Dx Launch Risks
This case study shows how a risk management strategy for possible recalls could prevent disruption, not only to a biomarker's market progress, but to ...
Aiden Flynn, Exploristics, highlights how early engagement with statisticians in personalized medicine planning would enable the identification of many more biomarkers with clinical utility and reveal the revenue opportunity for a particular therapy.
As a statistician, I believe I can add most value to an experiment during the planning phase and before any data are collected by ensuring the design offers the best chance of meeting the study objectives. Therefore, it is frustrating that statisticians are regularly ignored during the planning of a study and only get involved once the data have been collected. At this stage, it is often too late to salvage the experiment and it becomes an exercise in sifting through the detritus looking for remnants of information. Whilst statisticians can make retrospective analysis look like a well-executed, professional undertaking by producing some pretty pictures, this can amount to giving a pig a full makeover.
The concept of proper planning is particularly relevant in personalized medicine research because much of the focus to date has been on sample collection, the generation of huge volumes of biomarker data and the analysis of these data rather than the prospective design of the experiments. Whilst it is possible to identify some useful biomarkers from a retrospective analysis (e.g., KRAS), there is no doubt that retrospective analysis in poorly-designed experiments suffers from a very low chance of success. At Exploristics, we have shown that it is possible to design exploratory personalized medicine studies to achieve a high probability of success. We have developed and applied a novel approach, based on computer simulation, to optimize studies designed to meet multiple objectives, such as the investigation of efficacy in the entire study population, as well as in a subgroup. In one application, we estimated the probability of success in identifying a genetically defined subgroup to range from zero, for an entirely retrospective approach, to more than 60 per cent for a more prospective approach. This can be achieved by smarter stratification and allocation of subjects within and between treatment groups and by selecting a more appropriate method for analyzing the data, all without compromising the likelihood of success for the other objectives or adding to the cost of the study.
The early engagement of statisticians and the application of novel study designs will shift the likelihood of success to a level that will enable the identification of many more biomarkers with clinical utility, lead to a fold-change in the return on investment, create an irrefutable momentum that will change the personalized medicine cultural landscape and, consequently, will ensure that personalized medicine really is an integral part of drug development planning process. Personalized medicine would greatly benefit from a statistical makeover, but one with real substance – if it involves a pig, it’s too late.