High-Density Data Mining to Identify Unique Toxicities of Adolescent/Young Adult Cancer Treatment: A Report from the Children's Oncology Group

Abstract

Background: Adolescent and young adult (AYA) cancer patients, 15-39 years of age, experience excess treatment-related toxicity compared to children. However, the relatively low number of AYAs enrolled on individual Children's Oncology Group (COG) clinical trials limits thorough toxicity analyses. Our objective was to delineate a detailed signature of drug- and dose-specific toxicities experienced by AYAs. We created a novel high-density data informatic method of clustering and analyzing toxicity data across multiple clinical trials that permits the comparison of outcomes in large populations of AYA versus younger cancer patients. In this pilot study, we aggregated four different COG acute lymphoblastic leukemia (ALL) trials to identify new and previously described toxicities that correlated with age and drug exposure, and validated our findings with previously published toxicity data as traditionally reported.Methods: Toxicity data from four recent COG trials for newly-diagnosed high-risk (CCG-1961, AALL0232) and standard-risk (CCG-1991, AALL0331) ALL trials were analyzed (n=13,436). First, disparate toxicity scoring systems were harmonized and re-mapped to a unified system based on the Common Terminology Criteria of Adverse Events (v.4.0), and then clustered in hierarchical fashion to enhance the power analysis. Differential rates of age-specific toxicities were determined in phases of treatment that were common to all protocols. AYA was defined as per NCI (15-39 y). Fisher's exact test was used to determine significant toxicity differences by treatment and age. Grade 3-5 events were included for all toxicities except osteonecrosis (ON), which included grade 2-5.Results: Across all phases of treatment, toxicity rates among AYAs vs children are summarized in Table 1. During Induction, hyperglycemia and hyperbilirubinemia were more common among AYAs (25.3% vs 8.1% and 10.4 vs 2.4%, respectively; both p < 0.0001). Metabolic disorders, including hypoalbuminemia, were also more common in AYA patients for both dexamethasone and prednisone in Induction (AYA: 8.7% vs 2.2%; children: 2.7% vs 1.5; both p < 0.05). Post-Induction, mucositis and peripheral neuropathy were more common among AYAs (15.5% vs. 4.5% and 6.1% vs 2.9%, respectively, both p<0.0001). ON appeared as a delayed manifestation during Maintenance in patients who received dexamethasone in Induction; patients (≥10 y) showed increased rates (21.4% (dex) vs 12% (pred), p< 0.0001). In addition, this large-scale cross-study analysis identified novel AYA toxicity signatures not previously described, including pancreatitis and thrombosis, which were more common in the AYAs compared to children (8.5% vs 5% and 3.9% vs 1.2 %, respectively; both p < 0.0001).Conclusions: We used high-density bioinformatics analysis to identify unique and novel toxicity signatures of AYA leukemia patients. We identified side effects previously published using traditional biostatistical analyses in individual trials, validating this approach. In addition, we identified novel toxicities. These findings suggest that large-scale data analyses are a sensitive and powerful method for simultaneous analyses of outcomes across multiple clinical trials, and in otherwise small populations. It provides a framework for a comprehensive understanding of toxicity and outcome differences among AYAs. [Display omitted] DisclosuresNo relevant conflicts of interest to declare.