Machine Learning Derived Three-Parameter Prognostic Model for Survival in Patients with BPDCN

Introduction

Blastic plasmacytoid dendritic cell neoplasm (BPDCN) is a rare, aggressive hematologic neoplasm. The optimal first-line therapy in BPDCN is unknown. Therapeutic options include tagraxofusp, the first and only drug specifically approved for BPDCN, as well as regimens developed for acute myeloid leukemia (AML), acute lymphoblastic leukemia (ALL) and lymphoma. Furthermore, a prognostic model for overall survival (OS) does not exist. In this study, we compare patient outcome by first treatment, and employ a machine learning algorithm to generate a prognostic model that incorporates clinicopathological, molecular and therapeutic characteristics.

Methods

We included clinical, pathological, molecular and therapeutic data in patients with biopsy-confirmed BPDCN seen at Dana-Farber Cancer Institute between 2006-2022. We defined overt bone marrow (BM) involvement as ≥5% BPDCN cells in the marrow (previously shown to be prognostic by our group, Shimony et al, Blood Advances 2024). Next generation sequencing was performed on blood/BM at diagnosis. The primary outcome was OS, calculated from first treatment until death or last follow-up. We estimated OS across treatment groups using the Kaplan-Meier method, and differences in OS curves were assessed using log-rank tests. The supervised machine learning algorithm recursive partitioning was used to identify prognostic factors associated with OS.

Results

Among 66 patients included in our cohort, the median age in was 68 years (interquartile range [IQR] 61-76), the majority (n= 57, 86%) were male, with an ECOG performance status of 0 (66%). BM involvement was identified in 50 (76%), 39 (59%) had overt BM disease, most had normal karyotype (68%) and TET2 was the most common mutation (65%).

First treatment included either tagraxofusp (n=30, 46%) or an ALL regimen, AML regimen or other treatment (n=12 [18%] each). Patients who were treated with AML and ALL therapies were younger (median age 62 and 63 years, respectively) compared with those who were treated with tagraxofusp (median age 70) or other therapies (median age 78).

Complete remission (CR) after first treatment was documented in 35 (53%) patients and 20/35 (57%) were consolidated in CR1 with allogeneic stem-cell transplantation (SCT). The CR1 rates were comparable between ALL and AML-based therapy groups (n=9 each, 75%) and lower in the tagraxofusp group (n=10, 37%), although 10 (37%) additional patients in the tagraxofusp group achieved PR. The rates of SCT were similar between all treatment groups (p=0.3). The median OS was 17.7 months (95% confidence interval [CI] 12.3-27.2) with tagraxofusp, 23.3 months (95% CI 17.5-NA) with ALL therapy, and 10.9 months (95% CI 6.9-18.1) with AML therapy. In pairwise comparisons, there was no statistical difference in age-adjusted OS between tagraxofusp and either ALL-based (p=0.36) or AML-based (p=0.21) therapies.

In the recursive partitioning model, 3 variables emerged as prognostic for OS in a hierarchal approach. Age was the most influential predictor, followed by presence of overt BM disease and absence of activated signaling mutations (NRAS, KRAS, FLT3, CBL, CKIT, SH2B3). These results translated into 3 distinct prognostic groups. Favorable risk: those age < 50 had the most favorable outcome (1- and 2-year OS of 100%). Intermediate risk: defined as either a) patients age ≥ 50 without overt BM disease or b) age ≥ 50 with overt BM disease but without any signaling mutations (1- and 2-year OS of 75% [95% CI 61-90%] and 39% [95% CI 20-58%] respectively, p-value vs. favorable <0.001). Adverse risk: age ≥ 50 and not meeting criteria for intermediate risk (1- and 2-year OS of 38% [95% CI 16-60%] and 5% [95% CI 0%-16%], respectively, p-value vs. favorable <0.001).

Conclusions

In a cohort of 66 patients with BPDCN, we did not find any difference in OS between treatment modalities. We generated a prognostic model for patients with BPDCN based on age, bone marrow involvement and presence of signaling mutations, with 3 risk groups - favorable (age < 50), intermediate (age ≥ 50 without overt BM disease or without activated signaling mutations) and adverse risk (age ≥ 50, not meeting criteria for intermediate risk). This new BPDCN prognostic scoring system may be integrated into future research including clinical trials.

Luskin:Novartis: Honoraria, Research Funding; AbbVie: Research Funding; Jazz: Honoraria; KITE: Honoraria; Pfizer: Honoraria. Renosi:Incyte: Other: speaker for an event, participating to another; Servier: Other: congress registration. Deconinck:ImmunoGen: Research Funding; Menarini-Stemline: Other: consultancy, Research Funding; Abbvie: Research Funding; Novartis: Research Funding. Neuberg:Madrigal Pharmaceutical: Current equity holder in publicly-traded company. Lane:Menarini Group: Other: Steering Committee, Research Funding; IDRx: Consultancy; AbbVie: Research Funding; Cimeio Therapeutics: Consultancy; Jnana Therapeutics: Consultancy; ProteinQure: Consultancy; Qiagen: Consultancy; Stelexis BioSciences: Consultancy; Medzown: Current equity holder in private company.