Sir

In his Commentary (Nature 454, 692–693; 2008), Donald Berry warns about the dangers of poor statistical understanding and misinterpretation of drug-testing results in Olympic athletes. Unfortunately, this same problem arises on a daily basis around the world in medical clinics, often with even greater consequences.

Berry illuminates the failure to use proper Bayesian reasoning in interpreting doping tests and also the problem of not having sufficient control-population norms for the tests to determine correctly whether an athlete is taking a banned substance or not. Clinicians typically have less understanding of Bayesian statistics than drug-testing officials and even fewer resources to interpret or norm such tests.

Take urine testing of patients on opiate therapy to make sure that they test positive for opiates (to show the patient is taking the medicine rather than, say, selling it) and that they are not using illegal drugs. Either a negative test for opiates or a positive test for an illegal substance can typically be sufficient to preclude a patient from receiving another prescription for opiates or to put the clinician in the position of having to explain the test result before prescribing the medicine.

Such tests need to be reported with the appropriate Bayesian interpretation. Also, as Berry advocates for Olympic athletes, patients should have the right (and access) to a statistical 'consultation' if they feel the test is in error.

See also: Doping: a paradigm shift has taken place in testing Doping: probability that testing doesn't tell us anything new Doping: ignorance of basic statistics is all too common