102
Letters to the Editor
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.
REFERENCES
1. Milton RC. Re: "Diet and nuclear lens opacities." (Letter). Am
J Epidemiol 1996;143:101.
2. Mares-Perlman JA, Brady WE, Klein BEK, et al. Diet and
nuclear lens opacities. Am J Epidemiol 1995;141:322-34.
3. Mares-Perlman JA, Brady WE, Klein BEK, et al. Serum
carotenoids and tocopherols and severity of nuclear and cortical opacities. Invest Ophthalmol Vis Sci 1995;36:276-88.
4. Kruskal WH, Tanur JM, eds. International encyclopedia of
statistics. New York, NY: MacMillan Publishing Company,
1978:779.
5. Sperduto RD, Hu T-S, Milton RC, et al. The Linxian cataract
studies: two nutrition intervention trials. Arch Ophthalmol
1993;111:1246-53.
6. Mares-Perlman JA, Klein BEK, Klein R, et al. Relation between lens opacities and vitamin and mineral supplement use.
Ophthalmology 1994; 101:315-25.
Julie A. Mares-Perlman
Barbara E. K. Klein
Ronald Klein
William E. Brady
Department of Ophthalmology and Visual
Sciences
School of Medicine
University of Wisconsin
Madison, WI53705
Man Palta
Department of Preventive Medicine
School of Medicine
University of Wisconsin
Madison, WI 53706
RE: "THE SPECTRUM OF MEDICAL CONDITIONS AND SYMPTOMS BEFORE ACQUIRED
IMMUNODEFICIENCY SYNDROME IN HOMOSEXUAL AND BISEXUAL MEN INFECTED WITH THE
HUMAN IMMUNODEFICIENCY VIRUS"
We read with great interest the article by Holmberg et al.
(1) and the accompanying invited commentary by Fessel
(2). The study has provided much new information. The
authors are to be commended for the detail included in their
"Materials and Methods" section, which provides readers
with a good appreciation of the study design. We feel,
however, that it is important for authors to comment on the
limitations of their study design in the Discussion, to allow
readers to better interpret the results. Therefore, we wish to
raise the following points.
First, it seems that Holmberg et al. used different examination schedules for the three different study regions. All
Am J Epidemiol
Vol. 143, No. 1, 1996
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.
REFERENCES
1. Holmberg SD, Buchbinder SP, Conley LJ et al. The spectrum
of medical conditions and symptoms before acquired immunodeficiency syndrome in homosexual and bisexual men infected with the human immunodeficiency virus. Am J Epidemiol 1995;141:395-404.
2. Fessel WJ. Invited commentary: early symptoms of human
immunodeficiency virus infection. Am J Epidemiol 1995;141:
405-6.
3. Weiss NS. Clinical epidemiology: the study of the outcome of
illness. New York, NY: Oxford University Press, 1986.
4. Hulley SB, Cummings SR, eds. Designing clinical research:
an epidemiologic approach. Baltimore, MD: Williams and
Wilkins, 1988.
5. Fletcher RH, Fletcher SW, Wagner EH. Clinical epidemiology: the essentials. 2nd ed. Baltimore, MD: Williams and
Wilkins, 1988.
6. Rothman KJ. Modem epidemiology. Boston, MA: Little,
Brown and Company, 1986:66.
103
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.
REFERENCES
Salman MD, Hutchison JM, Smith M. Re: "The spectrum of
medical conditions and symptoms before acquired immunodeficiency syndrome in homosexual and bisexual men infected with the human immunodeficiency virus." (Letter). Am
J Epidemiol 1996;143:102-3.
Holmberg SD, Buchbinder SP, Conley LJ, et al. The spectrum
of medical conditions and symptoms before acquired immunodeficiency syndrome in homosexual and bisexual men infected with the human immunodeficiency virus. Am J Epidemiol 1995;141:395-404.
M. D. Salman
J. M. Hutchison
M. Smith
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
Am J Epidemiol
Vol. 143, No. 1, 1996
Scott D. Holmberg
Division of HIV/AIDS (E-45)
National Center for Infectious Diseases
Centers for Disease Control and
Prevention
Atlanta, GA 30333