Excitement around and investment in oncology drug development are at unprecedented levels. To maximize the health impact and productivity of this research and development investment, quantitative modeling should impact key decisions in early clinical oncology including Go/No-Go decisions based on early clinical data, and dose selection for late stage studies.

See related article by Bottino et al., p. 6633

In this issue of Clinical Cancer Research, Bottino and colleagues (1) integrate preclinical exposure-efficacy data from mouse xenograft models with clinical safety data using a modeling and simulation (M&S) framework (which leverages pharmacometrics techniques). They predict that the combination of TAK-117, a PI3Kα inhibitor, and TAK-228, a dual mTOR (mTOR1/2) inhibitor, is unlikely to be better than monotherapy. The combination showed benefit in preclinical efficacy models, but M&S predicted that the clinical exposures required for efficacy are above MTD. Confidence in the M&S prediction relies on assuming translatability of the murine exposure–efficacy relationship. Given the uncertainty associated with this prediction, the combination will be tested in phase Ib/II studies including 350 patients across three cancer indications.

The confidence in translatability of preclinical models in terms of efficacy (and pharmacology) is key to most translational M&S predictions. For new targets and many new combinations, it is almost impossible to demonstrate, or, in many cases, even to assess the prudence of such confidence. This is especially risky in novel immune-oncology targets where both translatability of preclinical syngeneic models and mechanisms of action are poorly understood (e.g., due to homogeneity of tumors in mice and significant differences in human and murine immune systems). Publications retrospectively reviewing ad hoc successes may present prescriptive translational pharmacokinetic–pharmacodynamic (tPKPD) modeling strategies, and such forward translation strategies can mislead drug development for novel targets.

A better scientific strategy is to use reverse translation to build confidence in the translatability of mechanisms of action and of preclinical data. The clinical data from the first-in-class molecules, either positive or negative in various patient populations, can assess proof of mechanism and/or proof of biology. Quantitative systems pharmacology (QSP) provides a quantitative framework to incorporate species differences in pharmacokinetics, pharmacodynamics, and biology. This approach generates mechanistic insights into the potential effects of biological differences between species. The information resulting from these approaches provides a more reliable guide for future development efforts.

In addition to being critical of preclinical efficacy models, we need to learn from gaps in efficacy assessments in early clinical studies. For example, combining cobimetinib (MEK inhibitor) and atezolizumab (anti-PDL1 mAb) showed promising overall response rate (ORR) of 17% in a phase Ib study in colorectal patients, but ORR dropped to 3% and the combination failed to improve overall survival (OS) more than regorafenib (a tyrosine kinase inhibitor) in a pivotal study (2). Another combination, epacadostat (IDO1 inhibitor) and pembrolizumab (anti-PD1 mAb), yielded ORR > 50% in a phase I study in patients with melanoma, but ORR dropped to 34% and the combination failed to improve OS compared with pembrolizumab monotherapy in the pivotal study (3).

Oncology is seeing large investments in clinical testing of monotherapy and combinations with 2,250 trials of PD-1/L1 agents alone, requiring >380,000 patients (4). These clinical trials include combinations of PD-1/L1 with therapies against as many as 240 different targets further demonstrating that preclinical models were not considered a reliable filter for target selection. In clinic, although benefit of a new combination is directly addressed by a randomized study, research and development budgets force prioritization, permitting this strategy only for the highest priority investments. Fierce competition for patient recruitment is also an important factor to consider for large randomized studies. As a result, better portfolio prioritization decisions based on phase I data are needed urgently.

Reliable signal assessment for prioritization based on phase I clinical data needs to explicitly consider potential confounding and the resulting possibility of an optimistic efficacy signal. The sources of biases in phase I are not well understood, but some of these biases can be removed by central versus investigator review of tumor imaging data. Some of the discrepancy between early and late studies can be due to the small size and few study sites.

Despite these inherent challenges, decisions can be improved by understanding of lesion-level efficacy data in early studies. ORR is typically used as a decision criterion for Go/No-Go (GNG) decisions due to historical use and simplicity. However, ORR's categorical nature makes it highly susceptible to noise, potential measurement errors, and marginal reductions in tumor size around the stable disease/partial response cutoff (−30%). Therefore, in some cases, improvements in ORR may not be associated with significant tumor reductions, and therefore poorly predictive of OS. Meaningful differentiation in OS will require deep, as well as durable reductions in tumor size which may be better understood by modeling tumor size–OS relationships.

Mitigating risk in GNG decisions also requires adequate understanding of biomarkers for patient population selection. Clinical trials with umbrella and basket designs can help identify both stratification biomarkers and responsive patient populations. Adaptive, platform trial designs (e.g., I-SPY2; ref. 5) using longitudinal data such as tumor size and predictive markers of disease progression can help reduce the number of patients required to ascertain a reliable efficacy signal.

Once the GNG decision is made, dose selection requires just as much attention, as seen, for example, in potentially counter intuitive data on the ipilimumab/nivolumab combination. Clinical data in non-small cell lung cancer have demonstrated that 1 mg/kg every 6 weeks/every 12 weeks is better regimen than previously tested regimens of 1–3 mg/kg every 3 weeks of ipilimumab in combination with nivolumab (6). Challenges in finding optimal dose for new agents, such as those in immune-oncology, include shortage of systemic biomarkers for dose–response relationship, and high variability (due to spatial heterogeneity) in intratumor biomarkers collected from biopsies. M&S approaches can help overcome some of these challenges. A biologically effective dose (BED) for mAbs can be identified using physiologically based pharmacokinetic models developed using phase I data. These semimechanistic models can be used to predict dose required for target saturation in systemic circulation and tumor (considering spatial heterogeneity) for antagonist mAbs. Target saturation can be used as a surrogate for maximizing pharmacologic effect. Acknowledging gaps in understanding of pharmacology, a randomized dose-finding study should compare BED and MTD/MAD to find the dose for a pivotal study. Finally, closer collaboration between biology, pharmacology, clinical, biostatistics, and pharmacokinetics-pharmacodynamics groups is essential to the successful use of model informed drug development (MIDD) for dose finding.

The current oncology landscape presents an unprecedented opportunity to expand the applications of quantitative thinking (Fig. 1). The credibility, and therefore long-term success, of MIDD relies in acknowledging the translation gaps and carefully choosing programs where MIDD can add value based on science. This will require MIDD groups in biopharmaceutical companies to go beyond standard pharmacokinetic-pharmacodynamic modeling skills by developing deep therapeutic area knowledge, including insights into trial designs and clinical endpoints. Drug development cannot be a pursuit driven solely by biology or clinical organizations. Higher impact on human health and return on research and development investment can be achieved by integrating MIDD into decision making at the team and organizational leadership levels.

Figure 1.

An overview of opportunities and risks for MIDD in oncology. PK, pharmacokinetics; PD, pharmacodynamics.

Figure 1.

An overview of opportunities and risks for MIDD in oncology. PK, pharmacokinetics; PD, pharmacodynamics.

Close modal

All authors are employees of and have ownership interest (including patents) in Merck & Co., Inc. No other potential conflicts of interest were disclosed.

The authors would like to thank Drs. Eric Rubin and Vikram Sinha for insightful comments.

1.
Bottino
DC
,
Patel
M
,
Kadakia
E
,
Zhou
J
,
Patel
C
,
Neuwirth
R
, et al
Dose optimization for anticancer drug combinations: maximizing therapeutic index via clinical exposure-toxicity/preclinical exposure-efficacy modeling
.
Clin Cancer Res
2019
;
25
:
6633
43
.
2.
Eng
C
,
Kim
T
,
Bendell
J
,
Argilés
G
,
Tebbutt
N
,
Bartolomeo
MD
, et al
Atezolizumab with or without cobimetinib versus regorafenib in previously treated metastatic colorectal cancer (IMblaze370): a multicentre, open-label, phase 3, randomised, controlled trial
.
Lancet Oncol
2019
;
20
:
849
61
.
3.
Long
GV
,
Dummer
R
,
Hamid
O
,
Gajewski
TF
,
Caglevic
C
,
Dalle
S
, et al
Epacadostat plus pembrolizumab versus placebo plus pembrolizumab in patients with unresectable or metastatic melanoma (ECHO-301/KEYNOTE-252): a phase 3, randomised, double-blind study
.
Lancet Oncol
2019
;
20
:
1083
97
.
4.
Tang
J
,
Yu
JX
,
Hubbard-Lucey
VM
,
Neftelinov
ST
,
Hodge
JP
,
Lin
Y
. 
The clinical trial landscape for PD1/PDL1 immune checkpoint inhibitors
.
Nat Rev Drug Discov
2018
;
17
:
854
5
.
5.
Harrington
D
,
Parmigiani
G
. 
I-SPY 2 — a glimpse of the future of phase 2 drug development?
N Engl J Med
2016
;
375
:
7
9
.
6.
Hellmann
MD
,
Rizvi
NA
,
Goldman
JW
,
Gettinger
SN
,
Borghaei
H
,
Brahmer
JR
, et al
Nivolumab plus ipilimumab as first-line treatment for advanced non-small-cell lung cancer (CheckMate 012): results of an open-label, phase 1, multicohort study
.
Lancet Oncol
2017
;
18
:
31
41
.