Selected Articles from the Nonclinical Biostatistics Conference 2021

We are pleased to present a special section of Statistics in Biopharmaceutical Research, consisting of three papers developed from material presented at the Nonclinical Biostatistics Conference of 2021 (NCB21). We are excited to call your attention to this exciting work; our summary here expands that of Kolassa and Pickering (2022).

Pourmohamad and Wang (2023) discuss methodology for sample size reduction. Efficient use of empirical information has been a concern since the dawn of statistics as a discipline. The importance of this concern is clear when datasets are constructed through costly and time-consuming experimentation. Efficient use of samples is a moral imperative when units of experiments are animals.

Bayesian approaches to data analysis are appealing for preclinical studies, in that decisions based on study results are generally internal to a single organization, avoiding difficulties in finding priors agreeable to all stakeholders. Bayesian approaches often allow for more precise conclusions, although the questions that are answered are different from those of a frequentist analysis.

Analysis approaches applied sequentially have the potential to allow for early termination of a study if the initial evidence is strong. This has the advantage of saving experimental animals in such cases of strong initial evidence. Frequentist analyses applied sequentially exhibit difficulties that are substantially removed in a Bayesian analysis.

Pourmohamad and Wang (2023) exhibit the combination of these important ideas, using Bayes factors for making decisions, and demonstrate how Bayesian updating allows for the sequential analysis of data. They apply their methods to simulated data of a sort commonly occurring in nonclinical animal studies. They reference computational tools contained in an R package, and demonstrate a significant savings in animals over a non-sequential frequentist analysis. Frequentist operating characteristics are also investigated.

Feng and Baumgartner (2023) investigate various semiparametric data analysis tools. The use of data analytic tools allowing for relationships of various complicated shapes is an important advance in modern statistics. These tools are used both for prediction and for classification. This article focuses on kernel-based methods, and their relation to tree-based methods.

Performance of various procedures are compared, via simulation, for various conventional feature configurations. Kernel methods are demonstrated to perform very well. Various methods compared are also illustrated on some real benchmark datasets, including a high-dimensional example. Applications to random forests are considered. Investigations are motivated using Fisher’s Iris data, and kernel techniques were compared via 14 benchmark datasets, and other tree-based methods are considered.

Hsieh, Chang, and Barron (2023) investigate the problem of bioassay. This bioassay problem consists in estimating the dose associated with a certain proportion of the maximal response or effect; often this proportion is one half. This is important for determining an optimal dose for the drug in question.

Hsieh, Chang, and Barron (2023) consider cases in which standard monotonic models do not fit. Such non-monotonic relationships are called “biphasic”. Nonmonotonicity is particularly problematic in the case of data amalgamated from a variety of sources. The authors note that most common parametric models do not allow for such an effect, address this problem by employing a larger model that adds non-monotonicity with additional parameters.

The problem is addressed by modeling the increasing and decreasing parts of the dose–response curve separately, and explicitly accounting for the process of switching between phases. Various parametric models are considered, and new switch-point models are proposed. An example using flow cytometry data is introduced to exhibit these concepts, and extension to more complicated data relations, such as robust regression, are discussed.

We are excited to present these important advances in preclinical experimentation, tree techniques, and bioassay.

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.