Alienor Berges, Martin Johnson and Alexander MacDonald
Quantitative Clinical Pharmacology, Early Clinical Development, IMED Biotech Unit, AstraZeneca, Cambridge, UK
Background
In Oncology drug development, the primary objective of most phase I clinical dose finding studies continues to be identification of a maximum tolerated dose (MTD): a dose/regimen found to have limited “dose‐limiting toxicities (DLT)”, in a small group of patients (typically ≤ 1 out of 6), over a finite period of time (usually 21-28 days following treatment). The scientific and statistical limitations of this approach are well-documented [1]. However, patient characteristics, such as cancer type, health status or prior anti-cancer therapy, may influence the MTD identified in these studies. We have previously reviewed the risk factors for neutrophil toxicity in cancer populations [2], a common DLT in phase I studies, and Lyman et al. have attempted to quantify risk of neutropenic complications in patients undergoing chemotherapy [3].
We present a simulation study to assess the effect of various degrees of patient’s risk to dose-limiting neutropenic complications (represented by severe or febrile neutropenia (SN/FB)) on the MTD of a theoretical anti-cancer agent.
Methods
We used a logistic regression model that includes i) background risk , ii) patient’s risk prior to treatment and based on the patient’s specific risk factors Xi=1,2,…m and iii) dose-related drug effect toxicity.
Logit(pSN/FB) = a + b1 * X1 + … + bm * Xm + C * DOSE
The probability of neutropenia (pSN/FB) was calculated from Logit(pSN/FB) and binary individual SN/FB events (DV=1 if presence or DV=0 if absence) were generated according to a binomial distribution given pSN/FB and a sample size per group. The binomial distribution led to natural variability in the toxicity for a given dose, in a given patient among the simulations.
The simulations were generated based on a typical dose escalation study design: 5 dose levels (0.1, 1, 3, 6 and 9 mg) and 6 patients per dose group. Four scenarios were selected, based on drug toxicity and patient’s risk prior to treatment:
- Low and high-toxicity drug, with a parameter value such as pSN/FB at the top dose was associated to 0.5 and 0.9 respectively;
- Low and high-risk population, with a combination of risk factors Xi such as pSN/FB prior to treatment was associated to 0.05 and 0.25 respectively. The risk factors and their regression coefficient values bi were obtained from the risk model from Lyman et al. [3]
For each study, 1000 replicates were simulated to obtain the predictive distribution of MTD values. The simulations were performed in R software (version 3.5.1). Count of SN/FB events was calculated from the simulated individual SN/FB events per dose group, and MTD was determine per study using the operational definition.
Results
Based on Lyman model [3], we simulated two sets of patient’s characteristics; one associated with the low-risk population (age <65 years, breast tumor, no prior chemotherapy, normal white blood cell and normal liver enzymes) and one associated with the high-risk population (age ≥65 years, small-cell lung cancer, prior chemotherapy, low white blood cell and elevated liver enzymes).
The distributions of MTD values were significantly different between the low and the high-risk populations. For the low-toxicity drug, the most likely MTD was at the high doses (9 mg and above) in the low-risk population, and at the low doses (0.1 and 1 mg) in the high-risk population. For the high-toxicity drug, the most likely MTDs in low-risk population and high-risk population were up to 6 mg and up to 1 mg respectively.
Additional simulations using other risk models from literature are planned to investigate the MTD impact for other risk factors and/or other degrees of correlation. In addition, trials combining low and high-risk patients would give a more representative Phase I patients population.
Conclusion
Those simulations including baseline patient’s risks, reveal a clear shift in the MTD distributions between the low and high-risk populations, irrespective of the level of drug toxicity. Although those simulated cases are the extreme SN/FB predicted risks, they illustrate the importance of accounting for risk factors in the MTD concept, determination and application for dose finding.
References:
[1]. Symposium. Challenging the maximum tolerated dosing paradigm in oncology: Threading the needle with targeted agents. ASCPT Annual Meeting (Atlanta, GA, March 2014).
[2]. Casabianca et al. Literature Review to Explore the Risk Factors of Neutrophil Toxicity in Oncology. PKUK conference November 2018
[3]. Lyman et al. Predicting Individual Risk of Neutropenic Complications in Patients Receiving Cancer Chemotherapy Cancer. 2011 May 1; 117(9): 1917–1927
Reference: PAGE 28 (2019) Abstr 8857 [www.page-meeting.org/?abstract=8857]
Poster: Drug/Disease Modelling - Oncology