Furthermore, for a surrogate to be valid, it should also be possible to use the effect of the treatment on the surrogate to estimate the amount of clinical benefit a patient will experience. Eikelboom et al1 have again presented correlations to suggest that such an estimate can be derived and, again, it is not clear that these correlations are sufficiently large. However, even large correlations are not sufficient to demonstrate the validity of a surrogate, because even a perfect correlation between the treatment-induced effects on a potential surrogate and a patient-oriented outcome does not guarantee the utility of that potential surrogate.4 Further, summary metaanalytic data, such as those reported by Eikelboom et al,1 can be misleading. Data that allow for a correct metaanalytic interpretation of the validity of surrogate measures do not appear to be available, but it is worth noting that the desired analysis would involve regressing the effect of an intervention on PE, and also regressing the effect of that intervention on DVT for each trial. One would treat the coefficients from these two regressions as random variables with a joint distribution over all trials. The estimated parameters from this joint distribution would then be used to predict mean PE rates from the expected DVT rates.5 A correct interpretation of the results of single studies requires similar analytic considerations.