Prevention Efforts

In order to prevent infection from contaminated blood, efforts have been made to reduce HIV prevalence in the population of donors and to screen donated blood for HIV. To increase the sensitivity of the screening procedure, blood is ejected if the first EI A is positive, even if that result is not later confirmed by a second EIA or by WB. Nonetheless, because of the "silent" interval, there remain rare instances of infection from blood products. Risks of failing to detect blood from infections in the silent interval are greater in populations with a preponderance of recent infections. Thus a donor population that contains 0.01% of infected people, most of whom are infected in the recent past, poses a greater risk than a donor population that contains 0.01% of infected people, most of whom were infected years before.

Several steps have been taken to prevent HIV transmission through the blood supply. In 1983, the U.S. Public Health Service first recommended that members of high risk groups be urged not to give blood. In the spring of 1985, compulsory ELISA screening of donated blood and blood products for HIV began. If the initial test was positive, the blood was not to be transfused or manufactured into other products capable of transmitting infectious agents (CDC, 1985a). This procedure maximizes sensitivity for detecting contaminated blood. To improve specificity for counselling donors, the test is only regarded as positive if the ELISA assay is repeatedly positive and if a subsequent positive WB result is obtained. Persons found to be positive are urged not to donate again. It is common practice for a blood bank to retain a confidential list of persons previously tested positive at that facility, and to use that list to prevent subsequent donations by such persons. Since the spring of 1985, the information given to potential donors on who should defer from donating blood has become more explicit (e.g., "males who have had sex with another male at any time since 1977"), and donors have been given the option to confidentially indicate that their blood should not be used for transfusion. The establishment of HIV testing facilities apart from transfusion centers has also been promoted to discourage the use of blood donation as a means of checking HIV status. These measures have reduced the risk of infection from blood transfusion dramatically.

There has been an important decline in HIV prevalence among donors and in the proportion of blood donations that are confirmed WB positive (Table 6.5). In every time period, repeat female donors have much lower seroprevalence than other classes of donors, and new male donors have much higher HIV antibody prevalence rates (Table 6.5). The effects of voluntary deferral are seen among new male donors and new female donors. There has been a 44% drop in seroprevalence among new male donors from 1985 to 1987, and a 10% drop among

Table 6.5 Rates of Positive Western Blot Results per Million Donations to the American Red Cross

New Male

Repeat Male

New Female

Repeat Female





















Source: From Cumming, Wallace, Schorr, and Dodd (1989).

Source: From Cumming, Wallace, Schorr, and Dodd (1989).

new female donors. Possibly there is less awareness among females of what constitutes risk of exposure. The largest percentage drops in seropositivity are seen in repeat donors, where the combined effects of voluntary deferral and blood bank imposed deferral operate. There has been a 69% drop in seropositivity among repeat male donors from 1985 to 1987 and a 29% among repeat female donors.

6.4.2 Effect of the "Silent" Window on the Sensitivity of the EIA Assay

If the EIA test had 100% sensitivity, all infected donors would be screened out. However, as we discussed in Section 6.2, the EIA assay will be negative in the silent time window between infection and the evolution of antibody to HIV. Consider a population of potential blood donors for whom the infection rate is g(s) (see Section 1.3) and for whom the probability of surviving greater than u years after infection is Jiu). Let ij/(u) be the sensitivity of the ELISA assay as a function of time, u, from infection. Define the normalized density of previous infection times among infected donors alive at calendar time T as g*(s\T) = g(s)J(T-s)l^j(s)J(T-s)ds. (6.5)

The density g*(i| T) describes the distribution of previous infection times among the infected members of the donor pool at time T. Then the effective sensitivity of the test at calendar time T for this group of infected donors is sens = Jr g*{s\T)il/(T-s)ds. (6.6)

To simplify, suppose ij/(u) =0 for 0 ^ u < w and 1 for u> w. The sensitivity is thus zero within a window of length w. Under this model, equation (6.6) reduces to namely the probability that infection occurred before T — w. For repeat donors it is reasonable to suppose that new infections occur uniformly in the interval between screens, which we take to be 54 weeks, following Cumming, Wallace, Schorr, and Dodd (1989). Since the chance of death is nearly zero within this time period, the sensitivity of the EIA assay is about (54-zt>)/54 for persons infected since their last donation, where the window width, w, is measured in weeks.

The situation is more complicated for first time donors because we have less knowledge about g*(s\T). To be concrete, suppose T = January 1, 1988. We make a rough estimate ofg*(i|T) by assuming it is proportional to the infection curve, g(s), in the United States. This assumption would not be true if a substantial fraction of those infected early in the epidemic had died, or if programs to discourage donations by high risk donors had been more effective for people infected early in the epidemic than for those infected later. Both these effects would produce distributions of times of infection more weighted toward recent infections. Back-calculated estimates of the infection curve in the United States (see Figure 8.11) suggest that g*(.r| T) can be approximated by the following piecewise linear model: g* (¿| T) = 0 for s < 1979, 0.0342 x (* - 1979) for 1979 < s < 1984, 0.1711 for 1984 ^ j < 1985, and 0.1711 -0.0250 x (s - 1985) for 1985 ^ s < 1988. Thus 59.9% of the infections among potential first time donors in January 1988, occurred before January 1, 1985. Assuming the window, w, is less than 3 years, the sensitivity of EIA is, from equation (6.6) 0.599 + 0.1711 x (3-zo/52) -0.025 x (3-w/52)2/ 2, where w is expressed in weeks. For w = 8 weeks, the sensitivity among first time donors is thus .9847, whereas for repeat donors, the sensitivity is only (54 — 8)/54 = 0.8519.

We can estimate the number of units of contaminated blood that get through the HIV screen, per million donated units, by considering the proportion of blood donated by various types of donors and the corresponding estimated EIA sensitivity (Table 6.6). The prevalences for each of these six donor groups in 1987 are taken from the Red Cross study by Cumming, Wallace, Schorr, and Dodd (1989), as are the proportions of donors constituted by these groups. We treat new and untested repeat donors as having the same high sensitivity, 0.9847, as calculated above, whereas tested repeat donors have the lower sensitivity 0.8519.

Table 6.6 Estimated Number of HIV-positive Units Entering the Blood Supply in 1987, per Million Donated Units


Proportion of



(per million)




New male





Untested repeat






Tested repeat






New female





Untested repeat






Tested repeat








Source: Adapted from table 5 in Cumming, Wallace, Schorr, and Dodd (1989). Mote: Prevalence is based on repeatedly EIA reactive Western blot confirmed units. "Calculated as prevalence times the proportion of donors times 1 minus the sensitivity.

Source: Adapted from table 5 in Cumming, Wallace, Schorr, and Dodd (1989). Mote: Prevalence is based on repeatedly EIA reactive Western blot confirmed units. "Calculated as prevalence times the proportion of donors times 1 minus the sensitivity.

These calculations lead to an estimate of 4.64 contaminated units per million donated units (Table 6.6). Ward, Holmberg, Allen, et al. (1988) outlined the window calculation and obtained a rate of 2.6 contaminated samples per million by simply dichotomizing the donor pool into first time donors and repeat donors and by assuming a sensitivity of 0.99 for first-time donors. Cumming, Wallace, Schorr, and Dodd (1989) obtained a somewhat higher rate of 6.5 contaminated units per million donations from the data in Table 6.6 because they assumed g* (s) was uniform over the 5 years preceding donation for first time donors and uninfected repeat donors, yielding an EIA sensitivity of (5 x 52 — w) /5 x 52 = 0.9692 for a window of 8 weeks, instead of the value 0.9847 in Table 6.6.

One can compare estimates of the chance of false negative screening results based on the window calculation with the scant available empirical data. Cohen, Munoz, Reitz, et al. (1989) found one seroconversion among cardiac surgery patients who received 36,282 units, for a rate of 28 per million units. Based on Poisson sampling, the corresponding 95% confidence interval is 6.7 to 154 per million units. Busch, Eble, Khayam-Bashi, et al. (1991) found one positive unit confirmed by culture and DNA amplication among 76,500 donations in San Francisco. Based on Poisson variation, the corresponding estimated rate of false negative results would be 13 per million units, with 95% confidence interval 3.2 to 73 per million units. The empirical data thus yield estimates of false negativity rates in reasonable concordance with the window calculation, especially in view of the considerable random uncertainty associated with empirical estimates.

Based on window calculations, one can assert that the risk of contracting HIV from a blood transfusion is small. A patient requiring 20 units of blood might have a risk on the order of 1-(1 — 4.6 x 10-6)2O = .00009. If the recent estimate (Petersen, Satten, and Dodd, 1992) that the average window width is only 45 days is correct, the risk would be even smaller. To get an idea of the maximum plausible risk, we assume a window of width w = 16 weeks. Then the sensitivity for first-time donors decreases to 0.9690, and the sensitivity for repeat donors decreases to 0.7037. The expected number of contaminated units per million donations increases to 9.33, and the chance of infection from receiving 20 randomly selected units increases to 0.00019. If we further assume that the HIV prevalences in Table 6.6 are too small because the testing procedure with Western blot confirmation used to estimate these prevalences has sensitivity 0.90, the chance of a false negative result is increased further to 9.33/0.90 = 10.37 per million, and the chance of infection from 20 units increases to 0.00021. If, based on empirical data, we assume the rate of false negative results is 20 per million units, then the chance of infection from 20 units increases further to 0.00040. Such risks are small compared to the risks of not receiving needed blood products.

The window calculations can be refined to take into account the fact that some persons who tested negative at year T— 1 and donated again at year T were, in fact, already infected at or before T— 1. The screening assay at T would have 100% sensitivity to detect such persons for window widths less than one year. Hence the sensitivity would be somewhat higher than the value (54 — m>)/54 = (54 — 8)/54 = .8519 calculated previously. The improvement is small, however, as illustrated by the following calculations. We assume that mortality is negligible and consider the population of individuals infected before T whose first donation is at T— 1, and who subsequently donated at T. We assume that the chance, n, of donating at T — 1 and T is the same for persons who were infected before T — 1 and tested negative at T — 1 as for persons who were first infected between T — 1 and T. Then, of the infected persons at 7* who tested negative at T — 1, a proportion y = A ¡(A + B) were already infected by T — 1, where A and B are approximately

A = it x {1 — sensitivity at T — 1 for a first time i donor who is infected} x g(s)ds and

B =n x {specificity for a first time donor at T — 1

who is uninfected} x I g{s)ds.


Supposing g(s) was proportional to g* (i| T = 1988), which was defined previously, and setting T — 1 = 1987, we obtain A = nc(\ — 0.9847) x (.8912) by assuming the sensitivity for first time donors is 0.9847 as before, and B = 7tc(0.98) x (.1088), by assuming the specificity for a single unconfirmed EIA is 0.98. Here c is a proportionality constant that cancels out when calculating y = 0.113. Based on this calculation, the window model with w = 8 weeks predicts a sensitivity of 0.113 + (1 - 0.113) x (52 - 8)/52 = 0.864 for persons screened negative one year before. This value is 1.4% greater than the value .852 obtained by ignoring the possibility of false negative screens at T — 1. Such calculations suggest that the previous simple window calculations are sufficiently accurate given other uncertainties, but that they slightly underestimate the sensitivity of the screening procedure. A simulation study by Le Pont, Costagliola, Massari, and Valleron (1989) yields results similar to calculations based on the window model.

6.4.3 Screening Blood in Developing Countries

In a developing country with high seroprevalence rates, screening can reduce numbers of transfusion-related infections drastically. Even in a rapidly growing epidemic in which there are many recent infections so that the dates of infection among infected donors are uniformly distributed over the previous 5 years, a window calculation with w = 2 months suggests an effective sensitivity of about {(60 — 2)/60} = 0.967. If the prevalence of infection among donors were 5%, screening could thus reduce the chance that a transfused unit was infected from 5% to 5% x (1 — .967) = .17%. The chance that a person receiving three units would become infected would thereby be reduced from 1 - (1 - .05)3 = .142 to 1 - (1 - .0017)3 = .005, a dramatic improvement. To achieve such gains, developing countries must be able to find economic and logistical support for screening facilities, and the tests must be simple and reliable enough to use without the need for elaborate laboratory facilities and quality control systems. Quinn and Mann (1989) discuss the role of transfusion ofblood products and other factors that promoted the spread of HIV in Africa.

0 0

Post a comment