E. Litvak et al., WHOSE BLOOD IS SAFER - THE EFFECT OF THE STAGE OF THE EPIDEMIC ON SCREENING FOR HIV, Medical decision making, 17(4), 1997, pp. 455-463
Background. With improvements in HIV antibody test (ELISA) performance
, the window of time between infection and seroconversion becomes a ma
jor source of error in HIV screening. The authors examined its impact
on the false-reassurance rate (FRR). Methods. Test sensitivity was mod
eled as the product of two factors: the inherent sensitivity (sensitiv
ity when antibody is present) and the probability that antibody is pre
sent in infected blood. A model of HIV and AIDS incidence was used to
derive an estimate of the probability of remaining in the seronegative
window (p(w)) among those who are infected. With plausible assumption
s, this probability approaches 0.03. The FRR was then estimated as a f
unction of the probability of remaining in the seronegative window, th
e prevalence of HIV, and the inherent sensitivity of the ELISA test we
re estimated. Results. The FRRs for two blood donor groups, one with a
n HIV prevalence of 0.004 and a typical probability of remaining in th
e seronegative window (p(w) = 0.03) and the other with a higher preval
ence of 0.017 but fewer donors in the window (p(w) = 0.003), are equal
(140 per million donors) if the blood is negative on a single ELISA t
est. After two negative tests or a single test that can detect antibod
y more reliably, however, the FRR is much higher in the group with the
higher p(w) (= 120 per million compared with 50 per million), because
the greater numbers of donors in the window more than offsets the low
er prevalence. Conclusions. With improvements in inherent sensitivity
of ELISA by virtue of technical progress or retesting, the prevalence
of HIV infection may no longer play the critical role in degrading the
results of blood screening. As inherent test performance improves, te
sts are increasingly likely to miss infected blood because of the sero
negative-window error rather than because of measurement error. Window
error plays a proportionally greater role during the early stages of
HIV dissemination in a population where the incidence of new HIV infec
tion is high relative to the incidence of AIDS. These findings may exp
lain, in part, the recent observation that cases of transfusion of con
taminated blood often take place in areas where AIDS epidemics have st
arted recently. They also suggest that the traditional strategy of sol
iciting blood donors from low-prevalence populations may not always be
optimal, unless such populations are truly low-risk.