As Dr. Murphy points out, the role of smoking in VTE is controversial. In our study, smoking overall was a risk factor for VTE (hazard ratio, 1.7 [95% CI, 1.2 to 2.2]) after adjustment for sex, factor V Leiden, body mass index, myocardial infarction, physical activity, estrogen-containing preparations, menopause, and year of study entry. Because smoking did not interact with factor V Leiden genotype or estrogen-containing preparations to affect risk for VTE, our data can suggest only additive effects of these 3 factors on VTE risk. However, because of limited statistical power, we are unable to provide accurate risk estimates for smoking as a risk factor for VTE within small subgroups such as factor V Leiden heterozygotes and homozygotes, and certainly not when these groups are further stratified for sex and use of estrogen-containing preparations. We agree with Drs. Vandenbroucke and Rosendaal that this does not mean that oral contraceptives do not play a role in VTE, since this association has been convincingly demonstrated in several previous studies.

We also agree with Drs. Vandenbroucke and Rosendaal that most previous case–control studies have used more stringent diagnostic criteria for VTE than we were able to do. However, end point misclassification in the Copenhagen City Heart Study is unlikely to explain the difference in risk estimates: Assuming a diagnostic false-positive rate of 5 events per 10 000 person-years (that is, incorrectly diagnosed VTE) and a sensitivity of 0.6 (that is, only 60% of VTE cases diagnosed), the true hazard ratio for VTE in factor V Leiden heterozygotes versus noncarriers would be only 4.3.

There are many other examples in which genetic risk estimates are higher in case–control studies than in prospective studies of the general population. In a study of hospital case-patients with hereditary hemochromatosis, 83% were homozygous for Cys282Tyr of the HFE gene (1). However, when we screened patients in the Copenhagen City Heart Study and identified 23 Cys282Tyr homozygotes, all had biochemical signs of iron overload but none had hereditary hemochromatosis (2). The most likely explanation for such discrepancies is ascertainment bias, that is, the fact that hospital case-patients have a more severe phenotype than the average case-patient in the general population and often are selected particularly to identify genetic risk. Therefore, case–control studies will often overestimate disease risk due to genetic factors in the general population. Consequently, advice on genetic risk factors intended for the lay public should be based on results from prospective studies of the general population.