The role of a statistician in drug development: Pre-clinical studies

Before a novel compound is evaluated in human trials, preliminary characterisation of the toxicology profile as well as the efficacy and drug metabolism should be assessed in pre-clinical studies. This can be performed using basic techniques or more advanced strategies such as pharmacokinetic-pharmacodynamic (PK-PD) modelling. The findings may be used to predict how humans would react to the drug, as the disease progression and the PK-PD relationship in humans could be consistent, to some extent, with animal study results. Furthermore, data accumulated from animal experiments that have been well designed, conducted and analysed will contribute to a more successful translation into humans. Innovative study designs and sophisticated statistical methods can therefore potentially improve the efficiency and accelerate the development of pharmaceuticals. However, recent reports (Roberts et al., 2002; Hackam and Redelmeier, 2006; American Council for Science and Health, 2006) show that the quality of animal experiments in general is dishearteningly unsatisfactory. Criticism also involves the deficiency of applying statistical approaches adequately and properly in pre-clinical research. These issues arise fundamentally due to the scant awareness that statisticians can play a significant role in this stage of early drug development.

A general introduction of how the followings questions are answered can be found here: General Introduction on Drug Development

 

1. What are the responsibilities of a pre-clinical statistician in this phase?

Role of Statistician

Pre-clinical statisticians give statistical support to in-vitro and in-vivo studies (i.e. studies performed with cells and animals respectively). The responsibilities of a statistician in this phase are varied: they can involve data analysis of toxicological studies, assay validation, synergy calculation, biomarker exploration, planning and analysis of dose finding studies and involvement in pharmacogenomics, pharmacokinetic and pharmacodynamic modelling. As in other phases, the pre-clinical statistician is involved in the design, power calculation, data management and analysis of studies, communication of results and report writing. Nevertheless, the role of pre-clinical statisticians is more “pioneering” (Lendrem, 2002), as the pre-clinical phase of drug development is characterized by poor knowledge of a compound’s mechanism of action and of new scientific technologies. Therefore, the inherent innovative nature of this phase also requires the development of new statistical methods, which is another important role of the pre-clinical statistician.

 

2. With whom do pre-clinical statisticians collaborate?

Pre-clinical statisticians mostly collaborate with pre-clinical scientists (project managers, laboratory heads, laboratory assistants, toxicologists, computational biologists etc.) in order to address statistical questions and communicate results, and with IT specialists in activities such as data generation and storage. Statisticians involved in the pre-clinical phase also work with later stage scientists and statisticians to assess translatability of the results.

 

3. What are the major challenges emerging from these collaborations?

 

challenges-2Pre-clinical statisticians sometimes struggle to demonstrate the importance and value of statistics to pre-clinical scientists. It is often difficult to work in an interdisciplinary working environment and to reach a common goal starting from different scientific backgrounds. Some pre-clinical scientists have limited statistical knowledge and conduct trials without adequately taking statistics into consideration. Therefore, statisticians are sometimes not involved, or they are not involved early enough to prevent statistical errors. Even in situations where pre-clinical statisticians are involved, pre-clinical scientists often require analyses which are too much focused on p-values and testing (Wasserstein and Lazar, 2016), whereas the true scientific question is about estimation.

 

4. What are major ethical challenges in this phase?

Ethical regulation in the pre-clinical phase depends on the sub-sector: for example, toxicological studies are heavily regulated, whilst studies which concern drug discovery (i.e., the earliest phase of drug development) are characterized by fewer regulations, as this allows the freedom to undertake more exploratory analyses. However, the “good ethical practices” and “good laboratory practices” still need to be followed and an ethics committee has to check and approve any animal study. The ethical issues concern finding a good trade-off between small sample sizes for animal studies and obtaining meaningful results. To save animal lives and resources, the unnecessary use of animals should be avoided. Animal data should be analysed as effectively as possible, in order to prevent the failure of compounds only in later phases. This requires sometimes creative and innovative statistical analyses which are very different from those in e.g. large phase 3 studies.

 

5. What are the major statistical challenges in this phase?
challenges-1

Unlike late phase studies, pre-clinical studies are characterized by small sample sizes. The importance of pre-clinical statisticians in this phase lies in drawing the big picture to answer a complex scientific question from small experiments. The difficulties are the distributional assumptions, many confounding variables and unknown sources of variability (these are unverifiable in small sample sizes). In such context, all additional sources of information (e.g. historical controls, additional/external databases) need to be combined, cleaned and corrected in order to support and assess the current experiment. Moreover, statistical models should be able to have also a biological (pharmacological), not only a statistical, meaningfulness. Strategies such as complex Bayesian (hierarchical) models and equivalence methods have great potential in this phase. Robust statistical methods (e.g. for microbiome data analysis), mixed effects models for excess zero/sparse data (data with a majority of cells with zero counts) and process validation are important as well. Often, a specific statistical approach has to be developed in a short period of time and improper post-hoc- and subgroup analyses have to be avoided.

 

6. Which innovative study designs are particularly important in this phase?

studydesign3There is no ‘one size fits all’ approach to the design of pre-clinical studies. Instead, the pre-clinical study must be tailored to the investigational agent and the proposed experiment. The study designs particularly used in this phase are: crossover design with repeated measurements, Latin square designs, (fractional) factorial designs, sequential or adaptive designs, in silico models, response surface models, D- and I-optimal designs or a mixture of designs. Analysis often involves in-silico models and response surface models.

 

7. Which statistical analyses are most important in this phase?

As previously mentioned in this post, pre-clinical studies consist of in-vitro and in-vivo experiments in a laboratory. While it is mainly biologists who contribute to the in-vitro experiments, in-vivo experiments are the stage where statisticians get involved more often. A well planned in-vivo experiment should formulate the hypothesis unambiguously and be conducted according to appropriate study designs proposed by statisticians. Meanwhile, correct sample size determination is thought to be essential, and should be consistent with the 3Rs principle; that is, Replacement, Reduction, and Refinement. Statistical methods including Bayesian analysis, parametric survival models (Tobit regression), mixed models, non-parametric tests, and multiple testing procedures are considered to greatly facilitate the design and analysis of in-vivo experiments.

 

8. Usage of Bayesian methods at this stage:

Baayesian-3In general, Bayesian methods enable the possibilities of incorporating historical data in a formal way and offer flexibility to cope with complex models. Moreover, Bayesian inference that relies on Markov Chain Monte Carlo (MCMC) technique, compared with classical maximum likelihood estimation, can achieve greater precision when estimating model parameters. Most statisticians feel it is very promising and strongly encourage the use of Bayesian approaches in pre-clinical studies. Though it is believed that Bayesian methods can significantly enhance the quality of pre-clinical studies, statisticians are regretful to admit these advanced methods have not yet been applied widely.

 

9. What are the statistical topics that are particularly “hot” in this phase?

As statistical methods have not been adequately developed/applied in pre-clinical studies yet, there is room for statisticians to make a large contribution. Among a number of plausibly hot topics are 1) the use of Bayesian adaptive methods for experimental designs, 2) refining the quantification of risk using tolerance interval techniques, 3) statistical modelling and hypothesis testing for high-dimensional multivariate data, which are not necessarily following normal distribution but can be, for example, dirichlet, in the case of microbiome, 4) modelling of longitudinal data with time-varying factors, and 5) bioassay validation.

 

10. Are new methods used regularly in practice?

Since the pre-clinical phase is not as highly regulated as clinical trials, the framework in which methodologies are chosen can be loose. This means that the applied strategies can be modified to a certain degree. Nevertheless, substantial changes of statistical methodologies will be performed only rarely, as, unfortunately, new methods are not usually used in this phase. Moreover, it is common that the investigator and the clinical team are reluctant to use innovative statistical methods. Another consideration for the statistician in the pre-clinical phase, is that new methods often need some time to be assessed in practice with regards to their advantages and limitations. Besides that, new methods do not always come with available validated software, or are too complex to be used in practice, or are not that robust. Therefore, new methods are usually applied in parallel with the established methods at the research level.

 

11. What is the connection between pre-clinical and Phase I studies?

preclinical to clinical

Statisticians working in the preclinical phase provide critical input into quantification of risk when moving from animal to human studies. Pre-clinical results evaluation leads to go/no go decisions about first-in-human studies.

 

 

 

 

 

“This blog was written by Eleni, Elvira, Fabiola and Haiyan”

pic1

 

References:

  1. Roberts I, Kwan I, Evans P, Haig S (2002). Does animal experimentation inform human healthcare? Observations from a systematic review of international animal experiments on fluid resuscitation. BMJ 324:474-6.
  2. Hackam DG, Redelmeier DA (2006). Translation of research evidence from animals to humans. JAMA 296:1731-2.
  3. American Council for Science and Health (2006). America’s war on carcinogens: Reassessing the use of animal tests to predict human cancer risk.
  4. Lendrem DW (2002). Statistical support to non-clinical. Pharmaceutical Statistics 1(2):71-73.
  5. Wasserstein RL, Lazar NA (2016). The ASA’s statement on p-values: context, process, and purpose. The American Statistician 70(2):129-133.

 

Leave a Reply

You must be logged in to post a comment.