Methods for identifying cases and estimating person time at risk must be detailed

EDITOR—Lawrenson et al reported a crude incidence of venous thromboembolism of 38 per 100 000 women who used combined oral contraceptives, based on information derived from the General Practice Research Database, without describing how they derived their estimate.1 This estimate is closely similar to the incidence reported in a prior publication of theirs, which was based on a different automated medical database.2 In their letter they provide no details on the validity and specificity of the diagnosis of venous thromboembolism or the presence of medical risk factors and no information on the person time at risk for current users of oral contraceptives.

An incidence must be based on person time at risk. Despite apparent deficiencies, the authors provide an incidence that is roughly twice as high as the incidence that colleagues and I reported in a paper derived from the General Practice Research Database; we included only idiopathic cases of venous thromboembolism (those possibly related directly to use of oral contraceptives).3 Lawrenson et al conclude that our reported estimate was substantially lower than the true incidence.

On the basis of their previous publication,2 Lawrenson et al seem to have included both outpatients and inpatients with any one of five computer recorded diagnoses as patients with venous thromboembolism, without documentation from clinical records. The five diagnoses include a computer recorded diagnosis of thrombophlebitis. Colleagues and I found that, on the General Practice Research Database, over 90% of subjects with this computer recorded …