A joint tumour dynamics and C-reactive protein turnover model to identify early longitudinal prognostic predictors of overall survival in advanced non-small cell lung cancer patients
Yomna M Nassar (1,2), Francis Williams Ojara (1,2,3), Alejandro Pérez Pitarch (4), Kimberly Geiger (5), Wilhelm Huisinga (6), Robin Michelet (1), Stefan Holdenrieder (5), Markus Joerger (7), Charlotte Kloft (1)
(1) Department of Clinical Pharmacy and Biochemistry, Institute of Pharmacy, Freie Universitaet Berlin, Germany; (2) Graduate Research Training Program PharMetrX, Berlin/Potsdam, Germany; (3) Department of Pharmacology and Therapeutics, Faculty of Medicine, Gulu University, Uganda; (4) Translational Medicine & Clinical Pharmacology, Boehringer Ingelheim Pharma GmbH & Co. KG, Ingelheim am Rhein, Germany; (5) Institute of Laboratory Medicine, Munich Biomarker Research Centre, German Heart Centre of the Free State of Bavaria, Technical University Munich, Germany; (6) Institute of Mathematics, University of Potsdam, Germany; (7) Department of Medical Oncology and Haematology, Cantonal Hospital, St. Gallen, Switzerland
Background
C-reactive protein (CRP), a marker of inflammation, positively correlates with disease aggressiveness and poor prognosis in advanced cancers including non-small cell lung cancer (NSCLC). So far, focus has been on the prognostic potential of baseline CRP rather than CRP concentrations across time despite the fact that time-varying CRP concentrations, during treatment, are more reflective of the disease dynamics. Hence, we aimed to investigate the impact of early longitudinal CRP metrics on patients’ overall survival (OS) by (a) characterising longitudinal CRP concentrations, informed by tumour dynamics and (b) exploring the prognostic impact of early CRP metrics, along with other covariates, on OS using a parametric time-to-event (TTE) modelling approach.
Methods
A clinical dataset including 257 advanced NSCLC patients of the CEPAC-TDM study [1] was used for model development. Patients received paclitaxel combined with a platinum-based drug every 3 weeks (i.e. cycle length) for up to 6 cycles. Tumour assessment was performed according to the RECIST criteria [2] at baseline and before the start of cycle 3, cycle 5, at the end-of-treatment, and during follow-up. CRP was measured using a validated commercial assay on day 1 of cycles 1, 2, 3; day 2 of cycles 1, 2; and at end-of-treatment. Patients were followed-up every 3 months for survival.
As a first step, a joint tumour size-informed CRP model was developed to derive longitudinal predictions for CRP concentrations. Longitudinal tumour size (TS) data were estimated using a previosly developed tumour growth inhibition model describing a linear tumour growth, and accounting for paclitaxel effect and development of resistance [3]. TS, as metric of tumour dynamics, was then linked to CRP synthesis described within a turnover model of CRP [4].
As a second step, a parametric TTE model of OS was developed, exploring constant, Weibull and Gompertz hazard functions. Besides baseline CRP, different CRP metrics were derived from early longitudinal CRP concentrations (i.e. first 3 cycles) and, along with other covariates of TS, disease aggresiveness and patient’s health status, were evaluated as prognostic factors on the hazard function describing OS using an SCM approach [5].
All modelling work was performed with NONMEM 7.4.3 using PsN 4.8.1 and visualisations were performed in R 3.5.3.
Results
A linear model linked the TS relative to baseline TS to CRP synthesis within the turnover model; where a unit change in the ratio of TS to baseline TS positively increased CRP synthesis by 81.9%. A Weibull hazard function adequatly characterised OS data. CRP concentrations on day 1 of cycle 3 (CRPcycle3), presence of liver lesions, difference in CRP concentration between cycles 3 and 2 (CRPcycle3-cycle2), and baseline TS significantly affected the hazard (magnitude of risk of death, λ) while the difference in CRP concentration between cycles 3 and 1 (CRPcycle3-cycle1) significantly affected the risk-trajectory (change in hazard across time, α). Baseline CRP concentrations or neutrophil-to-lymphocyte ratio had no effect on the hazard. The OS of a subpopulation of patients with CRPcycle3, CRP cycle3-cycle2, and baseline TS values corresponding to the 95th percentile of the respective distribution, in presence of liver lesions was 8.9 months shorter than the observed median OS (1.4 months vs. 10.3 months, respectively) whereas median OS was not reached after 32.5 months for a subpopulation of patients with CRPcycle3, CRP cycle3-cycle2, and baseline TS values corresponding to the 5th percentile of the respective distribution, in absence of liver lesions.
Conclusion
This developed tumour dynamics-CRP model adequately characterised the longitudinal CRP concentrations providing mechanistic linkage between chemotherapy-driven tumour size dynamics and CRP as a marker of inflammation and disease aggressiveness. Moreover, besides disease-related factors, longitudinal CRP metrics were stronger prognostic factors compared with baseline CRP conncetrations. Longitudinal biomarker data should be further exploited and chosen over baseline data for better disease- and patient-related predictions.
References:
[1] M. Joerger et al. Ann. Oncol. 27: 1895–1902 (2016).
[2] E.A. Eisenhauer et al. Eur. J. Cancer 45: 228–247 (2009).
[3] F.W. Ojara et al. CPT:PSP. doi:10.1002/psp4.12937.
[4] Y.M. Nassar et al. (2022) [https://sites.altilab.com/files/CONGRES/2022/PAMM-ABSTRACTS-BOOKLET.pdf ]
[5] J.S. Owen, J. Fiedler-Kelly. Wiley. (2014).
