Dose-escalation strategies which use subgroup information

2 stage subgroup-specific time-to-event (2S-Sub-TITE): An adaptive two-stage time-to-toxicity design for subgroup-specific dose finding in phase I oncology trials

For phase I trials, the subgroup-specific time-to-event (Sub-TITE) design identifies the maximum tolerated dose (MTD) separately in 2+ heterogeneous patient subgroups. Sub-TITE allows borrowing strength and dynamic clustering across subgroups from the trial's start, but delaying the initiation of borrowing and clustering may improve trial accuracy. We propose the 2-stage Sub-TITE (2S-Sub-TITE) design in which the trial starts by estimating separate models per subgroup, and then initiates the Sub-TITE design at some pre-specified point of patient accrual. We evaluate the operating characteristics of the 2S-Sub-TITE design using simulations. Nine configurations of the 2S-Sub-TITE design (varying in timing of initiation of borrowing/clustering and prior probability of subgroup heterogeneity, p_hetero) and three control methods were compared across 1000 randomly-generated true toxicity probability scenarios. Effects of priors, sample size, escalation rules, target toxicity probability, accrual rate, and number of subgroups were evaluated. Metrics included: proportion of correct selection (PCS) of the true MTD, and average number of toxicities incurred. Among the 5 2S-Sub-TITE configurations (out of 9 total) with the highest PCS (45%) when the subgroup heterogeneity assumption is correct (all of which out-perform the control methods by 2%-6%), the configuration which enables borrowing and clustering allowance with p_hetero = 0.7 starting at 75% patient accrual best minimizes toxicities as well as losses in accuracy if the heterogeneity assumption is incorrect. For trials with high confidence in subgroup heterogeneity, the 2S-Sub-TITE configuration enabling borrowing/clustering with p_hetero = 0.7 starting at 75% patient accrual exhibits superior dose-finding accuracy compared to existing methods.

Subgroup‐specific dose finding for phase I‐II trials using Bayesian clustering

In most models and algorithms for dose‐finding clinical trials, it is assumed that the trial participants are homogeneous—the optimal dose is the same for all those who qualify for the trial. However, if there are heterogeneous populations who may benefit from the same treatment, it is inefficient to conduct dose‐finding separately for each group, and assuming homogeneity across all subpopulations may lead to identification of the incorrect dose for some (or all) subgroups. To accommodate heterogeneity in dose‐finding trials when both efficacy and toxicity outcomes must be used to identify the optimal dose (as in immunotherapeutic oncology treatments), we utilize an adaptive Bayesian clustering method which borrows strength among similar subgroups and clusters truly homogeneous subgroups. Unlike methodology already described in the literature, our proposed methodology does not require the assumption of exchangeability between subgroups or a priori ordering of subgroups, but does allow for specification of different subgroup‐specific priors if prior information is available. We provide a comparison of operating characteristics between our method and Bayesian hierarchical models for subgroups in a variety of relevant scenarios. After simulation studies with four a priori subgroups, we observed that our method and the hierarchical models both outperform separate subgroup‐specific models when all subgroups have the same dose‐efficacy and dose‐toxicity curves. However, our method outperforms hierarchical models when one subgroup has a different dose‐efficacy or dose‐toxicity curve from the other three subgroups.

Bridging across patient subgroups in phase I oncology trials that incorporate animal data

In this paper, we develop a general Bayesian hierarchical model for bridging across patient subgroups in phase I oncology trials, for which preliminary information about the dose-toxicity relationship can be drawn from animal studies. Parameters that re-scale the doses to adjust for intrinsic differences in toxicity, either between animals and humans or between human subgroups, are introduced to each dose-toxicity model. Appropriate priors are specified for these scaling parameters, which capture the magnitude of uncertainty surrounding the animal-to-human translation and bridging assumption. After mapping data onto a common, 'average' human dosing scale, human dose-toxicity parameters are assumed to be exchangeable either with the standardised, animal study-specific parameters, or between themselves across human subgroups. Random-effects distributions are distinguished by different covariance matrices that reflect the between-study heterogeneity in animals and humans. Possibility of non-exchangeability is allowed to avoid inferences for extreme subgroups being overly influenced by their complementary data. We illustrate the proposed approach with hypothetical examples, and use simulation to compare the operating characteristics of trials analysed using our Bayesian model with several alternatives. Numerical results show that the proposed approach yields robust inferences, even when data from multiple sources are inconsistent and/or the bridging assumptions are incorrect.

Bayesian Semi-parametric Design (BSD) for adaptive dose-finding with multiple strata

In the era of precision medicine, it is of increasing interest to consider multiple strata (e.g. indications, regions, or subgroups) within a single oncology dose-finding study when identifying the maximum tolerated dose (MTD). We propose two Bayesian semi-parametric designs (BSD) for dose-finding with multiple strata to allow for both adaptively dosing patients based on various toxicity profiles and efficient identification of the MTD for each stratum. We develop non-parametric priors based on the Dirichlet process to allow for a flexible prior distribution and negate the need for a pre-specified exchangeability parameter. The two BSD models are built under different prior beliefs of strata heterogeneity and allow for appropriate borrowing of information across similar strata. Simulation studies are performed to evaluate the BSD model performance by comparing it with existing methods, including the fully stratified, exchangeability, and exchangeability-non-exchangeability models. In general, our BSD models outperform the competing methods in correctly identifying the MTD for different strata and necessitate a smaller sample size to determine the MTD. The BSD models are robust to various heterogeneity assumptions and can be easily extended to other binary and time to event endpoints.

Dose-escalation strategies which use subgroup information

Dose-escalation trials commonly assume a homogeneous trial population to identify a single recommended dose of the experimental treatment for use in future trials. Wrongly assuming a homogeneous population can lead to a diluted treatment effect. Equally, exclusion of a subgroup that could in fact benefit from the treatment can cause a beneficial treatment effect to be missed. Accounting for a potential subgroup effect (ie, difference in reaction to the treatment between subgroups) in dose-escalation can increase the chance of finding the treatment to be efficacious in a larger patient population. A standard Bayesian model-based method of dose-escalation is extended to account for a subgroup effect by including covariates for subgroup membership in the dose-toxicity model. A stratified design performs well but uses available data inefficiently and makes no inferences concerning presence of a subgroup effect. A hypothesis test could potentially rectify this problem but the small sample sizes result in a low-powered test. As an alternative, the use of spike and slab priors for variable selection is proposed. This method continually assesses the presence of a subgroup effect, enabling efficient use of the available trial data throughout escalation and in identifying the recommended dose(s). A simulation study, based on real trial data, was conducted and this design was found to be both promising and feasible.