Statistical issues in personalising medicine


There has been an explosion of studies in the field of pharmacogenetics in recent years and much has been made of the supposed potential for personalised medicine, whereby therapy would be tailored to the individual requirements of patients based upon genetic make-up. However, the practical obstacles are considerable. In fact the extent to which observed variation in clinical trials is due to genetic variation may have been overestimated due to a statistical fallacy[1]. What is clear is that traditional approaches to subgroups will be inadequate to address the issues of individualising therapy. It has been suggested that much more use should be made of repeated within patient cross-over studies in this field[2]. These could be the clinical trial equivalent of twin studies, much used in genetics[1, 3, 4]. What is clear, however, is that a more concerted attack on this problem, involving use of mixed models to identify components of variance but also decision analysis to establish logical approaches is required if progress is to be made.


Such repeated measures cross-over designs permit the various components of variation to be identified and thus allow patient by treatment interaction to be separated from pure within patient measurement error. Since patients differ by more than their genes, patient by treatment interaction provides an upper bound to gene by treatment interaction. Effectively, what the repeated cross-over provides is a form of whole genome matching. This will thus permit the identification of those disease treatment combination where the scope for targeted treatments is greatest[4].


However, this is only one strand of such an investigation and also important is a clear understanding of the economics of the situation. What does decision analysis say about the value of a theranostic approach? How prevalent does a sub-group have to be to make such an approach worth while, how different does the response to treatment have to be and what sensitivity and specificity is required from any test.




1. Senn, SJ. Individual response to treatment: is it a valid assumption?, British Medical Journal 2004; 329: 966-968.

2. Kalow, W, Tang, BK, Endrenyi, L. Hypothesis: comparisons of inter- and intra-individual variations can substitute for twin studies in drug research, Pharmacogenetics 1998; 8: 283-289.

3. Senn, SJ. Individual Therapy: New Dawn or False Dawn, Drug Information Journal 2001; 35: 1479-1494.

4. Senn, SJ, Statistical issues in Drug Development (second edition), Wiley: Hobokken, 2007.