C-Reactive Protein as an Early Predictor of Efficacy in Advanced Non-Small-Cell Lung Cancer Patients: A Tumor Dynamics-Biomarker Modeling Framework

Abstract

In oncology, longitudinal biomarkers reflecting the patient's status and disease evolution can offer reliable predictions of the patient's response to treatment and prognosis. By leveraging clinical data in patients with advanced non-small-cell lung cancer receiving first-line chemotherapy, we aimed to develop a framework combining anticancer drug exposure, tumor dynamics (RECIST criteria), and C-reactive protein (CRP) concentrations, using nonlinear mixed-effects models, to evaluate and quantify by means of parametric time-to-event models the significance of early longitudinal predictors of progression-free survival (PFS) and overall survival (OS). Tumor dynamics was characterized by a tumor size (TS) model accounting for anticancer drug exposure and development of drug resistance. CRP concentrations over time were characterized by a turnover model. An x-fold change in TS from baseline linearly affected CRP production. CRP concentration at treatment cycle 3 (day 42) and the difference between CRP concentration at treatment cycles 3 and 2 were the strongest predictors of PFS and OS. Measuring longitudinal CRP allows for the monitoring of inflammatory levels and, along with its reduction across treatment cycles, presents a promising prognostic marker. This framework could be applied to other treatment modalities such as immunotherapies or targeted therapies allowing the timely identification of patients at risk of early progression and/or short survival to spare them unnecessary toxicities and provide alternative treatment decisions.

Keywords: C-reactive protein; biomarkers; non-small-cell lung cancer; nonlinear mixed-effects modeling; overall survival; prognosis; progression-free survival; time-to-event analysis.

Figure 1

Schematic overview of the different stages undertaken to identify significant predictors of progression-free survival and overall survival. CRP: C-reactive protein.

Figure 2

CRP concentrations versus time (a) on a linear scale, and (b) as fold change in CRP concentrations relative to baseline, color-coded by sample time. Colored dots: CRP concentrations; gray lines: connected data points per individual. CRP: C-reactive protein.

Figure 3

Schematic representation of the coupled tumor dynamics-CRP model. In the CRP turnover model (dark blue), CRP concentration was influenced by two production rate constants ($K_{in}$ ) and ($K_{in,basal}$), the latter being unperturbed by influential variables or tumor size to ensure a basal level of CRP concentration. CRP production rate constant ($K_{in}$) was influenced by baseline interleukin 6 (IL-6), disease stage, and smoking status. The tumor size model-derived longitudinal ratio of tumor size to baseline tumor size (pink) informed CRP production through a linear relationship (i.e., slope parameter). $K_{out}$: CRP degradation rate constant. CRP: C-reactive protein.

Figure 4

Kaplan-Meier visual predictive checks and impact of identified predictors on median PFS. (a) Kaplan-Meier visual predictive checks (n = 250) comparing predictive performance of time-to-event final PFS model with lognormal hazard function and identified predictors, to the observed PFS data. Solid line: observed PFS data (thin vertical lines represent censoring times corresponding to the time of the patient’s last participation in the study); dashed line: median model-predicted profile, with 90% confidence interval (beige shade). (b) Forest plot of the impact of identified significant predictors on median PFS. Effects of continuous predictors (i.e., CRP_cycle3, CRP_cycle3-2) are shown at the 5th and 95th percentiles of the respective predictor. Black dots: predictor effects; horizontal lines: 95% confidence intervals. (c) Kaplan-Meier plots of simulated (n = 250) PFS profiles under the combined effect of the 5th percentile of the continuous predictors (CRP_cycle3: 1.34 mg/L, CRP_cycle3-2: −6.30 mg/L) and the 95th percentile of the continuous predictors (CRP_cycle3: 65.7 mg/L, CRP_cycle3-2: −0.121 mg/L). Dashed black line: simulated profile at the 95th percentile predictor level; solid black line: simulated profile at the 5th percentile predictor level; pink shaded area: 90% confidence intervals of simulated profiles at 95th percentile predictor level; purple shaded area: 90% confidence intervals of simulated profiles at 5th percentile predictor level; dotted horizontal line: 50% PFS; dashed vertical line: median PFS time at 95th percentile values of the predictors; dotted vertical line: observed median PFS time; solid vertical line: median PFS time at 5th percentile values of the predictors. CRP: C-reactive protein; PFS: progression-free survival.

Figure 5

Kaplan-Meier visual predictive checks and impact of identified predictors on median OS. (a) Kaplan-Meier visual predictive checks (n = 250) comparing predictive performance of time-to-event final OS model with Weibull hazard function and identified predictors to the observed OS data. Solid line: observed OS data (thin vertical lines represent censoring times corresponding to the time of the patient’s last participation in the study); dashed line: median model-predicted profile, with 90% confidence interval (green shade). (b) Forest plot of the impact of identified significant predictors on median OS. Effects of continuous predictors (i.e., CRP_cycle3, CRP_cycle3-2, baseline tumor size) are shown at the 5th and 95th percentiles of the respective predictor, and effects of categorical predictors (i.e., liver lesions) are shown relative to the reference category. Black dots: predictor effects; horizontal lines: 95% confidence intervals. (c) Kaplan-Meier plots of simulated (n = 250) OS profiles under the combined effect of the 5th percentile of the continuous predictors (CRP_cycle3: 1.40 mg/L, CRP_cycle3-2: −7.56 mg/L, baseline tumor size: 2.34 cm) in absence of liver lesions and the 95th percentile of the continuous predictors (CRP_cycle3: 80.7 mg/L, CRP_cycle3-2: −0.114 mg/L, baseline tumor size: 16.9 cm) in presence of liver lesions. Dashed black line: simulated profile at the 95th percentile predictor level in presence of liver lesions; solid black line: simulated profile at the 5th percentile predictor level in absence of liver lesions; pink shaded area: 90% confidence intervals of simulated profiles at 95th percentile predictor level in presence of liver lesions; purple shaded area: 90% confidence intervals of simulated profiles at 5th percentile predictor level in absence of liver lesions; dotted horizontal line: 50% OS; dashed vertical line: median OS time at 95th percentile values of the predictors in presence of liver lesions; dotted vertical line: observed median OS time; solid vertical line: median OS time at 5th percentile value of the predictors in absence of liver lesions. CRP: C-reactive protein; TS: tumor size.

Figure 6

Upper panel: Kaplan-Meier visual predictive checks (n = 250) of simulated (a) progression-free survival events and (b) overall survival events at different concentrations (i.e., percentiles) of CRP_cycle3. Different colors in the upper panel indicate different percentiles. Lower panel: Kaplan-Meier plots of observed distribution of (c) progression-free survival and (d) overall survival events stratified by model-estimated CRP_cycle3, color-coded by the different percentile intervals. Percentile intervals were chosen so that their median corresponds to the percentiles of the simulated profiles in the upper panel, i.e., 5th, 25th, 50th, 75th, and 95th. Solid colored line: observed progression-free survival or overall survival; thin vertical lines: censoring times corresponding to the time of the patient’s last participation in the study; different colors indicate different percentile intervals.