Assessment of a model to correlate early tumor size response to overall survival in relapsed or refractory diffuse large B cell lymphoma patients
Phyllis Chan (1), Rene Bruno (2), Laurent Claret (2), Tong Lu (1), Dale Miles (1), Chunze Li (1), Angelica Quartino (1), Jin Jin (1), Dan Lu (1)
(1) Genentech, South San Francisco, California, USA; (2) Genentech/Roche, Marseille, France
Objectives: To use tumor size metrics measured during treatment to predict overall survival (OS) in patients with relapsed or refractory diffuse large B-cell lymphoma (R/R DLBCL), as a potential approach to enable early decision making [1].
Methods: A sequential approach was used for model development. Longitudinal measurements of tumor size (square root of sum of product of diameters [SPD] of the target lesions) from 190 patients in two polatuzumab vedotin clinical trials (GO27834/NCT01691898 and GO29365/NCT02257567 [2-5]) were characterized by four tumor growth inhibition (TGI) models. The models comprised the simplified TGI model [6], Stein model [7], modified Stein model (using different shrinkage rate parameters to describe the treatment phase and post-treatment follow-up phase), and Chatterjee model [8]. Clinically relevant baseline covariates were assessed for their effect on the TGI model parameters, including tumor SPD at baseline, tumor growth, and shrinkage parameters, using univariate screening and backward elimination. Then a TGI-OS model describing the correlation between TGI model-derived parameters and OS was developed. The final model was determined by the following process: model-estimated tumor metrics and pre-specified clinically-relevant baseline covariates were screened as prognostic factors for OS by univariate Cox-proportional hazard models, based on the criterion of p-value < 0.005; next, backward elimination was applied on the selected prognostic factors, with selection criterion p-value < 0.001, in order to establish the final model using parametric survival analysis. Treatment group was evaluated as a covariate for both the TGI and the TGI-OS models. Model qualification was conducted through posterior predictive checks.
Results: Longitudinal tumor size data were adequately described by the four TGI models. The modified Stein model with normal distribution of shrinkage rate after the end of treatment was determined to be the best, based on the Bayesian Information Criterion. Statistically significant covariates included the effects of baseline lactate dehydrogenase (LDH), Eastern Cooperative Oncology Group (ECOG) performance status, and bulky disease on baseline tumor size. Treatment group was not a statistically significant covariate in the univariate screening step.
Based on univariate Cox model, among the TGI model-derived parameters, the tumor growth rate (KG), baseline tumor SPD, and week 8 to baseline tumor ratio (TR8) were most highly correlated with OS (p-values < 10-5), with KG being the most statistically significant predictor. Higher KG, baseline tumor SPD, and TR8 were all associated with shorter OS. Posterior predictive check plots stratified by KG or TR8 quartiles and baseline covariate categories (e.g. quartiles of time from most recent therapy to study start, LDH, hemoglobin, ECOG status, presence of bulky disease, age) showed that the prediction intervals simulated from the final TGI-OS model correspond well with the observed Kaplan-Meier curves pooled from the two studies. For the baseline covariates, increased model-estimated baseline tumor size was associated with shorter OS. No other statistically significant covariates, including treatment group, were identified for the TGI-OS model.
Individual patient KG or TR8 estimated from the final TGI model that incorporated multiple statistically significant covariates were compared directly among the six treatment groups in these two studies, and p-values from t-tests comparing two samples at a time indicated two treatment groups to be less effective than the other four treatment groups. These results showed TGI-based efficacy comparison between treatment groups is consistent with the comparison based on response criteria for malignant lymphoma [9].
Conclusions: This analysis demonstrated that model-estimated tumor-size metrics such as KG estimated from the TGI model, as early as 8 weeks after start of treatment, could be of value to predict OS. Both the TGI and TGI-OS model are treatment-group independent, which allow comparison of efficacy across different treatment groups based on TGI metrics.
References:
[1] Bruno R, Mercier F, Claret L. Evaluation of tumor size response metrics to predict survival in oncology clinical trials. Clin Pharmacol Ther (2014) 95:386-393.
[2] Morschhauser F, et al. Updated results of a phase II randomized study (ROMULUS) of polatuzumab vedotin or pinatuzumab vedotin plus rituximab in patients with relapsed/refractory non-Hodgkin lymphoma. Blood (2014) 124, Abstract 4457.
[3] Phillips T, Brunvand M, Chen A, et al. Polatuzumab vedotin combined with obinutuzumab for patients with relapsed or refractory non-Hodgkin lymphoma: preliminary safety and clinical activity of a phase Ib/II study. Blood (2016) 128: 622.
[4] Matasar M, Herrera AF, Kamdar M, et al. Polatuzumab vedotin plus bendamustine and rituximab or obinutuzumab in relapsed/refractory follicular lymphoma or diffuse large B-cell lymphoma: updated results of a phase 1b/2 study. Haematologica (2017) 102(s2): 173. abstract n. S468.
[5] Sehn LH, Herrera AF, Matasar MJ, et al. Addition of polatuzumab vedotin to bendamustine and rituximab (BR) improves outcomes in transplant-ineligible patients with relapsed/refractory (R/R) diffuse large B-cell lymphoma (DLBCL) versus BR alone: results from a randomized phase 2 study. Blood (2017) 130:2821.
[6] Claret L, Gupta M, Han K, et al. Evaluation of tumor-size response metrics to predict overall survival in Western and Chinese patients with first-line metastatic colorectal cancer. J Clin Oncol (2013) 31:2110-2114.
[7] Stein WD, Gulley JL, Schlom J, et al. Tumor regression and growth rates determined in five intramural NCI prostate cancer trials: the growth rate constant as an indicator of therapeutic efficacy. Clin Cancer Res (2011) 17:907-917.
[8] Chatterjee M, Turner DC, Felip E, et al. Systematic evaluation of pembrolizumab dosing in patients with advanced non-small-cell lung cancer. Annals Oncol (2016) 27:1291-1298.
[9] Cheson BD, Fisher RI, Barrington SF, et al. Recommendations for initial evaluation, staging, and response assessment of Hodgkin and non-Hodgkin lymphoma: the Lugano classification. J Clin Oncol (2014) 32:3059–3068.
