American Journal of Epidemiology
Copyright O 1997 by The Johns Hopkins University School of Hygiene and Public Health
All rights reserved
Vol 145, No. 7
Printed in U S.A
Progression to Acquired Immunodeficiency Syndrome Is Influenced by CD4
T-Lymphocyte Count and Time Since Seroconversion
Kamilla Begtrup,1 Mads Melbye,1 Robert J. Biggar,2 James J. Goedert,2 Kim Knudsen,1 and
Per Kragh Andersen 13
Progression to acquired immunodeficiency syndrome (AIDS) among persons infected with human immunodeficiency virus (HIV) varies considerably and may be influenced by factors such as age, smoking, number
of male partners per year, and CD4 T-lymphocyte count. The loss of CD4 lymphocytes is known to be the
dominant factor in the progression to AIDS. However, it is unclear whether the effect of the CD4 lymphocyte
count is of such importance that persons with similar CD4 cell counts who have been infected for widely
different lengths of time have the same nsk of AIDS. While a CD4 count is easily obtainable, the precise amount
of time since HIV infection is in most circumstances difficult to assess. In the present analysis, 259 Danish and
245 Amencan homosexual men were followed for up to 14 years from 1981 to 1995. Two hundred and one
persons seroconverted during the study penod, and 112 had developed AIDS before the end of follow-up. CD4
lymphocyte count was highly con-elated with the risk of developing AIDS (p < 0.001), but AIDS risk was not
affected significantly by either age at infection, smoking, or number of male partners per year (p > 0.20 in all
cases). Controlled for CD4 lymphocyte count, time since seroconversion was significant in explaining the risk
of AIDS (p = 0.018), with a lower risk being seen dunng the first 3 years after seroconversion but no effect
thereafter. These data confirm the central importance of CD4 lymphocyte level in the progression of HIV
disease to AIDS, and suggest that rapid progression within 3 years of infection may be related to factors other
than CD4 cell count. Am J Epidemiol 1997;145:629-35.
acquired immunodeficiency syndrome; CD4 lymphocyte count; cohort studies; HIV; homosexuality; risk
factors
The incubation period between human immunodeficiency virus (HTV) infection and the development of
acquired immunodeficiency syndrome (AIDS) is several years in most cases, but a continuous decline in
immunity begins shortly after seroconversion. The
great importance of the CD4 T-lymphocyte count and
its central role in the immunopathology of AIDS was
recognized early in the epidemic (1-3). One of the
problems in analyzing the progression of HTV infection to AIDS is that it is not possible to observe the
exact time of onset of HIV positivity. For each individual, one might typically know that the event oc-
curred either before a certain date or between two
specific time points. If knowledge of individuals'
dates of seroconversion adds nothing to our ability to
predict the onset of AIDS, this problem becomes irrelevant. Phillips et al. (4-6) have suggested that time
from seroconversion is of little importance when the
CD4 cell count is known, hi a cohort of persons with
hemophilia (4) and a cohort of Italian drug users (5),
they found that the effect of time from seroconversion
was insignificant when data were adjusted for CD4
lymphocyte count. This implies that persons infected
long ago have the same risk of AIDS as those recently
infected when their CD4 cell counts are the same. On
the basis of these findings, Phillips et al. have described the application of Kaplan-Meier analysis to
estimate the cumulative risk of AIDS as a function of
declining CD4 cell count (6).
In this report, we investigate the prognostic effect of
time from seroconversion to AIDS after controlling for
CD4 lymphocyte count in HIV-positive homosexual
men. We also analyze the effect of possible cofactors
on the progression to AIDS, such as age at seroconversion, smoking status, and number of male partners
Received for publication November 2, 1995, and in finaJ form
September 3, 1996.
Abbreviations: AIDS, acquired immunodeficiency syndrome; HIV,
human immunodeficiency virus.
1
Danish Epidemiology Science Centre, Statens Seruminstrtut,
Copenhagen, Denmark
* Viral Epidemiology Branch, National Cancer Institute, Bethesda, MD.
3
Department of BiostatJstics, University of Copenhagen, Copenhagen, Denmark.
Reprint requests to Dr. Mads Melbye, Danish Epidemiology Science Centre, Statens Seruminstrtut, Artillerivej 5, DK-2300 Copenhagen S, Denmark.
629
630
Begtrup et al.
per year. Age has been reported to affect progression
to AIDS in some studies, such that older persons have
faster progression than younger persons. An age effect
has particularly been observed in study populations
with a wide age range—e.g., persons with hemophilia
(7). Previous analyses (8) have indicated that smoking
increases CD4 lymphocyte count. Since the CD4 cell
count is known to be associated with progression to
AIDS, smoking might delay the onset of AIDS.
MATERIALS AND METHODS
Subjects
The study population consisted of 259 Danish and
245 American homosexual men. The Danish cohort
was established in 1981 (9, 10), and follow-up was
performed in 1982, 1983, 1984, 1987,1989, and 1992.
Participants donated blood for serum samples and
completed a self-administered questionnaire on lifestyle, sexual practices, smoking status, and exposure
to drugs. The serum samples were screened for HTV
antibodies, and, in all years except 1981, the CD4 cell
counts of participants were determined. The US cohort
was established in 1982 and consisted of 160 homosexual men from Washington, DC, and 85 homosexual
men from Manhattan, New York City (11, 12). Examinations took place every year from 1982 to 1992. The
questionnaire and diagnostic criteria used were the
same as those in Denmark. In Denmark and the United
States, data on the cohorts were updated with respect
to AIDS and vital status on January 1, 1995, and
January 1, 1994, respectively.
The age profile was the same in the two cohorts. At
enrollment, the participants ranged in age from 17
years to 73 years (median, 33 years), but 70 percent
were between ages 25 and 40 years. In both cohorts,
the number of male partners per year was very high in
the early 1980s. In 1982 and 1983, 61 percent of the
men in Denmark and 68 percent of the US men had
more than 10 partners per year, while later only 20
percent of the men had more than 10 partners per year.
Therefore, the number of male partners per year was
treated as a time-dependent covariate. In the United
States, 60 percent of the men were nonsmokers at their
first examination, whereas in Denmark this figure was
only 40 percent.
Twenty-one Danes were HIV-positive at the first
examination, and by 1994 another 48 were known to
have become HTV-infected. AIDS had developed in 37
Danish men by the closure of the study. In the US
cohort, there were 84 HTV-prevalent cases, and another 49 persons seroconverted between 1982 and
1994. In total, 76 US men developed AIDS. However,
one prevalent case from New York had AIDS at entry
into the study and was not included in the analyses.
Estimation of date of seroconversion
For the 97 incident HTV cases, the dates of their last
negative and first positive HTV tests were available.
Data of this kind are known as "interval-censored."
For the 104 prevalent cases, only the date of their first
positive test was available. Since the cohorts were
established in the very early part of the AIDS epidemic, these were cases who had been HIV-positive
for, at most, 3 years, and their dates of seroconversion
could therefore be reasonably estimated. Thus, in Denmark, the dates of HTV seroconversion for the prevalent cases were estimated as the midpoints between
July 1, 1980, and the dates on which they were first
seen to be HIV-positive. This gave an estimated date
of seroconversion of March 1981. The US cohort
consisted of persons from both Washington, DC, and
New York City. Since the disease is assumed to have
appeared slightly later in Washington, DC, the date of
seroconversion for these persons was estimated as the
midpoint between January 1, 1979 (New York City) or
January 1, 1980 (Washington, DC) and the date on
which they were first seen to be HIV-positive. Thus,
persons from New York City who were HTV-seropositive at their initial visit in 1982 were assumed to have
serocon verted in July/August 1980, while persons in
Washington, DC, were presumed to have seroconverted approximately 7 months later.
In estimating dates of seroconversion for the incident cases, we used a method which addressed the
interval censoring problem. This methodology is
based on a generalized linear model with a logarithmic
link function developed by Becker and Melbye (13)
for cases in which the subjects are examined at equal
time points, and has been further developed by
Carstensen (14) to cover situations where the persons
are not examined at equal time points. In short, this
results in a piecewise exponential "survival function"
for the time of seroconversion, and the date of seroconversion is estimated as the median "survival time"
in die interval wherein the person is known to have
seroconverted.
Adjustment of the CD4 lymphocyte count
During the years of the study, there was close cooperation between the research teams in Denmark and
the United States to facilitate comparison of results
from the two countries. Prior to 1987, HTV serology
and CD4 measurements were evaluated in the same
laboratory using similar methodology. Most recently,
these analyses have been standardized and are now
undertaken in each individual country. However, it is
possible that the change in methodology over time
might have influenced the variability of the CD4 cell
Am J Epidemiol
Vol. 145, No. 7, 1997
Time Since Seroconversion and Risk of AJDS
count. Furthermore, the individual variability of CD4
counts can be substantial and may be influenced by
initial clinical status. Studying the HTV-negative samples gave us an estimate of the magnitude of variation.
Analysis of variance on log-transformed CD4 measures from HTV-negative samples showed significant
effects of both subject and time of measurement in
both countries {p = 0.0001), and this led to adjustment of the measured CD4 counts. Estimating the
effect, 6j,j = 1982,..., 1992, of calendar time (time of
measurement), based on the CD4 counts for the HTVnegative persons, all CD4 counts were adjusted by this
factor:
Adjusted CD4, = 0,CD4,.
Follow-up
Since the main purpose of this investigation was to
study the influence of certain risk factors on AIDS risk
in HIV-positive homosexual men, we assumed in the
further analyses that the time of seroconversion was
known. We then considered only the 201 HIV-positive
homosexual men and information collected from the
estimated date of seroconversion to the development
of AIDS or loss to follow-up. Each citizen in Denmark
is assigned a unique 10-digit identification number,
which permits accurate linkage of information from
different registries. Information on AIDS and vital
status was obtained via the national AIDS registry
(mandatory registry) and the Central Person Registry.
The Central Person Registry keeps updated files on all
residents of Denmark and documents such demographic variables as death and migration. Follow-up of
the US cohort was conducted annually with interview
reports of events by subjects or their doctors. AIDS
was defined as clinically manifest AJDS; thus, a diagnosis based solely on a low CD4 cell count (using the
1993 definition) was not considered ATDS-defining in
our analysis. In Denmark and the United States, four
and three persons, respectively, died of causes other
than AIDS. These persons were censored at the time of
death.
Statistical methods
To investigate which factors might have an effect on
progression to AIDS, we used the Cox proportional
hazards model with time-dependent covariates (15).
The basic time scale used was time since estimated
date of seroconversion. Prevalent cases were observed
with delayed entry (left-truncated), and therefore their
"at-risk" period started only at the time at which they
were first seen to be HIV-positive. In the analyses
including CD4 cell counts, the individual "at-risk"
period started at the time of the first CD4 lymphocyte
Am J Epidemiol
Vol. 145, No. 7, 1997
631
measurement after the first positive HTV test. The
covariates were analyzed separately because of the
relatively few AIDS cases. Age at seroconversion was
analyzed as both a continuous and a dichotomous
variable: age minus the mean age, and age above or
below the mean age. Number of male partners per year
was defined as a dichotomous variable depending on
whether the number of partners was above five or less
than or equal to five. CD4 cell count was treated as a
continuous (linear) variable, a log- or square roottransformed variable, and a grouped variable. CD4
count was divided into five groups: 0-100, 100-200,
200-500, 500-800, and >800 cells/mm3. CD4 count,
smoking status, and number of male partners per year
were all fitted as time-dependent covariates; i.e., the
individual values were continually updated throughout
the follow-up period.
The proportional hazards assumption was investigated using log-log survival plots, and results seemed
satisfactory, whereas neither a logarithmic transformation nor a square root transformation of the CD4 count
data satisfied the assumption of a log-linear effect.
Using calendar time as the basic time scale in the
Cox regression model, as suggested by Phillips et al.
(5), it is possible to investigate the effect of time from
seroconversion as well. However, this approach was
not suitable for the present data, since most individuals
were infected during the early part of the study, making time from seroconversion and calendar time highly
correlated. Therefore, for analysis of the effect of time
from seroconversion, we used a Poisson regression
model (15). The AIDS rate was assumed to be piecewise constant in different groups of CD4 counts and
different time intervals since seroconversion. Again,
the CD4 count was continually updated throughout the
follow-up period and then data were grouped into the
five categories described above. Time since seroconversion was divided into number of years from seroconversion (range, 3-9 years), and furthermore data
were grouped by country. The groups were dynamic in
the sense that not only did a person change time group
during follow-up but he might change CD4 group as
well, when a new CD4 count was made.
Although Cox regression analysis and Poisson regression analysis model the hazard of AIDS and the
rate of AIDS, respectively, we will, for convenience,
refer to the estimates as relative risks.
RESULTS
One hundred and twelve of the 201 HIV-positive
subjects had developed AIDS before the end of the
study. Because seven persons died of other causes
before AIDS was diagnosed, AIDS survival time was
632
Begtrup et al.
regarded as progression time to AIDS provided that
the person did not die from other causes. A log-rank
test for the effect of country was borderline significant
(p = 0.08), and there was a tendency for the Danish
men to have a longer time to progression than the US
men. Therefore, in further analyses we stratified by
country. When we fitted a Cox model stratified by
country, no significant effect of smoking status, drug
use, age at seroconversion, or number of male partners
per year was found. However, there was a clear effect
of CD4 lymphocyte count. The relative risk estimates
are shown in table 1. Analyzing CD4 count as an
untransformed time-dependent covariate, we found
that the overall risk of AIDS increased by almost 60
percent for each CD4 count decline of 100. Investigation of the grouped CD4 count data showed no significant increase in the risk of AIDS when the CD4 count
dropped from >800 to 500-800 (p = 0.72) or 200500 (p = 0.13). However, a CD4 count between 100
and 200 increased the risk more than nine times, and a
CD4 count less than 100 increased the risk 30 times
relative to the risk for a person with a CD4 count
greater than 800. We investigated the effect of the
CD4 count measured at the last negative test and the
first positive test, dividing the measured CD4 counts
into two groups. To make the groups approximately
equally sized, values for CD4 count at the last negative
test were divided at 700 cells/mm3 and values for CD4
count at the first positive test were divided at 550
cells/mm3. Time to AIDS was not affected by a high
or low CD4 count prior to seroconversion. However, if
the first postseroconversion count was less than or
equal to 550, the risk of AIDS doubled. Results were
similar when the unadjusted CD4 counts were analyzed.
Fitting the Poisson regression model, a test for interaction between time from seroconversion and CD4
count showed that there was no interaction (p = 0.40).
This means that the effect of the CD4 count on the
progression to AIDS is the same at any time. Interactions between country and the other two covariates
(time from seroconversion and CD4 count) were also
investigated, but no interactions were found. (Because
these tests had little power to detect possible interactions, they were performed in a model with only half
the number of groups.) Table 2 shows the numbers of
person-years at risk and AIDS cases according to CD4
count and time from seroconversion.
In the model without interaction terms, we found
that both CD4 count and time from seroconversion
were significant in explaining the risk of AIDS (p =
0.018 for time, p < 0.001 for CD4 count), but there
was no effect of country (p — 0.58). There was a clear
trend in the effect of the CD4 count such that the risk
increased with decreasing CD4 count. Table 3 shows
the estimated relative risks of AIDS. The estimated
risks in the different CD4 groups are to be interpreted
as die risk in each group relative to the risk associated
with a CD4 count greater than 800 for CD4 counts
observed at the same time from seroconversion. Similarly, the expected relative risk in each time interval
represents the risk relative to that seen 5-6 years after
seroconversion at the same CD4 cell count. The table
indicates that the risk of AIDS was four times lower
TABLE 1. Estimated relative risk of developing AIDS* among 201 HIV*-poalttve Danish and US
homosexual men, 1981-19951
Covariate
Relative
risk
estimate
95%
confidence
interval
P
value
No. of male sex partners per year (>5 vs. 55)
1.04
0.70-1.55
0.84
Smoking (yea/no)
1.26
0.84-1.88
0.27
Age at seroconversion^
1.08
0.95-1.23
0.47
Age greater than mean age at seroconversion
1.27
0.87-1.86
0.21
1.59
1.00
1.33
3.05
9.92
30.66
1.39-1.82
< 0.001
0.28-6.33
0.72-13.0
2.27-43.3
6.89-136.0
0.72
0.13
< 0.001
< 0.001
CD4 lymphocyte count§ (cells/mm3)
>800
>200-5500
>100-<200
5100
CD4 cell count <700 at last negative test
1.26
0.60-2.66
0.54
CD4 cell count 5550 at first positive test
2.37
1.40-J.01
0.0013
• AIDS, acquired Immunodeficiency syndrome; HIV, human Immunodeficiency virus,
t Single factor analyses were performed. The variables (except age) were fitted as time-dependent covariates
in a Cox model stratified by country (Denmark vs. the United States).
t Relative risk per 5-year increase.
§ Relative risk per 100 cells/mm3 decrease.
Am J Epidemiol
Vol. 145, No. 7, 1997
Time Since Seroconversion and Risk of AIDS
633
TABLE 2. Number of AIDS* cases, number of person-years at risk, and yearly rate of AIDS In a group of 201 HIV*-posrUve
Danish and US homosexual men, according to CD4 lymphocyte count and number of years from seroconverslon, 1981-1995
CD4 lymphocyte count (cells/mm 1 )
200-500
>500
No
of
cases
PY*
AIDS
rate
No.
of
cases
0-200
AIDS
rate
PY
No.
of
cases
PY
Total
AIDS
rate
No.
of
cases
py.
AIDS
rate
Years from seroconversion
0-3
3-6
6-9
9-14
Total
1
0.04
137.0
190.1
131.1
401
002
15
14
5
0.03
37
498.3
0.07
1
7
3
1
110.8
135.2
80.0
27.1
0.01
0.05
0.04
12
353.1
3
0.08
0.11
0.12
8.6
023
4
47.2
41.8
25.4
0.49
0.45
0.16
6
45
36
10
256.4
372.5
252.9
926
0.02
0.12
0 14
0.11
48
123.0
0.39
97
974.4
0.10
2
23
19
AIDS, acquired Immunodeficiency syndrome; HIV, human Immunodeficiency virus; PY, person-years at risk.
TABLE 3. Estimated relative risk of AIDS* according to CD4
lymphocyte count and time from seroconversion in 201 HIV*po*JtJvs Danish and US homosexual man in a midtivarlate
Poisson regression model (without including country as a
factor), 1981-1995
Covartate
Relative
risk
estimate
and
subgroLp
95%
confidence
Interval
rears from seroconversion
0-3
3-4
4-5
5-6
6-7
7-8
8-9
9-14
CD4 lymphocyte count
(cells/mm3)
>800
500-800
200-500
100-200
0-100
0.26
1.02
0.91
1.00
0.96
1.17
0.86
0.58
1.00
1.87
3.45
11.10
31.20
0.10-0.68
0.50-2.09
0.45-1.86
0.45-2.05
0.57-2.43
0.38-1.91
0.26-1.29
0.42-8.24
0.86-13.80
2.71-45.30
7.58-129.0
* AIDS, acquired immunodeficiency syndrome; HIV, human
immunodeficiency virus.
within 3 years than it was 5-6 years from seroconversion, while the risk was steady after 3 years {p =
0.80).
Figure 1 shows the estimated AIDS risks by time
from seroconversion relative to AIDS risk 5-6 years
from seroconversion without controlling for the CD4
count (dotted line), as compared with the relative risks
presented in table 3 (full line). The figure shows that
the risks after 8 years were overestimated when we did
not control for the CD4 count. This is probably explained by the fact that after 8 years, the CD4 count
has become very low in most cases—i.e., when controlling for the CD4 count, the effect of time from
seroconversion is reduced.
Am J Epidemiol
Vol. 145, No. 7, 1997
9 2
CD
•£ 1
"
i
o
n 0.5
B
CO
E
M
HI
2
4
6
8
10
Years from seroconversion
12
FIGURE 1. Estimated nsk of developing acquired immunodeficiency syndrome (AIDS) relative to the risk 5-6 years (•) from
human immunodeficiency virus (HIV) infection, according to time
since seroconverslon, among 201 HlV-posrtrve Danish and US homosexual men, 1981-1995. Data are shown after controlling for
CD4 lymphocyte count (—) and without controlling for CD4 lymphocyte count (
).
Primary prophylactic treatment may delay the onset
of AIDS. However, since treatment approaches are
complex, including many regimens and variable
lengths of prophylaxis, treatment was not considered
directly as a cofactor in the analysis. Treatment became generally available by January 1, 1988, but we
found no evidence of a decreased risk of AIDS after
this date.
The effect of time from seroconversion was only
seen within the first 3 years and could have been due
to the large number of prevalent cases. When we
analyzed the prevalent and "incident" cases separately,
the findings were almost similar. When we controlled
for the CD4 count, however, the effect of time from
634
Begtrup et al.
seroconversion was only borderline significant (p =
0.048) among the "incident" cases. (The tests were
performed in a model with only half the regular number of groups.)
The robustness of the estimated date of seroconversion was investigated using, respectively, the earliest
possible and latest possible dates of seroconversion
instead. However, the basic conclusions remained the
same.
As noted above, there was no significant difference
between the effects of time in the two countries. However, analyzing the two countries separately, we found
that the effect of time was insignificant in the smaller
Danish cohort (although with a trend in the same
direction) but significant in the US cohort.
Assuming that the CD4 count can only be considered valid for 1 year after the date of measurement, an
analysis was performed censoring subject follow-up at
1 year after the most recent CD4 count. This censoring
pattern reduced the number of "observed" AIDS cases
somewhat, but gave results similar to those described
above.
DISCUSSION
Our results suggest that none of the covariates studied here (smoking status, drug use, number of male
partners per year, and age at seroconversion) had any
additional value over that of CD4 count in assessing
the risk of ADDS. Other studies, particularly those
conducted among hemophiliacs, have shown that age
at seroconversion is especially correlated with progression to AIDS (7). The insignificant result found in
this study (although with a trend in the same direction)
might have been caused by the relatively small variation in age between subjects, 70 percent of whom were
between ages 25 and 40 years at HTV seroconversion.
There was a tendency for the Danish men to have a
slower progression to AIDS than the US men. We
therefore kept country as a cofactor in our analyses,
even though there was no significant difference between the two countries. It is possible that the socialized health care system in Denmark, which offers the
same high standard of medical care to all citizens free
of charge, may have resulted in more extended prophylactic treatment in the Danish men after they were
infected with HTV. This could have postponed their
time of AIDS diagnosis. Because of the follow-up
procedure for the US cohort, it is also possible that
HIV-positive men who dropped out of the cohort and
as such were censored in our analyses had a better
outcome than those who remained in the cohort.
The analysis showed that the CD4 lymphocyte
count is highly associated with progression to AIDS,
and this corresponds well with previous findings (1-3,
11). The results were not based on any assumptions
concerning the pattern of CD4 lymphocyte loss, but
we adjusted the data for a possible calendar effect due
to changes in measurement techniques. We found no
association between CD4 count prior to HTV infection
and the risk of AIDS, but definitive conclusions on
this point would require a larger data set concerning
measurement prior to HTV infection.
Primary prophylactic treatment may delay the onset
of AIDS and may have influenced these findings.
However, we found no decrease in the risk of AIDS
after January 1, 1988, when prophylactic treatment
became generally available. This is probably due to a
strong correlation between calendar time and time
from seroconversion in closed cohorts such as these.
In the Poisson regression analysis, we found that the
CD4 count, as expected, was related to the risk of
AIDS. Time from seroconversion was an important
factor as well, even when data were controlled for
CD4 count. The effect of time was only seen within
the first 3 years, and the result should therefore be
interpreted with some caution because of the relatively
large fraction of prevalent cases in the two cohorts
analyzed. However, analyzing the prevalent cases separately gave us results similar to those from the original model. Furthermore, using an estimated date of
seroconversion as the earliest or latest possible date
did not change the basic findings. In these subjects,
CD4 count was measured, at most, every year. Thus,
since we updated the CD4 count continually throughout the follow-up period, the CD4 groups to which we
assigned people could in some cases have been quite
unrealistic. However, censoring subject follow-up 1
year after the most recent CD4 count did not change
the conclusions.
Our results differ from those reported by Phillips et
al. (4, 5). However, the cohorts we studied were substantially different in terms of HTV exposure route
(sexual activity vs. exposure to blood products or
sharing of injecting equipment) and other aspects of
lifestyle and medical care. In addition, we studied
events over a 14-year time span from seroconversion.
In the study by Phillips et al. (4), time zero was taken
as August 1, 1985, and time from HTV seroconversion
to AIDS then ranged from 0 to 6 years, while in our
data, it ranged from 0 to 14 years.
One can obtain enhanced effects when the measurement error of a numerical explanatory variable in a
regression model is correlated with the variable being
examined (16). In our data, the measurement error of
the CD4 counts may have been correlated with the true
CD4 values (and thereby time), since low counts are
likely to have smaller variance than high counts. To
Am J Epidemiol
Vol. 145, No. 7, 1997
Time Since Seroconversion and Risk of AIDS
minimize any such exaggeration, we treated the CD4
count in our model as a categorical covariate.
We conclude that a low CD4 count is the most
powerful predictor of AIDS risk, but knowledge mat a
person has only recently seroconverted (within 3
years) adds additional information—namely, that he is
at lower risk of AIDS than someone with a longer
duration of HIV infection and the same CD4 count. In
the present data, a small fraction of approximately 12
individuals had one or two transiently low CD4 counts
close to seroconversion. Possibly, the strain of HTV
and the host's response to initial infection cause aberrations in the CD4 count that, in the early years of
infection, are not tightly linked to functional cellmediated immunity. These topics may provide fruitful
avenues for further research.
ACKNOWLEDGMENTS
This study was supported by grants from the Danish
Medical Research Council (nos. 1993-69-000-8 and 121617-1), the Danish National Research Foundation, and the
US National Cancer Institute (contract no. l-CP-61013).
The activities of the Danish Epidemiology Science Centre
are financed by a grant from the Danish National Research
Foundation.
REFERENCES
1. Ehmann CW, Eyster ME, Wilson SE, et al. Relationship of
CD4 lymphocyte counts to survival in a cohort of hemophiliacs infected with HTV: Multicenter Hemophilia Cohort Study.
J Acquir Immune Defic Syndr 1994;7:1095-8.
2. Farzadegan H, Chmiel JS, Odaka N, et al. Association of
antibody to human immunodeficiency virus type 1 core protein (p24), CD4+ lymphocyte number, and AIDS-free time.
J Infect Dis 1992;1661217-22.
3. Hoover DR, Graham NM, Chen B, et al. Effect of CD4+ cell
count measurement variability on staging HTV-1 infection J
Am J Epidemiol
Vol. 145, No. 7, 1997
635
Acquir Immune Defic Syndr 1992;5:794-802.
4. Phillips AN, Sabin CA, Elford J, et al. Acquired immunodeficiency syndrome (AIDS) risk in recent and long-standing
human immunodeficiency virus (HTV-l)-infected patients
with similar CD4 lymphocyte counts. Am J Epidemiol 1993;
138870-8.
5. Phillips AN, Pezzotti P, Lepn AC, et al. CD4 lymphocyte
count as a determinant of the time from HTV seroconversion to
AIDS and death from AIDS: evidence from the Italian Seroconversion Study. AIDS 1994;8-1299-305.
6. Phillips AN, Lee CA, Elford J, et al. The cumulative nsk of
AIDS as the CD4 lymphocyte count declines. J Acquir Immune Defic Syndr 1992;5:148-52.
7. Rosenberg PS, Goedert JJ, Biggar RJ. Effect of age at seroconversion on tie natural AIDS incubation distribution. Multicenter Hemophilia Cohort Study and the International Registry of Seroconverters. AIDS 1994;8.803-10
8. Burns DN, Kramer A, Yellin F, et al Cigarette smoking: a
modifier of human immunodeficiency virus type 1 infection?
J Acquir Immune Defic Syndr 1991;476-83.
9. Melbye M, Biggar RJ, Ebbesen P, et al. Seroepidemiology of
HTLV-HI antibody in Danish homosexual men: prevalence,
transmission, and disease outcome. Br Med J 1984;289:573—5
10. Melbye M, Biggar RJ, Ebbesen P, et al Long-term seroposltivity for human T-lymphotropic virus type III in homosexual
men without the acquired immunodeficiency syndrome, development of lmmunologic and clinical abnormalities. A longitudinal study. Ann Intern Med 1986,104:496-500.
11. Goedert JJ, Biggar RJ, Melbye M, et al. Effect of T4 count and
cofactors on the incidence of AIDS in homosexual men infected with human immunodeficiency virus. JAMA 1987;257'
331-4
12. Kramer A, Biggar RJ, Hampl H, et al. lmmunologic markers
of progression to acquired immunodeficiency syndrome are
time-dependent and illness-specific. Am J Epidemiol 1992,
136:71-80.
13. Becker NG, Melbye M. Use of log-linear model to compute
empirical survival curve from interval-censored data, with
application to data on tests for HTV positivity Aust J Stat
199O;33-125-33.
14 Carstensen B. Regression models for interval censored data.
Stat Med 1996;15:2177-91.
15. Breslow NE, Day NE. Statistical methods in cancer research.
Vol 2. The design and analysis of cohort studies. Lyon,
France: International Agency for Research on Cancer, 1987.
178-92. (IARC Scientific Publication no. 82).
16. Wacholder S. When measurement errors correlate with truth'
surprising effects of nondifferential misclassification. Epidemiology 1995;6:157-61.