Alina Sofronova (1), Kirill Zhudenkov (1), Sergey Gavrilov (1), Oleg Stepanov (1), Kirill Peskov (1, 2)
(1) M&S Decisions LLC, Moscow, Russia (2) Computational Oncology Group, I.M. Sechenov First Moscow State Medical University, Moscow, Russia
Introduction: Tumor size is a validated surrogate biomarker for assessment of treatment efficiency and survival in patients with non-small cell lung cancer (NSCLC) [1,2]. However, many other biomarker measurements are available both in baseline and longitudinal setting. Lactate dehydrogenase (LDH), neutrophil count (NEUT), creatinine (CREAT) and alkaline phosphatase (ALP) may be of interest. In certain way, these biomarkers represent the oxidative stress level, blood status, kidney and liver function.
Objectives: The current research quantified the association of baseline and longitudinal biomarkers (SLD, LDH, NEUT, CREAT, ALP) with overall survival (OS) in patients with NSCLC and compared predictive efficacy of the selected biomarkers using Cox proportional hazards, linear univariate and multivariate joint models (uJM and mJM, respectively) [3].
Methods: Data from the control arms of two phase 3 clinical trials in second-line metastatic NSCLC were obtained from Project Data Sphere repository. The training dataset included 596 patients treated with standard chemotherapy docetaxel from ZODIAC trial (NCT00312377) and was used for the qualification of Cox proportional hazards and joint models. The second dataset was used for model validation. It comprised the data for 512 patients treated with targeted therapy erlotinib (NCT00364351).
Cox models of SLD biomarker in combinations with other baseline biomarkers (LDH, NEUT, ALP, CREAT) were tested first. The optimal Cox model was selected based on the lowest Bayesian Information Criterion value and statistical significance of considered covariates. Similar tests were performed for univariate and multivariate joint models given SLD as longitudinal biomarker and other biomarkers were considered either baseline or longitudinal. Longitudinal biomarker dynamics was described with natural splines.
To compare the quality of survival prediction, we estimated model performance by using area under the receiver-operating characteristic curve (ROC AUC) and Brier score (BS). ROC AUCs and BSs were calculated for patients from training and validation datasets with 90-day longitudinal data cut-off (for respective Cox models only baseline biomarker values were used). Multiple study-level simulated survival probability distributions were compared to observed KM curve using the logrank test. All models were developed and tested in JMbayes package in R [4].
Results: Patients in both considered datasets were similar in their disease status, demographics and baseline characteristics. In particular, they had similar baseline values of most biomarkers, in spite of different types of treatment.
The optimal Cox model, uJM and mJM models were identified. Among considered biomarkers (LDH, NEUT, CREAT, ALP) only NEUT and LDH could provide the model performance gain, though NEUT could not be considered as longitudinal due to neutropenia effects in the training dataset with chemotherapy treatment. As a result, Cox model with baseline SLD, LDH and NEUT was selected as optimal (Cox). Similar univariate JM with longitudinal SLD was selected (uJM SLD). The optimal mJM included longitudinal SLD and LDH biomarkers without any additional baseline biomarkers (mJM SLD&LDH).
Joint models outperformed Cox model according to ROC AUC and BS scores. Accumulation of longitudinal data up to 90 days enhanced prediction accuracy of uJM SLD and mJM SLD&LDH on the training dataset as compared with Cox model. Moreover, uJM SLD and mJM SLD&LDH outperformed Cox model on the validation dataset with another type of treatment.
Accounting of SLD and LDH dynamics up to 90 and 180 days in uJM SLD and mJM SLD&LDH improved accuracy of OS predictions, as compared to survival models with only baseline covariates (Cox).
Conclusions: The joint modeling platform for the multivariate search of static and dynamic biomarkers has been proposed. Platform utilized the information from different baseline and longitudinal biomarkers and their combinations by means of Cox proportional hazards and joint models. Inclusion of longitudinal biomarkers (SLD and LDH, for the selected datasets) provided higher model predictive performance, particularly, in ROC AUC and overall survival, in comparison with baseline values only in Cox proportional hazards models.
References:
[1] Gavrilov S et al. Longitudinal assessment of tumor size and neutrophil count in multivariate joint models are more predictive of survival than their baseline values in patients with non-small cell lung cancer. PAGE Meeting, 2019.
[2] Gavrilov S et al. Longitudinal tumor size and NLR are more predictive of individual survival than their baseline values in patients with non-small cell lung cancer treated with durvalumab. 2019 ASCO Annual Meeting.
[3] Rizopoulos D. Joint Models for Longitudinal and Time-to-Event Data. CRC press,2012,ISBN 9781439872864.
[4] Rizopulous D. The R Package JMbayes for Fitting Joint Models for Longitudinal and Time-to-Event Data Using MCMC. J Stat Soft, 2016, 72, 7.
Reference: PAGE () Abstr 9488 [www.page-meeting.org/?abstract=9488]
Poster: Drug/Disease Modelling - Oncology