The hypothesis underpinning the biomedical passport is as simple as obvious: Normal blood values vary between sportsmen. This requires limits that are specific for each individual person. How to establish these limits? By a regular check of blood values outside competition. How to trace a sportsman outside competition? We need to know her/his “whereabouts”. For the sportsman there is no other possibility: “Accept or Stop your sport!”
“Now I know, now I know...” thinks the UCI. But do they realize that Brassens' lyric continues with: ….now I know that I will never know”? Remember, hundred percent certainty does not exist in science.
The BMP is nothing else than a simple plot of a blood (or urine) value as a function of time. In this plot two important lines are constructed. A line indicating the mean value (the m-line) and a line at a certain distance above (or below) the mean indicating the upper (or lower) limit (the Lu-line) under “normal conditions”. So far so good. Exceeding the upper limit means an “abnormal” value and indicates the usage of doping until convincing proof of the contrary. Still good? It depends.
Let us consider the upper limit (Lu) which is given by: Lu = m + k.s (m is the mean of a series of observations and s is the standard deviation of that series).
As we can see we need to quantify three parameters: m, s and k! So, there is enough room for mistakes. First we should realize that m and s are estimates of the true but unknown mean and standard deviation. The less observations we make the less precise these estimates are. A rule of thumb often used is that we should perform 20 observations at least before positioning the m-line in the plot. However, these observations should be made under “normal” and stationary conditions. This means that we assume that the mean and standard deviation do not vary in time.
Second, we have to select the value k. For that we need to consider the decision to be made when an observation exceeds the limit. In quality control, control charts are already used for a long time. Here, the decision is that the exceeding observation indicates that the 'system' may be not-normal (they say: out-of-control). In other words, the system is not any longer in the “normal” state. Either the mean or the standard deviation has changed. This leads to an inspection for the cause of that deviation. If the value k is chosen too low, the system is often unnecessarily inspected. This is a false alarm. If the value k is too high, products may be produced under abnormal conditions leading to out-of-spec products. This is the result of a missed abnormal situation.
In order to evaluate the probabilities of a false alarm (in the context of doping this is a false accusation) and a missed alarm (in the context of doping this is a missed case of doping) we need to assume the probability density function (pdf) of the observations. In quality control this is done by plotting the mean of a (small) number of observations instead of plotting the individual observations. Distributions of these means are well-known and independent of the distribution of the individual values. However, in doping control not the means but the individual values are plotted. In that case the distribution is unknown! It may have a very long tail (skewed) or be strongly asymmetric. This is a real problem, asking for caution. It is certainly not allowed to take a value equal to 2.5 or 3 and attribute a risk of a false accusation of 0.5% or less.
So my message to every sportsman is: if you are falsely accused for doping, put your finger on possible flaws of your Biomedical Passport:
-
how and when was assessed that your “normal”values were stationary, and the collection of observations to calculate the mean and standard deviation could begin: probably you will never get the answer, as this maybe the major flaw of the BMP.
-
on how many observations are mean and standard deviation based: an answer less than 20 is unacceptable
-
which distribution of your values was assumed to decide on the value of k. An answer “normal distribution” is not acceptable.
-
If Lu derived from your BMP is more critical than the accepted value for the population of sportsmen (or any individual), the above points need special attention.
But be aware, despite these difficulties the BMP time plot is a powerful tool, if correctly applied. It may also suspect doping even when your values are well below Lu. For instance, it is highly unlikely that several values in a row are above (or below) the mean. The values should remain randomly distributed (correlation between two successive observations should be small). I will discuss this in a next post.