Rosenberg PS, Biggar RJ. Trends in HIV Incidence Among Young Adults in the United States. JAMA. 1998;279(23):1894-1899. doi:10.1001/jama.279.23.1894
From the Biostatistics Branch (Dr Rosenberg) and the Viral Epidemiology Branch (Dr Biggar), National Cancer Institute, Bethesda, Md.
Context.— Behaviors that result in potential exposure to human immunodeficiency
virus (HIV) usually begin in adolescence or young adulthood, but trends in
HIV incidence in young people remain unclear.
Objective.— To estimate trends in HIV incidence in teenagers and young adults.
Design and Setting.— Back-calculation of past HIV incidence in persons born between 1960
and 1974 using US national acquired immunodeficiency syndrome (AIDS) incidence
data and estimates of the distribution of times between HIV infection and
Main Outcome Measures.— Incidence and prevalence of HIV in 1988 and 1993 in persons aged 20
and 25 years, respectively, in each of those years.
Results.— As of January 1993, about 22000 men and 11000 women aged 18 to 22 years
were living with HIV infection in the United States. Homosexual contact was
the leading route of infection among young men. Heterosexual contact was the
leading route of infection among young women. The HIV incidence attributed
to homosexual contact or injection drug use decreased among persons aged 20
and 25 years between 1988 and 1993, but HIV incidence attributed to heterosexual
contact was stable or increasing. Notably, in men aged 20 and 25 years, HIV
prevalence declined by about 50% in white men but was relatively stable in
black and Hispanic men. In contrast, HIV prevalence in women aged 20 and 25
years rose by 36% and 45%, respectively, because of increasing heterosexual
transmission. Overall, HIV prevalence in persons aged 20 and 25 years declined
by only 14% between 1988 and 1993.
Conclusions.— In young persons, HIV incidence in homosexual men and injection drug
users was slowing by 1993; this favorable trend was offset by increasing heterosexual
transmission, especially in minorities.
THE INCIDENCE of human immunodeficiency virus (HIV) in the United States
during the 1990s has fallen to levels below the peak of the mid-1980s,1- 3 but much of this decline
reflects trends in white homosexual men older than 30 years.2
Because high-risk sexual behaviors and drug use are usually initiated in adolescence
or young adulthood,4 the epidemic continues
to be propagated in young people.
Until now, trends in the number of young people who become infected
remained unknown. Surveillance of HIV incidence based on direct testing for
HIV is currently limited to small cohorts at risk.5
There have been numerous surveys of HIV prevalence in selected populations
but the results are difficult to generalize. Prevalence estimates from the
National Health and Nutrition Examination Survey (NHANES) III may be relatively
representative overall, but that survey excludes persons younger than 18 years
and samples too few persons aged 18 to 25 years to provide reliable estimates
specifically for this age group.3,6
To assess national HIV incidence trends in young people, we used a statistical
method known as back-calculation to estimate the
incidence of infection by working backward from acquired immunodeficiency
syndrome (AIDS) surveillance data on the basis of the natural history.1- 3 We used this approach
to reconstruct the HIV epidemic among successive birth cohorts who have entered
adolescence and young adulthood since the onset of the HIV epidemic in 1978.
Previous reports that used this technique2,3
provided no estimates of trends in HIV incidence since 1987. The estimates
of HIV incidence trends derived in this article are, to our knowledge, the
most current estimates because they exploit new methods7- 9
developed by the Centers for Disease Control and Prevention (CDC) to control
for the effects of the 1993 revised AIDS surveillance case definition10 that limited previous analyses. Therefore, we are
now able to estimate incidence trends by age, sex, race/ethnicity, and HIV
exposure category through 1993. In the present analysis, we estimate HIV incidence
and prevalence in 1988 and 1993 in persons aged 20 and 25 years in each of
those years and identify the significant changes that have occurred.
Back-calculation was used to estimate the past incidence of HIV infection
that best predicts the observed epidemic of AIDS cases on the basis of the
distribution of the incubation period between HIV infection and AIDS diagnosis.
Knowledge of the incubation distribution is derived from natural history cohort
studies that use the 1987 AIDS surveillance case definition promulgated by
the CDC. To ensure consistency between AIDS incidence and the incubation distribution,
we define AIDS according to the 1987 definition. All analyses exclude persons
infected by perinatal transmission or transfusion of blood or blood products.
We assumed that the study population was free of HIV infection before the
age of 13 years.
In 1993, the CDC expanded the AIDS surveillance case definition to include
HIV-infected persons with severe immunosuppression.10
To track AIDS using a consistent definition, the CDC has recently developed
statistical methods7- 9
to estimate the incidence of AIDS-defining opportunistic illnesses listed
in the previous 1987 definition. The method distributes forward in time cases
diagnosed solely on the basis of severe immunosuppression according to the
likelihood of an AIDS opportunistic illness, accounting for death before the
onset of an AIDS opportunistic illness as defined in 1987 (competing mortality)
and for delays in the reporting of new AIDS diagnoses.
Using the CDC estimates of 1987-defined AIDS opportunistic illnesses,
AIDS epidemic curves were tabulated for January 1982 through December 1995
using all cases reported to the CDC through December 1996. Cases with no identified
risk of infection as of December 1996 were redistributed to established categories
of HIV transmission according to estimates of the proportions who were eventually
reclassified on the basis of more complete information.7
AIDS epidemic curves were truncated after December 1995 because new highly
active combination therapies11 may have reduced
AIDS incidence and mortality since then12 and
available estimates of the incubation distribution might no longer apply.
Because the hazard of AIDS is low during the first few years after HIV infection,
back-calculation cannot provide reliable estimates of HIV incidence during
the recent past. For this reason, the estimated HIV incidence curves are presented
only through January 1993.
Subgroups of adolescent and adult cases aged 13 years or older were
analyzed according to birth cohort (1960-1964, 1965-1969, and 1970-1974),
sex, race/ethnicity (non-Hispanic white, non-Hispanic black, Hispanic, and
other), and hierarchical mode of HIV transmission (homosexual contact, injection
drug use, men with homosexual contact who also reported injection drug use,
and heterosexual contact). Persons were classified as infected via heterosexual
contact if they reported no male-male sex or injection drug use but did report
heterosexual contact with a person with AIDS or HIV infection or belonging
to an established HIV risk group.
For use in back-calculation models, the distribution of incubation times
between HIV infection and 1987-defined AIDS was estimated from follow-up through
December 1995 of 481 seroconverters in the Multicenter AIDS Cohort Study (MACS).13 The subjects were homosexual men who became infected
between 1984 and 1994 (median year of infection, 1986). In these men, the
median age at infection was 33 years and one quarter of the subjects were
aged 28 years or younger. As in other natural history studies,14- 16
a younger age at HIV infection was associated with slower progression to AIDS.
The estimated median time to AIDS was 9.4 years for persons infected at age
25 years. The observed progression rates in this cohort are well characterized
by a Weibull distribution with a scale parameter of 0.003 and a shape parameter
of 2.448 for persons aged 25 years at seroconversion. Similar to other studies,
the hazard of AIDS increased by the factor 1.025 for each 1-year increase
in age at infection. It is likely that the hazard of AIDS levels off about
8 years after infection.17 The incubation distribution
appears to be similar in men who have sex with men and in men and women infected
through injection drug use.16,18
Hence, these age-adjusted estimates of the incubation distribution derived
from the MACS were applied to each subgroup of AIDS cases determined from
CDC surveillance data.
The HIV incidence curve was modeled as a flexible cubic spline to obtain
smooth estimates without making strongly parametric assumptions.19
Variance calculations used a bootstrap procedure to incorporate uncertainty
about AIDS hazard rates intrinsic to the MACS seroconverter cohort, as well
as random variation of AIDS incidence.19 Injection
drug users who are infected with HIV are at high risk of death from causes
that are not AIDS defining, such as suicide, drug overdose, and bacterial
endocarditis.20 To avoid underestimating HIV
incidence, the back-calculation models for this group incorporated a competing
non-AIDS mortality rate of 1.5% per year from the time of HIV infection, a
figure consistent with findings from cohorts of HIV-infected injection drug
Prevalence of HIV as of January 1983, January 1988, and January 1993
were estimated for persons aged 18 to 22 years (midpoint, 20 years) in each
of those years by summing the monthly incidence of infection through those
dates in the successive birth cohorts and subtracting the corresponding number
of persons with AIDS who died. The number of deaths among HIV-infected injection
drug users was increased to reflect the number expected to die before the
onset of an AIDS opportunistic illness. Similar calculations determined the
prevalence in persons aged 23 to 27 years (midpoint, 25 years) in 1988 and
1993 in each of those years. All estimates were multiplied by the factor 1.18
to adjust for AIDS cases that never get reported to the CDC.22
Because the number of persons engaging in HIV-transmission risk behaviors
is unknown, the number of persons in the general population23
served as the denominator to calculate comparative incidence rates, such as
for the incidence of HIV infection attributed to heterosexual contact or injection
drug use in men compared with the corresponding incidence in women. Point
estimates are provided in the text and 90% confidence limits are provided
in the figures.
Among the 38 million persons in the United States born between 1965
and 1974, 42240 persons had developed AIDS opportunistic illnesses as of January
1996. On a per capita basis for this birth cohort, black men and women had
the highest incidence of AIDS, with 124 per 100000 black men and 60 per 100000
black women developing AIDS during 1995. Hispanic men and women had AIDS incidence
rates of 56 and 20 per 100000, respectively, and white men and women had AIDS
incidence rates of 23 and 5 per 100000, respectively.
Homosexual contact was the leading HIV exposure category in young white,
black, and Hispanic men with AIDS (Figure
1, A-C). In 1995, the incidence of AIDS attributed to homosexual
contact was 4-fold higher in black men and 2-fold higher in Hispanic men than
in white men. Injection drug use was the next largest exposure category in
men with AIDS, followed by heterosexual contact. Heterosexual contact was
the leading HIV exposure category in young white, black, and Hispanic women
with AIDS (Figure 1, D-F). During
1995 the incidence of AIDS attributed to heterosexual contact was 36 per 100000
in black women, 12 per 100000 in Hispanic women, and 3 per 100000 in white
Estimates of HIV incidence are derived from trends in AIDS cases. Compared
with men born between 1960 and 1964, HIV incidence attributed to homosexual
contact in the younger cohorts was substantially higher during the teenage
years but declined to lower levels as these men entered young adulthood (Figure 2). This pattern was observed in white,
black, and Hispanic men. Comparing men who were all aged 20 years, HIV incidence
was statistically significantly lower in 1993 than in 1988. In white men aged
20 years, the rate fell from about 56 to 24 per 100000, a reduction of 57%.
Statistically significant declines were also apparent in black and Hispanic
men. Similarly, HIV incidence was lower among men aged 25 years in 1993 than
in 1988. Among white men aged 25 years, about 110 per 100000 became infected
via homosexual contact in 1988, compared with 27 per 100000 in 1993, a reduction
A similar decline over time was observed in successive birth cohorts
of young white, black, and Hispanic men and women infected via injection drug
use (Figure 3, A-F). For example,
in men aged 20 years in 1988, HIV incidence attributed to injection drug use
was about 7, 55, and 68 per 100000 white, black, and Hispanic men, respectively.
Corresponding rates in 1993 were about 3, 33, and 35 per 100000, respectively,
a reduction of about 50%.
In contrast, HIV incidence attributed to heterosexual contact was stable
or increasing in the successive birth cohorts (Figure 4). Comparing the consecutive birth cohorts during their
teenage years, there was a striking increase in HIV incidence attributed to
heterosexual contact, especially in women (Figure 4). Black women had the highest incidence of infection attributed
to heterosexual contact. In 1993, about 1 in 1000 black women aged 20 years
became infected via heterosexual contact, compared with 1 in 2800 Hispanic
women and 1 in 15000 white women.
Incidence of HIV was higher in young minority men than in white men
from each route of transmission, including homosexual contact. Incidence of
HIV was also higher in young minority women than in white women, both from
injection drug use and from heterosexual contact.
During 1992, about 22000 men and women born between 1965 and 1974 became
infected with HIV. Of them, about 43% were homosexual men, 20% were injection
drug users, and 34% were infected through heterosexual contact. Seventy-three
percent were black or Hispanic persons and 28% were women.
Estimates of HIV prevalence are determined from cumulative HIV incidence
minus deaths. Despite the declines in HIV incidence in young homosexual men
and injection drug users, about 150000 persons born between 1965 and 1974
were living with HIV as of January 1993, and, assuming that the incidence
did not change from 1992, about 200000 persons were living with HIV as of
Prevalences of HIV as of January 1983, January 1988, and January 1993
in persons aged 18 to 22 years in each of those years are shown
in Table 1. In men aged 18 to 22 years, HIV
prevalence doubled between 1983 and 1988, from 14900 to 30300, but decreased
by 27% between 1988 and 1993, to 22100. The reduction in prevalence in young
white men contributed most to this decline. Prevalence of HIV declined only
slightly in black and Hispanic men. In contrast, in women aged 18 to 22 years,
HIV prevalence increased by 36% between 1988 and 1993, from 8100 to 11000,
entirely because of the increasing heterosexual transmission to young women.
Trends in HIV prevalence in persons aged 23 to 27 years were similar (Table 2). The prevalence of HIV decreased
by 45% in white men, increased slightly in black men and decreased slightly
in Hispanic men, and increased by 45% in women because of increasing heterosexual
transmission. Overall, in men and women combined, prevalence in persons aged
18 to 22 years and 23 to 27 years (inclusive of persons aged 20 and 25 years)
declined by 13.8% and 14.2%, respectively, between 1988 and 1993.
We have estimated HIV incidence in successive birth cohorts of persons
who became adults since the onset of the HIV epidemic in 1978. The cohort
born between 1960 and 1964 served as our comparison group. Persons in this
cohort became 18 years old between 1978 and 1982, the period when the HIV
epidemic was just starting but the virus was spreading rapidly. Few were infected
before the age of 18 years but many became infected in their 20s. The younger
cohorts (born between 1965 and 1969, and 1970 and 1974) became 18 years old
between 1983 and 1987, and 1988 and 1992, respectively. For these cohorts,
the pattern of HIV incidence reflects a balance between opposing factors.
The increasing HIV prevalence3 meant that each
risk-taking act conveyed a higher likelihood of exposure, while growing HIV
awareness caused behaviors to change to reduce exposure.
We found 2 major trends in HIV incidence in young persons. First, compared
with the cohort born between 1960 and 1964, younger cohorts more frequently
became infected as teenagers. Incidence of HIV was higher in these younger
cohorts of teenagers regardless of sex, race/ethnicity, or exposure group,
an unfortunate consequence of engaging in risk behavior at a time of high
HIV prevalence. Second, in young adults aged 20 and 25 years, HIV incidence
attributed to homosexual contact and injection drug use was lower in 1993
than in 1988, but the incidence attributed to heterosexual contact was stable
Prevalence reflects both new infections and the number living with HIV.
In young black and Hispanic men, the incidence in teenagers and heterosexual
men offset reductions in incidence in homosexual men and injection drug users
in their 20s. Consequently, prevalence was relatively stable between 1988
and 1993 in minority men aged 20 and 25 years. In contrast, declines in HIV
incidence in young white men were so profound that HIV prevalence in this
group declined by about 50% between 1988 and 1993, in both persons aged 20
years and persons aged 25 years. Occurring at a time of high overall HIV prevalence,
this decline marks a notable prevention success. Unfortunately, HIV prevalence
in women aged 20 and 25 years increased between 1988 and 1993, entirely because
of increasing heterosexual transmission to women. Heterosexual transmission
was the leading route of infection in young women.
During 1992, about 22000 persons born between 1965 and 1974 became infected
with HIV. We estimate that about 43% were men infected via homosexual contact
vs about half of the entire population living with HIV at that time.3 We also found that a higher proportion of HIV-infected
young people had acquired the infection heterosexually than in the entire
HIV-infected population, about 1 in 3 young people vs about 1 in 7 in the
entire population.3 Furthermore, about two
thirds were black or Hispanic, although these minority groups constitute only
27% of the US population born during those years.
Previous studies based on back-calculation estimated the average incidence
of infection between 1987 and 19922,3
and provided no information about trends during this period. The current study
was able to identify these trends because of the use of more recent AIDS surveillance
data. However, 3 potential biases may have affected our results. First, ascertainment
of mode of HIV transmission is subject to misclassification of unknown extent.
In particular, heterosexual transmission may have become increasingly recognized
over time. If so, our estimates of HIV incidence attributed to heterosexual
contact may be too high. Second, CDC data for 1987-defined AIDS opportunistic
illnesses are estimates and may be imprecise. Third, natural history could
vary across risk groups or in adolescents and young adults. If AIDS progression
rates are faster than we estimated, back-calculated HIV incidence will be
Uncertainty about the effects of pre-1996 therapy on time to AIDS probably
had only a minor impact on our analyses for 3 reasons. First, changes in the
incubation period have been subtle and hard to detect in natural history cohort
studies. The benefits of zidovudine monotherapy are greatest in persons with
AIDS or with severe immunosuppression, and prophylaxis against Pneumocystis carinii pneumonia does not prevent other AIDS-defining
opportunistic illnesses from occurring. Second, our natural history estimates
derived from MACS seroconverters. This cohort has therapy effects incorporated
into the incubation distribution because these men have had good access to
treatment from early in the course of infection. Third, the younger cohorts
we studied could not have had HIV infection for many years and their experience
is determined by AIDS events occurring within the first decade after infection.
This part of the natural history curve is well estimated from the MACS and
probably has been less affected by therapy.
The main limitation of our study is that back-calculation cannot identify
HIV incidence trends during the most recent several years. Prevalence of HIV
in young people could be lower or higher today than in 1993. Favoring a lower
prevalence, HIV incidence was slowing by 1993 in young cohorts of homosexual
men and injection drug users; it is hoped that this trend has continued. Furthermore,
new combination therapies that lower the viral level11
may reduce the infectiousness of treated persons who expose others.24 Prevention can be strikingly effective in young persons,
as demonstrated by the marked decline in HIV prevalence in young white men.
However, complacency25 about exposure risks
could have led to a resurgence of the epidemic since 1993. Despite some prevention
successes, the rate of heterosexual transmission in young minority persons
is a particular cause for concern. To guide current prevention efforts, and
to help plan future vaccine trials, we urgently need new studies that provide
consistently obtained, up-to-date data on HIV incidence, prevalence, and risk
behaviors, especially among youth.