AIDS IN THE U.S.: A GROWING EPIDEMIC?
Rethinking AIDS Aug. 1992
The following analysis is based on AIDS cases reported to the Center for
Disease Control (CDC) through December 31, 1991. In order to obtain a clear
picture of the growth of the epidemic, cases must be viewed by diagnostic
year. This is accomplished by analyzing the lag between the date an AIDS
cases is diagnosed and the date it is reported to CDC. A series of "lag
completion" factors can be generated from an analysis of prior lag
patterns. For example, by the end of 1991, virtually all cases diagnosed
prior to 1987 have been reported. On the other hand, based on prior years'
actual runout, we can expect approximately 20% of cases diagnosed in the
4th quarter of 1990 are still unreported.
In the exhibits that follow, actual cases reported by diagnostic year
have been increased by "lag completion factors" to reflect expected
ultimate reporting. So much of 1991 diagnoses remain unreported that 1990
is me last year used.
The most elementary evaluation of the growth of the epidemic entails
solving for the annual growth rate. The growth rate is the number of cases
diagnosed in the current year divided by the number of cases diagnosed
in the prior year. Converting these numbers to percentages (as sbown in
the graph), one can readily see that the growth rate of the epidemic has
decreased every year since 1982 and is approaching zero. The inevitable
conclusion is that the epidemic is cresting, and the number of cases diagnosed
will begin to decline. This can already be seen in certain patient risk
The homosexual AIDS cohort, which accounted for 55% of all cases diagnosed
in 1990, is shown in the accompanying graph. When all cases are counted,
the cohort is seen as approaching a crest. However, when only those homosexual
AIDS cases meeting the more limited 1985 definition are counted (which
provides consistency throughout all diagnosed years), the crest is clearly
seen in 1990. The expanded definition has added 38% more cases in 1990
and delayed the epidemic's crest.
Graph 3 shows that the hemophiliac cohort and the blood transfusion
cohort both crested in 1988. This is a most curious result.
"Evidence from several studies indicates that exposure to HIV began
in 1978 for the U.S. hemophiliac population and that 70-85% of this population
had been infected with the virus by the end of 1984, most becoming infected
in 1981-82" (McGrady, et al., 1987, American Journal of Epidemiology,
126:25). The U.S. hemophiliac population was estimated to be 14,467
as of 1980 (National Heart, Lung and Blood Institute, DHEW pub no. NIH
77-1274). This equates to about 11,000 HIV+ hemophiliacs, infected for
10 years, only 1,713 (15.6%) of whom developed AIDS. The median incubation
period implied by these statistics begins to approach 25 to 30 years. This
is tantamount to saying that for a significant percentage of hemophiliacs,
HIV infection does not lead to AIDS, because the incubation period exceeds
the life expectancy.
lt is estimated that 12,000 transfusion recipients were infected with
HIV in the early 1980s before regular screening of the blood supply was
implemented (Ward et al., 1989, NEJM, 321:947). Through 1990, about
35% had been diagnosed with AIDS. This would imply a 10 year median incubation
period. Perhaps the more interesting observation with respect to transfusion
AIDS is that HIV seropositivity does not appear to confer increased mortality.
A 1989 study by Ward and colleagues at the CDC, designed to ascertain whether
HIV increases risk of death after transfusion, reportcd that "of 233
recipients [HIV+ transfusion cases] whose medical records were located,
95 (41%) had died within one year of transfusion .... By comparison, 73
(50%) of 146 recipients of components from a random selection of donors
not known to be infected with HIV died in the year after transfusion"
(Ward et al., ibid.). Thus transfusion cases with HIV actually had; better
mortality than transfusion cases without HIV (41% vs. 50%).
The intravenous drug user cohort evidences the same pattern as the homosexual
cohort, but is less mature and thus funher from its crest. And the heterosexual
cohon is the least mature of all. The growth rate in this cohort has decreased
every year, but the growth rate in 1990 is still a substantial 39%.
Along these lines, it is of interest to note that heterosexual AIDS
remains virtually nonexistent among teenagers. There have been only 5 cases
ever reported of teenage male heterosexuals contracting AIDS through sexual
activity (1 case in 1990 and no cases diagnosed in 1991), and only 89 female
cases (26 cases in 1990 and 13 diagnoses so far in 1991).
It is difficult to reconcile estimates of HIV infectivity with the above
analysis of AIDS cases. One reason is that the CDC changes its estimate
of HIV prevalence retroactively. e.g., in MMWR Vol. 38/No. 5-4 (May 12,
1989), CDC said, "Thus, while variable, the observed HIV antibody
prevalence data are compatible with CDC's 1.0 million - 1.5 million working
estimate, particularly with the lower end of the range." Then, in
MMWR Vol. 39/No. RR-16 (November 30, 1990), "CDC now estimates that
approximately 750,000 persons in the United States were infected with HIV
at the beginning of 1986." They also estimated HIV infections as of
Juy 1989 at "approximately 1,000,000."
These CDC estimates are based largely on back-calculation methods. Back-calculation
assumes that the incubation curve is known and that the epidemic's curve
is known (from reporting of actual AIDS cases). Thus, one can solve for
the implied HIV infection curve that produces the incubation period assumed
and the number of cases actually materializing. It is apparent that if
one changes the incubation assumption, HIV prevalence changes.
The CDC also uses extrapolation from empirical data as a method to arrive
at HIV prevalence. Such empirical data include military studies, Red Cross
blond donor studies and studies of homosexual male cohorts. Interestingly,
a pilot survey of Dallas County, Texas in 1989, designed to measure HIV
prevalence in the general population, did not corroborate the back-calculation
models used by CDC. In fact, even after adjusting upward to reflect non-response
bias, the survey estimated HIV infections at a number 50% to 70% lower
than three back-calculation models. Similarly, data from the six pilot
sentinel hospitals, which are located in cities that in the aggregate have
an HIV prevalence in applicants for military service of approximately the
national average, show a median prevalence of .24%. This would translate
to 600,000 HIV infections in the U.S. general population of 250,000,000.*