Reports on HIV/AIDS: 1990

NOVEMBER 30, 1990, MMWR, Vol. 39, RR-16 [Inclusive page numbers]
The estimate of HIV cumulative incidence for early 1986 depends partly on the
series of AIDS cases modeled. Hay estimated approximately 500,000 infections from
modeling all cases diagnosed through December 1989. Hyman also estimated
500,000 infections but modeled cases diagnosed through June 1989. Their estimates
were substantially less than Brookmeyer's and Rosenberg's estimates of approximately 650,000 infections, estimates obtained from modeling cases diagnosed
through June 1987 that are consistent with the pre-1987 definition. Hay's and
Hyman's estimates are lower partly because they used AIDS cases diagnosed after
mid-1987 and assumed that the distribution of the incubation period did not change
with time. With this assumption, back-calculation accommodates the change in the
trend in AIDS incidence in 1987 (Figure 1) by estimating that infections earlier in the
epidemic, rather than recent infections, tend to be responsible for the AIDS cases
diagnosed during June 1987-June 1989. After subtracting 12,000 deaths among
diagnosed AIDS cases and adjusting for the estimate that reported AIDS cases
represent approximately 75% of all persons with life-threatening symptomatic HIV
infection (Appendix A, Section 4), the range of 500,000-650,000 cumulative infections
yields an estimated HIV prevalence of 650,000-900,000 for early 1986.
There is greater variation in the estimates of HIV cumulative incidence for early
1989 than in the comparable estimates for early 1986. As shown by Brookmeyer's
results, much of this variation is due to differences in the incubation period
distributions used; back-calculation estimates are known to be sensitive to the form
of this distribution (B1,B2). Hay's estimate is low partly because it is based on smaller
adjustments to observed AIDS incidence for reporting delays. His estimate would be
approximately 650,000 cumulative infections if it were based on the same adjusted
AIDS incidence as the other estimates (Appendix A). After adjustments for deaths and
incomplete ascertainment of life-threatening symptomatic HIV infection, this estimated range of 530,000-1,110,000 cumulative infections in early 1989 yields an
estimated HIV prevalence of 630,000-1,380,000 in early 1989. If Hay's estimate of
530,000 infections were replaced by an estimate of 650,000, the lower bound for
estimated HIV prevalence obtained from back-calculation would be 78b;000.
These analyses demonstrate that results from back-calculation must be evaluated
carefully. Analyses based on AIDS incidence for different periods gave markedly
different estimates of cumulative HIV infections, depending on the time period
modeled and the incubation period distribution used. Results also varied among
analysts who used the same time period and similar incubation period distributions.
AIDS case projections: all cases
The number of AIDS cases diagnosed in the United States and reported to CDC
was higher in 1989 than in 1988 (B3). These numbers are expected to continue to
increase through 1993 (Table B2), with 52,000-83,000 cases diagnosed during 1993,
before adjustments for underreporting. Projections adjusted for underreporting
appear in Table B3. The lower bounds for these projections are from Hyman's and
Hay's analyses, which gave nearly identical results. The upper bounds are from Gail's
analyses, assuming 150,000 new infections per year.
The new projections are lower than the corresponding PHS projections made in
1988. For each year in the period 1990-1992, the midpoint of the plausible range is
approximately 15% lower than the projection made in 1988. For example, the
midpoint (71,500 cases) of the range for 1992 is 17% less than the 1988 prediction that

174