ACQUIRED

qualitative differences between men and women were observed for many nutrients. A second reason for reporting associations of dietary nutrients by sex was to facilitate comparisons with the results of other studies (3). The use of the word "moderate" to characterize nuclear opacities at levels 2 and 3 of severity was intended to refer to intermediate levels between the two extremes which could be measured in this study, rather than to describe a clinically moderate degree of opacity. Opacities at these levels of severity are not expected to impair vision significantly. The use of four levels of severity, rather than two, as Dr. Milton suggests, permitted us to take advantage of information about dietary influences on earlier stages of nuclear opacification. When severity levels were dichotomized (level 3 or lower vs. higher than level 3), the odds ratios changed little. The assumption of the ordinal logistic regression model, that the relation of nutrition to severity of nuclear sclerosis is the same over all levels of severity, was tested using the proportional odds test. As we stated in our paper, no significant (p < 0.05) violations of the proportional odds assumption were observed. The definition of odds ratios in the paper was indeed erroneous. The sentence on page 324 (top of the second column) should read "The odds ratios represent the odds of being at or above a given level of opacity for a given quintile divided by the odds for the group in the lowest quintile." The use of the word "quintile" to reflect a region, rather than outpoints, is supported by a definition of this term which can be found in the International Encyclopedia of Statistics (4). Furthermore, this common usage permits a more straightforward discussion of results. The traditional approach of collapsing nutrient intake data into quintiles permits visual inspection for potential thresholds that may exist—that is, associations with only very high or very low levels of intake. While collapsing data on any exposure variable into categories carries with it a risk of losing an overall association caused by strong associations with very high or very low intakes, increasing the number of quantile categories used to the maximum allowable by the sample size minimizes this risk. Quintile classification is commonly used by many researchers in the field. Thus, an additional advantage of categorizing by quintiles is that it permits comparisons across studies with comparable quintile cutoffs.

qualitative differences between men and women were observed for many nutrients. A second reason for reporting associations of dietary nutrients by sex was to facilitate comparisons with the results of other studies (3).
The use of the word "moderate" to characterize nuclear opacities at levels 2 and 3 of severity was intended to refer to intermediate levels between the two extremes which could be measured in this study, rather than to describe a clinically moderate degree of opacity. Opacities at these levels of severity are not expected to impair vision significantly.
The use of four levels of severity, rather than two, as Dr. Milton suggests, permitted us to take advantage of information about dietary influences on earlier stages of nuclear opacification. When severity levels were dichotomized (level 3 or lower vs. higher than level 3), the odds ratios changed little. The assumption of the ordinal logistic regression model, that the relation of nutrition to severity of nuclear sclerosis is the same over all levels of severity, was tested using the proportional odds test. As we stated in our paper, no significant (p < 0.05) violations of the proportional odds assumption were observed.
The definition of odds ratios in the paper was indeed erroneous. The sentence on page 324 (top of the second column) should read "The odds ratios represent the odds of being at or above a given level of opacity for a given quintile divided by the odds for the group in the lowest quintile." The use of the word "quintile" to reflect a region, rather than outpoints, is supported by a definition of this term which can be found in the International Encyclopedia of Statistics (4). Furthermore, this common usage permits a more straightforward discussion of results. The traditional approach of collapsing nutrient intake data into quintiles permits visual inspection for potential thresholds that may exist-that is, associations with only very high or very low levels of intake. While collapsing data on any exposure variable into categories carries with it a risk of losing an overall association caused by strong associations with very high or very low intakes, increasing the number of quantile categories used to the maximum allowable by the sample size minimizes this risk. Quintile classification is commonly used by many researchers in the field. Thus, an additional advantage of categorizing by quintiles is that it permits comparisons across studies with comparable quintile cutoffs.
Dr. Milton notes that we did not include another study on this topic (5) in our Discussion. Results of our study (2) and that of Sperduto et al. (5) are complementary and contribute to the evidence that multinutrient supplementation may reduce the development of nuclear opacities. Our study showed that multivitamin use (6) and the intake of several nutrients 10 years in the past (2) were inversely associated with the severity of lens opacities in middle-aged and older adults. In the Linxian Cataract Study, a randomized, doublemasked clinical trial of multinutrient supplementation in a chronically malnourished Chinese population, nuclear cataracts were less common among older participants (65-74 years of age) taking multinutrient supplements than among those who had been taking a placebo for approximately 5 years (5). This evidence supports the likelihood that multinutrient supplements, rather than factors characteristic of supplement users, reduce the development of opacities in this region of the lens. Results of the study done in Beaver Dam, Wisconsin, suggest that this relation may extend to the relatively better nourished American rural population and to people under 65 years of age. Thus, consideration of the results of these two differently designed studies together enhances our understanding of the potentially protective nature of vitamin supplements on the development of nuclear lens opacities. This seems to be a good example of how interpretation of results from clinical trials and observational studies, considered together, can strengthen our overall understanding of relations between diet and disease.
Letters to the Editor participants were interviewed and examined twice per year in Chicago and Denver, but only once per year in San Francisco (1). Since the purpose of the study was to evaluate the medical conditions and symptoms associated with human immunodeficiency virus (HIV), the variations in the examination schedule should have been adjusted for prior to calculation of the incidence density. The numerator for this rate is influenced markedly by the examination schedule (3).
Second, the examinations were conducted by physicians in one study region and by physician assistants in the other two (1). Thus, some variation could be expected in the symptoms reported by the examiners. The authors did not discuss the potential for bias created by differences in the experience level of the examiners. Several textbooks and published articles have discussed this possible bias (4,5).
Third, it appears that the interval and conditions present before a man was first found to be HTV-seropositive were not used in the calculation of episodes and person-years of seronegativity. It has been suggested that each case in a case-control study should be eligible for inclusion as a control until the time of disease onset (6). No mention of this consideration is made in the Discussion.
The findings reported in the manuscript were used as inferential information on the development of HIV disease by Fessel (2). The above points were not considered in his interpretation of the data. We recommend that incidence density and incidence risk ratio be calculated for each study region individually rather than combining all of the regions; then a trend in each of the three regions can be determined.

Center of Veterinary Epidemiology and Animal Disease Surveillance Systems Department of Clinical Sciences College of Veterinary Medicine and Biomedical Sciences Colorado State University Fort Collins, CO 80523-1620 THE FIRST AUTHOR REPUES
We thank Salman et al. (1) for their comments on our paper (2). It is not surprising that there would be modest . variations reflecting practical problems in a prospective study examining more than 1,000 men over a 5-year period in three cities, and we listed those variations in the Materials and Methods of our paper. At one site, there were yearly, not biannual, examinations of men during the first 2 years of the study. However, men were asked about the occurrence of conditions in defined time periods. Thus, as in the Discussion, we were more concerned about recall biases than about (unavoidable) variations in interview scheduling. The physical examinations performed were external and limited; whether they were carried out by physicians or by others (who communicated with one another frequently), they were standardized, with examiners receiving special training in performing oral examinations. The long latency period of the virus and the difficulty of precisely defining dates of infection' limit the ability to count as controls persons who later seroconvert; thus, to keep our analysis and comparison as "clean" and defensible as possible, we did not include as controls men who later seroconverted. Finally, incidence densities and risk ratios were certainly calculated for each site independently, "but since these analyses did not show any evident or important differences between sites and would have added extensively to an already long paper, the data were not shown.
In the final analysis, one needs to look at the highly significant differences found between the seropositive and seronegative men in our study and ask whether they would be credibly and substantially influenced by the considerations raised by Salman et al.'s letter.