With an estimated 19 million new cancer cases each year and almost 10 million deaths worldwide, there is a huge need for an optimised development of new therapeutic anticancer compounds as well as prediction of cancer patient response at bedside. Pharmacometric modelling and simulation allow to optimally leverage the rich longitudinal data from clinical studies and the different oncology variables and endpoints as tumour response, biomarker concentrations and survival information to better optimise clinical decision-making in oncology at its different stages: the development of new therapeutic compounds with optimised dosing selection and during therapeutic use to predict patient prognosis for improved treatment decisions at bedside.
Beginning with the development of new therapeutic compounds, identification of the appropriate/optimal dosing for the confirmatory phase III trials that maximises efficacy and minimises toxicity remains the most challenging component of clinical drug development. For this reason, a better understanding of the impact of different dose levels on efficacy and toxicity is needed to characterise the dose-response relationship and offer a more rational derivation of optimal doses/dosing. Compared to traditional pairwise comparisons between different study arms of dose-finding studies, statistical and model-based approaches have been shown to best leverage phase II dose-finding study data and characterise the dose-response relationship. These different approaches have been endorsed by the US Food and Drug Administration (FDA) which has recently initiated “Project Optimus” with the aim to optimise dose finding in oncology. Therefore, project I aimed to compare the performance of two model-based approaches within the oncology setting: the recently proposed combined Likelihood Ratio Test (cLRT) which leverages longitudinal phase II data and has shown high power detecting a dose-response relationship—but is computationally expensive, and the Multiple Comparison Procedure (MCP), the earlier and more established approach that has gained the FDA’s qualification as “fit-for-purpose” for the design and analysis of phase II studies. A simulation-based framework of a dose-finding phase II study under different study design considerations was established and applied to investigate cLRT and MCP performance. The results showed that, in general, cLRT was associated with higher power (=1−type II error) compared with MCP (89.8% vs 27.0%); however, its type I error (i.e. false positive) was not well controlled (mean: 13%) compared with MCP (<4%). Moreover, cLRT power was less sensitive to the different study design variables (e.g. number of patient with respect to number of dose levels) in contrast with MCP. Therefore, based on these results, before cLRT can be recommended to analyse dose-finding studies in oncology, further investigation of its robustness to different model complexities and study design variables as well as investigation of conditions that would better control type I error are needed to justify its high power at the expense of its computational demands.
Equally important to the development of new therapeutic compounds with optimised dose selection, is the accurate and early prediction of patient prognosis to monitor patient response and improve treatment decisions at bedside. Non-small cell lung cancer (NSCLC), the leading cause of cancer-related death, represents a disease of high burden and poor prognosis. Therefore, project II was conducted with the aim to identify early predictors of efficacy for NSCLC patients to spare them the unnecessary exposure to toxicities and contribute to better prediction of treatment outcomes. Clinical data from patients with advanced NSCLC receiving first-line combination therapy with paclitaxel and a platinum-based drug, were leveraged to characterise and quantify the relationships between anticancer drug exposure, tumour dynamics and C-reactive protein (CRP) concentrations—as a measure of the inflammatory level, using pharmacometric modelling. Model-derived variables were then investigated as potential predictors of the most important and commonly adopted efficacy endpoints progression-free survival (PFS) and overall survival (OS) by means of parametric time-to-event modelling, with a special focus on the potential of early longitudinal biomarker information as a potential early prognostic predictor. The results of our modelling framework in which longitudinal CRP concentrations were leveraged for the first time for prognostic investigations, identified the inflammatory level at treatment cycle 3 (CRPcycle3), i.e. day 42 from start of treatment, and the extent of absolute reduction in the inflammatory level between treatment cycles 3 and 2 to be the most significant predictors of PFS. Besides the CRP-related metrics, baseline tumour size and presence/absence of liver lesions were found to be predictors of OS. Nevertheless, CRPcycle3 was by far of the highest impact. The identification of CRP at treatment cycle 3 points to the potential and more informative value of longitudinal biomarker data compared to the commonly applied approach in which only baseline (pre-treatment) measurements are investigated and which do not reflect the patient situation and dynamic evolution of the disease. Measuring longitudinal CRP as a routine biomarker allows for the monitoring of inflammatory levels and, along with its reduction across treatment cycles, presents a promising prognostic marker for the timely identification of patients at risk of therapeutic failure, early progression and/or short survival to spare them unnecessary toxicities and provide alternative treatment decisions.
Overall, the two projects presented in this thesis, acknowledged and addressed, by leveraging pharmacometrics methodologies, two critical needs within the scope of oncology drug development and therapeutics that require early and more optimised clinical decisions. First, was the need to optimally identify the drug effect for an accelerated and optimised development of efficacious compounds for oncologic indications. Based on a scientific understanding of the dose-response relationship during early clinical drug development, a more informed and optimised dosing selection can be achieved. Within the investigated approaches, MCP, a robust but less powered approach currently exists. However, for better power, systematic investigations of cLRT are needed to achieve a more robust performance. Through the establishment of a proper understanding of the dose-response relationship, decisions regarding optimal dosing selection can be better informed that would maximise the drug’s efficacy, safety and tolerability and would consequently reflect on more successful phase III trials and lower attrition rates. Second, was the need to accurately and timely predict treatment outcomes and prognosis to monitor patient response, and improve treatment decisions at bedside. Our developed modelling framework, successfully identified the minimally invasive and cost-effective biomarker, CRPcycle3, as a significant predictor of PFS and OS. It also proposed that monitoring the inflammatory level was of prognostic value. The application of this modelling framework goes beyond NSCLC and is envisioned to suit other treatment modalities such as immunotherapies or targeted therapies for prediction of patient response and treatment outcomes in different settings. Finally, with use of proper methodologies as pharmacometric modelling and simulation, different data were successfully leveraged to optimise dosing selection and patient monitoring for a pharmacometric-based optimisation of early clinical decision-making in oncology.
