IV-034 Florencia Tettamanti

Predicting Progression Free Survival from Measurable Residual Disease (MRD) in Chronic Lymphocytic Leukemia (CLL)

Florencia A. Tettamanti (1), Holly Kimko (2), Shringi Sharma (3), Giovanni Di Veroli (1)

(1) CPQP, CPSS, BioPharmaceuticals R&D, AstraZeneca, Cambridge, UK, (2) CPQP, CPSS, BioPharmaceuticals R&D, AstraZeneca, Gaithersburg, USA (3) CPQP, CPSS, AstraZeneca, South San Francisco, California, USA

Introduction/Objectives: 

Measurable Residual Disease (MRD) is detectable and often quantifiable remnant cancer cells following treatment. The association between MRD and survival outcomes in haematological cancers, including Chronic Lymphocytic Leukaemia (CLL), has often been reported [1,2]. However, limited quantitative analyses has been undertaken to establish the predictive power of MRD in relation to progression. Better understanding of the predictive ability of MRD and its features (e.g., absence vs baseline vs. longitudinal changes) may also help improve our prognostic potential both at a trial and individual level. To this end, we collected published CLL data (the indication where data was most abundant) and conducted analyses to evaluate the potential of MRD to assess early futility of clinical trials and to predict individual risk of progression.

Methods: 

We undertook a systematic review of MRD in CLL. Our analysis is comprised of two parts. The first, is a meta-regression model implemented in R package meta 6.5.0 [3] which predicts median PFS from undetectable MRD rates at 3-6 months follow-up using study level data from eight clinical trials. The second, is a joint modelling approach, in which we investigated the evidence for predicting PFS from competing MRD derived metrics using four small studies. The data used for this second analysis consisted of individual MRD trajectory data, and study population PFS data. In lieu of linked individual progression and MRD trajectory data, we adapted the standard joint modelling approach as follows. First, we fit an MRD dynamic model to individual data using a non-linear mixed effects approach implemented in Monolix [4,5].  Then, for each individual in a study, we projected their PFS based on three competing joint models. In Model 1 survival was independent of any MRD metric, whereas in Model 2 and Model 3 the survival was informed by baseline MRD value and normalized instantaneous MRD value derived from fits of the MRD dynamic model, respectively. For each joint model, the mean PFS of the individuals in a study was then compared to the reported study PFS by residual sum of squares. We implemented an Approximate Bayesian Computation rejection algorithm in MATLAB [6,7] for joint model selection and parameter estimation.

Results: 

Using data from eight clinical trials, we constructed a meta-regression model which shows that undetectable MRD rates at 3-6 months post follow-up are highly correlated to median PFS (R2 = 0.94, p = 0.004). We evaluated our meta-regression model on two small studies not used for model building and found the predictions to be adequate. The resulting model can be used to predict the probability of technical success of a planned clinical trial.

We also investigated the evidence for predicting PFS from competing MRD derived metrics via a joint modelling approach. Using data from four small studies, we found strong evidence that including MRD derived metrics in joint models improves predictions of PFS compared to not including them. Specifically, in all but one study the joint model informed by the continuous MRD metric, Model 3, was the most strongly supported by the data (Bayes Factor ≥ 2). For the remaining study, which had smaller follow-up times, no single model was better supported by the data although Model 3 was the least well supported. Our analyses suggest that MRD metrics are consistent with observed population PFS and their inclusion could improve PFS predictions. It is therefore proposed that systematic MRD metric collection is accompanied by modelling for specific therapies and populations to generate algorithms that, once validated, inform patients’ progression.

Conclusion: 

Our analysis provides a comprehensive assessment of published MRD data to explore the utility of MRD metrics in the prediction of PFS. It provides further evidence that incorporating MRD-derived metrics such as baseline levels and longitudinal changes improves predictions of PFS. It demonstrates how these could be used to predict survival at a population and individual level. Future access to more comprehensive data may provide the opportunity to strengthen these findings.

References:

  1. Kwok M, Rawstron AC, Varghese A, Evans PAS, O’Connor SJM, Doughty C, et al. Minimal residual disease is an independent predictor for 10-year survival in CLL. Blood. 2016;128(24):2770–3.
  2. Santacruz R, Villamor N, Aymerich M, Martínez-Trillos A, López C, Navarro A, et al. The prognostic impact of minimal residual disease in patients with chronic lymphocytic leukemia requiring first-line therapy. Haematologica. 2014;99(5):873–80.
  3. Balduzzi S, Rucker G, Schwarzer G. How to perform a meta-analysis with R: a practical tutorial. Evidence Based Mental Health. 2019. 153-60.
  4. Zhudenkov K, Gavrilov S, Sofronova A, Stepanov O, Kudryashova N, Helmlinger G, et al. A workflow for the joint modeling of longitudinal and event data in the development of therapeutics: Tools, statistical methods, and diagnostics. CPT: Pharmacometrics & Systems Pharmacology. 2022 Apr 1;11(4):425–37.
  5. Antony 2019 France: Lixsoft SAS. Monolix version 2018R1. http://lixoft.com/products/monolix/; 2018.
  6. Csilléry K, Blum MGB, Gaggiotti OE, François O. Approximate Bayesian Computation (ABC) in practice. Trends in Ecology & Evolution. 2010 Jul 1;25(7):410–8.
  7. MATLAB [Internet]. Natick, Massachusetts, United States: The MathWorks Inc.; 2022. Available from: https://www.mathworks.com

Reference: PAGE 32 (2024) Abstr 11054 [www.page-meeting.org/?abstract=11054]

Poster: Methodology - Other topics

PDF poster / presentation (click to open)