Intended for healthcare professionals

Rapid response to:


Applying results of randomised trials to patients

BMJ 1998; 317 doi: (Published 22 August 1998) Cite this as: BMJ 1998;317:537

Rapid Response:

The mystery of an "overall main effect"

In their reply to Senn, Guyatt et al. have identified another source of variation in the response to a treatment -- the "overall main effect of treatment" -- which supposedly is something quite different from the second and third sources in Senn's list (within-patient and between-patient heterogeneity of response).

Assuming, for a moment, that I am holding an indeterministic (stochastic) view of causation, where the response of each patient to the treatment in question (at a given point of time) is a random variable that has some expected value if the patient were treated, and possibly a different expected value if the patient were not. These expected values might vary from patient to patient, depending on (possibly unknown) numerous patient characteristics and the causal propensity of the treatment in each patient. For example, the causal propensity might be zero for some patients, might produce an inevitable effect in others, or might be one of an infinite number of intermediate propensities in-between. It is also possible of course -- although not a logical must of Nature -- that some treatments have an identical causal propensity in all human beings (whether that propensity is zero or has a non-zero value).

Under this model of the universe (which is not that fictional for anyone who knows something about quantum mechanics), can Guyatt et al. tell me what exactly they have in mind when they speak, with some statistical confidence, about the "overall main effect of treatment"?

Competing interests: No competing interests

25 August 1998
Eyal Shahar
Associate Professor
Division of Epidemiology, University of Minnesota, 1300 South Second Street, Suite 300, Minneapolis,