Enhanced Outcome Prediction in Early Stage Classical Hodgkin Lymphoma Using Pre-Treatment Biomarkers and Interim PET (BioPET); A Sub-Analysis of the UK NCRI RAPID Trial

Introduction: Interim PET identifies patients with early stage classical HL (cHL) suitable for risk-adapted treatment escalation or de-escalation, but relapse-free survival remains inferior for patients with a negative interim PET who omit radiotherapy. Genetic risk predictors have demonstrated potential to enhance the negative predictive value of interim PET. In the BioPET study reported here, we evaluated the association between a priori selected candidate genes with interim PET and cHL-specific event free survival (cHL-EFS) for patients enrolled on the UK NCRI RAPID trial (NCT00943423).

Methods: Patients with stage 1A or 2A cHL treated with 3 cycles of ABVD followed by interim PET assessment using a 5-point scale, full clinical data and available diagnostic biopsy material were included. Patients with a score of 1-2 (PET ‘negative’) were randomised (1:1) to involved field radiotherapy (IFRT) or no further treatment (NFT); those with a score of 3-5 (PET ‘positive’) received a further cycle of ABVD plus IFRT. Pre-treatment diagnostic FFPE material was obtained for 227 patients (21 with cHL events). Tissue homogenates were prepared and analysed using Quantigene 2.0 (QG_2.0) Plex for expression of 57 candidate genes known to be associated with treatment response and survival in cHL. QG_2.0 data were generated for experimental samples (n=227), RNA controls (n=15) and FFPE controls (n=12). Data were capped at both upper and lower limits of detection. Four housekeeper genes with the lowest variance (GUSB, TBP, HMBS, ABL1) were used for normalisation using the geometric mean. Candidate genes were ranked according to variability of expression. The association between gene expression, PET outcomes and cHL-EFS (disease progression or death) in the three treatment groups was evaluated in a series of regression analyses (Cox and binary logistic), both in univariable and multivariable settings using stepwise procedures, taking baseline EORTC and GHSG risk stratification into account. Analyses were run on the full dataset as there were insufficient cases for a training:validation split.

Results: In total, cHL events were observed in 10/121 (8.3%) PET score 1, 4/53 (7.5%) PET score 2, 2/33 (6.1%) PET score 3, 1/10 (10.0%) PET score 4 and 4/10 (40.0%) PET score 5 respectively. Several genes were found to be associated with PET response after ABVD, and two genes remained in the multivariable model: PRF1 increased the risk of PET score 3-5 (OR=1.49, 95% CI: 1.05-2.13, p=0.03); BCL2L1 decreased risk (OR=0.65, 95% CI: 0.44-0.96, p=0.03). BCL2L1 was also strongly associated with a lower PET score (OR=0.62, 95% CI: 0.46-0.83, p<0.01) in multivariable analysis of ordinal PET scores (PET score 1/2/3/4/5). We found no evidence of an association between any of the genes and PET score 4-5 only, likely due to the small number of cases in this group (n=20). Three genes (CD22, BID, IL15RA) were associated with cHL-EFS and included in the multivariable model after adjusting for PET ordinal score (CD22 (HR=0.54, 95%CI: 0.41-0.72, p<0.001); BID (HR=4.04, 95%CI: 1.79-9.14, p=0.001; IL15RA (HR=0.39, 95%CI: 0.16-0.97, p=0.04); PET score p=0.06). Using this model, a cut-off value of 0.69 for the predicted probability performed best in a time-dependent ROC curve analysis of cHL-EFS with a true positive rate of 67.7% and a false positive rate of 9.5% (Figure 1). There was no evidence of an association with baseline EORTC and GHSG risk scores, and the observed results for genes were similar after adjusting for risk score. We also considered a combined variable of treatment failure (PET score 4-5 or cHL event; n=36) versus success (PET score 1-3 and no cHL event; n=191), adjusting for study group; five genes were associated with treatment failure (CD22, BCL2, SH2D1A, ITGA4, CD3D), and CD22 (HR=0.66, 95% CI: 0.51-0.85, p=0.02) remained in the multivariable model.

Conclusions: A combined multivariable model using interim PET and selected pre-treatment genes shows promising utility for enhanced prediction of cHL-EFS in early stage cHL that is independent of EORTC and GHSG pre-treatment clinical risk scores. These findings warrant further evaluation in an independent cohort with a view to more precisely individualising treatment, improving disease control and minimising late toxicity for patients with early stage cHL.

Figure 1. Kaplan-Meier survival plot of cHL-EFS based on expression of CD22, BID and IL15RA