Despite the appeal of personalized medicine (that is treatment selection based on the presence of a particular marker), uncertainty remains regarding the broad utility of this selection strategy in oncology. A recent meta-analysis by Jardim et al. in the Journal of the National Cancer Institute attempted to provide some clarity by comparing efficacy outcomes between personalized and non-personalized clinical trial designs leading to the new FDA approval of drugs between 1998-2013. The publication concluded that using a biomarker-based selection strategy led to improved response rate, progression free survival and overall survival across a range of cancer subtypes and selection biomarkers.
The study should be applauded for its unique approach in trying to determine the benefit of personalized drug development, the paper’s conclusions are qualified by five issues.
No information about drugs that do not receive license
The study only evaluated efficacy outcomes for trials directly leading to the FDA approval of the drug (the authors acknowledge this). This may prevent generalizability of conclusions, as it does not capture drugs that failed during testing. However this search strategy also excluded studies earlier in the development of approved drugs, where they were explored unsuccessfully for various indications or biomarker subgroups. In contrast to FDA approval for non-personalized drugs, which just requires identifying the proper indication, personalized strategies in addition require finding optimal test conditions for biomarkers used in patient selection. It is therefore conceivable that greater failed exploration goes into the development of a personalized strategy and therefore that an overall comparison of efficacy outcomes between personalized and non-personalized designs may not reach the same conclusions as a comparison of the FDA approval trials.
Doesn’t address dangers of premature biomarker enrichment
While there may indeed be a benefit to using biomarker-based trial designs, the study does not encompass potential harm that can arise when trials prematurely enrich for a particular biomarker population. Early enrichment precludes evaluating the drug in biomarker ‘negative’ patients and can prolong uncertainty regarding a drug’s utility in biomarker negative groups. The approval of Trastuzumab for HER2+ breast cancer provides an example of this. The two clinical trials leading to the FDA approval of the drug were based on a personalized strategy, but now nearly 20 years later the biomarker originally used for patient selection is being reevaluated in a large-scale phase 3 study.
May not properly classify “personalized therapy”
A third issue concerns the authors’ classification of “personalized therapy”. The paper’s definition includes both trials selecting patients who express rare biomarkers along with studies in which at least 50% of the patient population is known to harbor the mutation (in the study just over half of the personalized trials fell into the latter category, with a number of those including markers present in nearly 100% of the patients). While a biomarker is implicated in the response to therapy in both situations, comparing these two groups may not be appropriate. As there was no selection process needed to identify patients from the overall population to include in 50% criteria trials, they more appropriately reflects a “population based-” rather than a “personalized-” medicine. One of the most pressing issues in developing personalized treatments is grappling with properly selecting the patients who have increased chance of benefit. It is conceivable that the risk/benefit of personalized trials using low frequency mutations (which requires applying often complex selection criteria to identify the proper population) may not be comparable to the “population” marker trials.
Doesn’t quantify clinical benefit post-approval
Another issue not addressed in their conclusion is the actual clinical impact of biomarker-based treatment selection once a treatment has been approved. There is general concern over the current unbalanced cost/benefit of drug development and as many biomarkers exist in low frequencies in the population it is conceivable that the net benefit of drugs approved based on personalized strategies is lower than that of non-personalized strategies – or that the impact of drugs approved based on the 50% criteria is greater than that of other biomarker-based drugs. It is therefore unclear whether a biomarker-based study design is just better for getting drugs approved, or better for getting better drugs approved.
May not predict the future of personalized medicine
Finally, several commentators have noted that large scale trials (such as those evaluated in this study) may not be sustainable for the future of personalized medicine drug development. There is a growing trend in the use of next generation clinical trials, which include N of 1 trials, basket designs and adaptive treatment allocation. Each of these enroll small populations because the frequency of patients expressing the biomarkers of interest is generally very low, and therefore one should be cautious in extrapolating the methods and conclusions of the publication (especially due to the inclusion of “population” markers) to future evaluations of the efficacy of personalized medicine.
While not complete, this publication is the first step in a much-needed rigorous evaluation of utility of biomarker-based strategies in cancer treatment and drug development.