February 14th 2018. Pitfalls in precision medicine: How do we know individuals differ in their response to a health intervention?


Prof Geg Atkinson, Teeside University, Health & Social Care Institute


Personalised medicine is relevant to any researcher who is interested in quantifying individual differences in the response to an intervention or treatment and identifying any individual moderators or mediators of that response.  In this seminar, it is demonstrated how popular plots used to present individual differences in intervention response can be contaminated by random within-subject variation and the regression to the mean artefact1,2. Large individual differences in the response to an intervention can be suggested by some common plots and analyses, even when the true magnitude of response is exactly the same in all individuals. These flaws have translated to incorrect inferences in important previous trials (e.g. the HERITAGE study on individual exercise response heterogeneity3), as well as in governmental guidelines on individual weight loss objectives4. The appropriate designs (e.g. replicated crossover studies5,6) and analysis approaches for quantifying this true inter-individual variation in treatment/intervention response are then presented. It is imperative to analyse the data from a control group/condition (or derive information from a prior relevant repeatability study). The most important statistical comparison is the standard deviation (SD) of changes in the intervention arm vs the same SD derived from the control arm or relevant repeatability study. Only if the difference between these SDs is clinically relevant is it logical to explore any moderators or mediators of the intervention effect, e.g. phenotypes or genotypes. When the SDs are similar, true individual response differences are clinically unimportant and further analysis is unwarranted and potentially unethical.


Phil Tucker