Abstract
When I was a medical student (this was way back in the days that non-consultant medics were called junior doctors and both they and medical students wore those lovely bug-infested white coats outside of laboratories), I can remember having a copy of the first Oxford Handbook of Clinical Medicine. It was very handy as it nicely fitted into the white coat pockets. Since then, the series has grown to over 50 books and even includes the Oxford Handbook for the Foundation Programme!
Given the greater number of individuals having problems with maths and statistics, there is an increased requirement in health care for concise and easily available sources that cover the whole breath of medical statistics in an easily readable form. The format of the Oxford Handbook series lends itself quite nicely to this aim, although this book is unlikely to be carried around, even in the pockets of lab white coats. The inside cover has a very useful table that serves as a quick and simple reference text for what statistical test to use depending on the scenario in question. Another very helpful concept is the use of a white exclamation mark in a black circle to highlight points where readers must be cautious and also a black filled-in triangle to highlight important points.
What about specifics that might be of more interest to those in the labs? Yes, it does mention Bland–Altman and correctly states that Bland–Altman plots assess clinical and not statistical agreement between two methods. While there is mention of log transformation, there is no information on the more useful relative (or percentage) difference plots. Having said this, the text correctly points out the problem with the use of the correlation coefficient in assessing agreement but does give the impression that linear regression and correlation should not be performed in method comparison studies. In another section, least-squares regression is covered but not Deming's or Passing–Bablok regression. Thus, the whole opportunity for a discussion about different types of residual data and the type of error in data is missed. The concept of confidence intervals for ‘r’ is appropriately mentioned, but the formula is not provided despite adequate space remaining. The F ratio for the analysis of variance is mentioned, but unfortunately is in a totally different section – to access this section, one would have to be aware of the concept as the text does not point the reader towards the correct section.
Finally, the chapter on diagnostic studies surprisingly does not mention the STARD initiative. I think that it would also have benefited from a sentence or two on the difference between analytical and clinical sensitivity and specificity in view of the increasing number of ‘high sensitivity’ assays available. Otherwise, the chapter is quite comprehensive and clearly shows the effect prevalence has on the performance of a test.
So, overall, the Oxford Handbook concept still has its uses but the Handbook on Clinical Statistics sadly lacks important and basic information. Hopefully, the next edition will be better?