ABSTRACT
Introduction In the later years of my career in clinical pharmacology, I was trans-
ferred to a strategic job on a committee which oversaw early clinical development of drugs in humans. A big part of this job was managing the interactions of the biostatistics and data management organization with a bunch of ‘data-happy’ clinicians. This adjective ‘data-happy’ refers to medics who love to collect data and want someone to analyse it until, as they say, the data pleads for mercy. Most often it is the statistician involved who ends up pleading for an end to the analysis. Data seldom speaks for themselves; someone usually has to interpret them. It is beneficial when working with such ‘data-happy’ people to train them to perform such exploratory statistical analyses themselves. Such an action tends to cure their state of ‘data-happiness’ quite effectively. To clarify, clinical pharmacologists are paid, and in many cases earn
their higher educational degree, developing new markers of clinical activity. As with QTc (described in Chapter 8), these take on the attributes of alphabet soup, in most cases, with the addition of numbers where the pharmacologists run out of letters - for example, CRP, IL8, IL5, LDL, VLDL, VLDL1, VLDL2, etc. Unlike statisticians, in general, they do not seem to have thought to introduce Greek letters, instead they just add more letters and numbers. My personal belief is that this is because the word-processing software packages they most use make it difficult to use Greek letters.... In any event, the point of measuring such markers in humans, and
describing their behavior over time and relative to dose, is to detect the clinical effect of drugs on the body. This obviously is of great potential benefit. If one can measure such activity in the body, and if such activity is predictive of clinical outcomes (like stroke or myocardial infarction), then one could, in theory, predict the efficacy of drugs early in drug development! Even if it is not directly predictive, such knowledge should, in theory, allow one to improve understanding of how a drug works. Such knowledge of method of action is hoped to be beneficial. My ‘data-happy’ clinicians were always excited about such endpoints,
and often wondered why I was not. They usually put it down to, ‘Statisticians are just not interested in the science of such matters....’ In truth, I was interested, but after many years in clinical pharmacology, I had made a conscious decision not to get excited about (or too involved in) such ‘data-happy’ clinical things because: 1. There is a lot more involved in predicting clinical outcomes than just
showing that a marker is correlated to clinical outcome, and 2. One comes to realize that efficacy is all well and good, but safety comes
first (and, as we saw in Chapter 8, is an evolving topic). If one cannot find a safe and well-tolerated dose range (which is what early phase development is all about), then it really does not matter how efficacious the drug is. In my experience, most drugs fail in drug development because one cannot achieve a dose that is high enough such that the drug works without untoward side-effects, not because the drug does not work. All this said, evaluation of drug efficacy and method of action data is
an important part of clinical pharmacology, and this chapter will cover some methods for modelling the behavior of such data. We first briefly review some topics related to nomenclature, assumptions, and the statistics employed for this purpose.
