Identifying Trial-Eligible High- and High-Intermediate Risk Patients with Diffuse Large B-Cell Lymphoma Using Prognostic Models: A Real-Life Comparative Analysis of IPI, aaIPI and NCCN-IPI from Danish Study Group

Background: The International Prognostic Index (IPI) and age-adjusted IPI (aaIPI) have been used to select patients in recent frontline clinical trials, aiming to challenge R-CHOP as the standard of therapy. As the prognosis following R-CHOP treatment is favorable for low-risk patients, predictive tools to identify high-risk trial-eligible diffuse large B-cell lymphoma (DLBCL) patients most likely to benefit from experimental treatments are warranted. The lack of overall survival (OS) benefit may partly result from non-optimal patient selection based on prognostic models. Currently, an ongoing frontMIND (NCT04824092) clinical trial comparing tafasitamab + lenalidomide + R-CHOP vs. R-CHOP selected high-intermediate (HI) and high-risk (H) patients using IPI (score 3-5 in patients >60 years) and aaIPI (score 2-3 in patients ≤60). However, it remains unclear if aaIPI and other models can accurately identify high-risk patients for clinical trials.

This study compared the International Prognostic Indices (IPI, aaIPI, NCCN-IPI) used to identify trial-eligible HI and H-risk DLBCL patients. We also investigated the models' prognostic ability in trial-eligible patients aged ≤60, identified through the Danish Lymphoma Registry (LYFO).

Methods: Patients diagnosed between 2000 and 2020 were included if they fulfilled the following criteria extracted from the frontMIND trial: 1) newly diagnosed DLBCL; 2) age 18-80 years; 3) Eastern Cooperative Oncology Group performance status of 0-2; 4) required laboratory at screening (blood counts, renal and liver function); 5) no CNS involvement; 6) no history of malignancy (excluding non-melanoma skin cancer). Moreover, only patients treated with R-CHOP (+/- etoposide) were included.

Results: Of 5,188 patients with newly diagnosed DLBCL treated with R-CHOP/(+/-etoposide) identified in LYFO, 2,961 (57.0%) had sufficient registry data and fulfilled frontMIND trial inclusion criteria (excluding IPI/aaIPI criterion). The median age at diagnosis was 65. Males (n=1,719, 58.1%) and patients diagnosed in advanced stage 3/4 (n=1,900, 64.2%) were predominant. The median follow-up was 70.0 months (IQR, 32.8-112.9).

All three models stratified patients into four risk groups. NCCN-IPI was better in discrimination risk groups (c-index=0.680, p<0.05) than IPI (c-index=0.676) and aaIPI (c-index=0.636). The 5-year OS for H and HI risk patients, according to IPI, aaIPI, and NCCN-IPI, was 57.0% and 71.2%, 63.6% and 75.7%, and 46.1% and 71.3%, respectively.

When H and HI risk patients were selected according to frontMIND inclusion criteria (IPI 3-5 in patients>60 years and aaIPI 2-3 in patients≤60), almost half of the initial population was identified as trial eligible (1,426/2,961, 48.2%) with the 5-year OS of 70.1% (95% CI, 67.7-72.7). If IPI were used to identify HI and H risk patients, fewer patients (n=1,202, 40.6%) with poorer 5-year OS of 67.1% (95% CI, 64.4-69.9) would enter the trial than planned as per the original protocol. A similar 5-year OS of 67.4% was found when NCCN-IPI (score 4-8) was used (n=1,323, 44.7%).

Of 2,961 initially recognized trial-eligible patients, 1,072 were ≤60 years. Of them, 415 (38.7%) were eligible for the frontMIND trial if aaIPI was used to select H and HI risk patients (scores 2-3). The 5-year OS for these patients was 83.7% (95% CI, 80.1-87.4). However, IPI and NCCN-IPI identified fewer patients at H and HI risk, accounting for 17.8% (191 patients) and 18.4% (197 patients). 5-year OS in these patients was slightly lower than that of aaIPI, specifically 80.5% (95% CI, 75.0-86.4) for NCCN-IPI and 81.2% (95% CI, 75.7-87.2) for IPI. The discriminatory ability of aaIPI was not significant compared to IPI and NCCN-IPI (p>0.05).

Conclusion: This retrospective real-world study on 2,961 trial-eligible DLBCL patients investigated the impact of prognostic models as selection criteria for identifying high-risk candidates for clinical trials. Although some recent trials used aaIPI to select younger patients, we found that aaIPI fails to identify true HI and H-risk younger patients as 5-year OS was generally favorable (83.7%) in this real-world population. Although none of the three analyzed models could identify truly high-risk patients ≤60 years, IPI and NCCN-IPI identified fewer trial-eligible patients with a slightly worse prognosis than aaIPI and could also be used to select high-risk trial-eligible patients aged 60 or less.