Precision Medicine Approaches and the Health of Populations: Study Design Concerns and Considerations

Biomedical advances in the past decade have aimed to capitalize on two movements that have dominated the research conversation: precision medicine and the ascent of big data. These emerging shifts have resulted in growing confidence that we can better characterize health, predict who will get ill and with what, develop new treatments which exploit genetic, metabolic, and other vulnerabilities in cancers and infectious agents, and tailor some of these treatments to match characteristics of the individual patient and their specific disease. However, we suggest that there are important cautions. Weaknesses in the data and the methods used to study them raise three potential concerns. First, any data collected, and analysis attempted, will have limited utility absent internal validity, unless fundamental issues of accurate and consistent measurement can be addressed. Second, lack of attention to external validity limits generalizability beyond the narrow (even if large) samples in hand, so that the utility of inference that can emerge from these approaches remains limited. Third, the proposed approaches seldom include consideration of ubiquitous forces that can determine whether observed associations are truly attributable to the innovation or to other, unmeasured forces. This essay discusses these limitations and explores how they can influence inference drawn from big data precision medicine science.