For these types of projects we have developed a structure approach involving multiple steps:
Summary statistics can often mask the existence of patient subgroups. As a start, we like to visualise patient level data to show variability in response, to identify subjects with clinically meaningful responses and to evaluate the proportion of patients with clinically meaningful responses.
Prior to generating any biomarker data, it is useful to look for evidence of hidden structures within the data. Without such evidence it is unlikely to identify a biomarker that can classify patients into subgroups. We use methods such as principal components analysis and latent class modelling to latent classes (subgroups) in the data. These approaches also give insights into the properties of the subgroups in terms of proportions and treatment effects.
Where there is evidence of subgroups, it is useful to evaluate other possible explanatory variables before embarking on an expensive biomarker study. Indeed, there may be other factors that can explain the existence of subgroups in the response data. These factors are often as important as biomarkers in stratified medicine research.
Quantifying the likelihood of statistically identifying a subgroup in a retrospective analysis is a useful exercise as it supports go/no go decisions and helps to prioritise use of resources. We utilise simulation approaches to evaluate the ability to detect subgroups given the study design, availability of samples, data variability and a range of scenarios representing subgroup characteristics. Often the power to identify subgroups is so low that the study is not feasible or that the data need to be pooled across studies to improve the power.
Our experience enables a very pragmatic approach to the integration of stratified medicine into clinical development programmes. We are able to make help focus efforts on opportunities that are most likely to succeed. We are also able to ensure stratified medicine studies have the best chance of success through the prospective design of individual studies and research programmes.
We have developed our own proprietary modelling and simulation platform specifically for the purpose of evaluating stratified medicine scenarios. We have applied this platform for numerous clients for training purposes, to help them quantify the likelihood of success and to optimise the design of new studies.
Our experience has taught us that it is best to be open about the risks our clients are taking when they make decisions. Our clients appreciate our honesty when we advise them not to waste their time and money by pursuing opportunities that are not likely to proceed. It may seem strange that we would want to turn away potential revenue but we believe that those clients will come back to us again soon.
Our structured approach to evaluating opportunities for stratified medicine has broad utility to other similar applications. We have used the same approaches on multiple projects supporting the rescue of failed treatments and the identification of new indications for existing drugs. These projects have involved both clinical trial data and real world evidence.