Dr. Sigurdsson would like clarification of the covariates that made the effect of retinol significant in the multivariate analysis in the bone mineral density study. He would also like to see the results after adjustment for vitamin D intake.

We tried to include the same set of covariates in both the hip fracture and bone mineral density studies (established osteoporosis risk factors). No major confounding was observed in the hip fracture study, but in the linear regression analysis of the bone density study, inclusion of possible confounders did change our estimates. Energy intake was the covariate with the most pronounced effect on the ß estimates in the latter analysis. The bone density study included a more heterogeneous sample of women with regard to age and menopausal status. When we restricted the analysis of retinol intake to a dichotomous variable (> or < 1.5 mg/d), no major differences were seen between univariate and multivariate analysis (Table 5 of our article).

Dr. Sigurdsson argues that because the correlation between the intake of vitamin A and intake of vitamin D is usually 0.8 to 0.9, our result may be explained by a high vitamin D intake. The correlation coefficient between these two nutrients in the bone mineral density study was 0.4. In the last sentences of the Results section, we note that adjustment for intake of several other nutrients does not significantly change the association of retinol intake and bone mineral density. In fact, the ß-coefficients shown do include adjustment for vitamin D intake. Unfortunately, we did not include this information in the final version of the article. We thank Dr. Sigurdsson for drawing our attention to this.

With regard to the fracture study, Dr. Sigurdsson states that the main finding is based on the group with an intake greater than 1.5 mg/d, which must have contained fewer than 10 case-patients. Table 3 clearly indicates that this calculation is incorrect. The true numbers are 33 case-patients and 66 controls.