Correlative analysis of response to treatment and biomarker levels in a setting with a time-to-event efficacy outcome and sparse biomarker data
I. Ludwig (1), J. Zhou (2)
(1) Novartis Pharma AG, Biostatistical Sciences and Pharmacometrics, Basel, Switzerland, (2) Novartis Pharmaceuticals Corporation, Oncology Clinical Pharmacology, East Hanover, NJ; Current affiliation: Clinical Pharmacology, Biopharm Clinical Development, Sandoz, Princeton NJ, US.
Objectives: The identification of reliable pharmacodynamic markers of drug efficacy can be of great value, e.g. to explore optimal drug schedule options in oncology applications. However, correlative analysis of biomarker levels and efficacy outcomes can be challenging in settings with time-to-event based efficacy endpoints and limited availability of biomarker samples. The goal of this analysis is to assess a potential correlation between biomarker levels and tumor response as characterized by time to progressive disease in patients treated with an investigational drug, making use of predicted biomarker response.
Methods: Data from 230 patients were evaluated for this analysis. Only few biomarker samples per patient were collected, therefore biomarker levels could not directly be related to efficacy outcomes (time of progressive disease observation). Instead, biomarker levels at the time of disease progression were estimated from an Emax response model established from earlier studies, using Ctrough estimates from an existing population PK model. Exploratory graphical analysis of these predictions was performed to compare predicted biomarker levels at the time of disease progression with the levels of a corresponding at-risk population.
Results: The analysis approach allowed to further assess hypotheses initially generated upon visual inspection of patient-level data, summarizing and putting in context information from the whole study population. E.g. whether, after treatment discontinuation, elevated biomarker levels may be associated with an increased risk of disease progression. In the studied case, this hypothesis was supported, which encourages further exploration of the biomarker as a potential marker of drug efficacy.
Conclusions: Model-based prediction of biomarker levels using prior knowledge from population PK and PK-biomarker models allowed assessment of a potential correlation with tumor response outcomes (time of progressive disease) despite limited availability of biomarker data. The proposed graphical analysis is a well suited means to explore general trends in the data, support hypothesis generation and make a basis for fruitful discussions with the clinical team. We conclude that this approach may be valuable also in a wider setting of exposure-response analyses with time-to-event endpoints.