocha:AbbVie: Consultancy; Takeda: Consultancy; Pfizer: Consultancy; Astellas: Consultancy; Kite: Consultancy; Amgen: Consultancy.

This content is only available as a PDF.
2024
Sign in via your Institution