TEMA: Estudio del Estándar de Televisión Digital Terrestre ISDB-

The most unsatisfactory features of the new method described in Chapter 2 to estimate the HIV vertical transmission rate and the AIDS incubation period are (i) the assumptions about the sensitivity and specificity of virus tests, and (ii) a limitation in accommodating only a single categorical covariate. The first problem could be addressed by introducing a latent distribution that allows for individual-specific sensitivities, similar to the models of Chapter 4. The

presence and effect on estimates of serial correlation within individuals would also need to be examined. Concerning the second problem, k in equation (2.1) could, in principle, be replaced by an arbitrarily complex function of covariates. With these extensions to the basic model an EM algorithm, or even direct maximisation of the likelihood using standard subroutines, is likely to be so complex that Markov Chain Monte Carlo methods may be a more attractive approach to estimation. However, for reasons outlined below it is not clear if it would be worthwhile investing considerable effort in these lines of research.

The analysis of Chapter 3 indicates that the sensitivity of PGR after the

neonatal period is apparently close to 100%, which has also been reported in related studies (Section 3.5.3). This would appear to conflict with the results of Chapter 2 where PGR sensitivity was estimated as only 92%. A likely

explanation for this discrepancy is that the tests in the European Collaborative Study were performed for routine clinical diagnosis whereas the data in Chapter 3 were obtained primarily from research laboratories. The collection of an early blood sample for PCR analysis should now be a standard feature of all vertical transmission study protocols. Where this is done, and provided laboratory quality control procedures are maintained, a reliable early diagnosis should be available for all children. This effectively eliminates any difficulties in the statistical analysis and thus nullifies advantages of the method described in Chapter 2. This is unlikely, therefore, to find wide application in the context of HIV infection, although ad hoc applications could be envisaged.

such as Toxoplasma gondii. HTLV-1, and hepatitis C virus have generally been unsophisticated. In principle, the new method could, with suitable modification, be applied to any of these infections. A common feature is that reliable

diagnosis can eventually be made using antibody assays but that direct tests for the virus or parasite (e.g. PCR) are highly inaccurate. Detailed consideration would need to be given to the biology of, and tests for, each particular infection. For example, a model for hepatitis 0 would need to take account of the

possibility of complete clearance of the virus (Ni et al. 1994).

Returning attention to HIV infection, it is generally held, partly on the basis of the analysis of Chapter 3 and related studies, that transmission around the time of delivery is more frequent than transmission in utero. This is the rationale for a number of studies that have evaluated or are evaluating interventions against intrapartum transmission, including antiseptic cleansing of the birth canal, zidovudine started shortly before the expected date of delivery, and caesarean section delivery (Biggar et al. 1996, Cohen 1995). Results from these studies should shed further light on the critical question of the timing of transmission.

Early PCR tests still have an important role to play. For example, caesarean section delivery is of no value if the fetus has already been infected. A powerful way of assessing the effectiveness of this intervention would be compare

transmission rates by mode of delivery in infants who apparently avoided prepartum transmission, as reflected by an early negative PCR test result. Also, data which have been used to assess the sensitivity of PCR largely pre­ date publication of the ACTG-076 trial (Connor et al. 1994) and thus mainly include children of women who did not receive zidovudine during pregnancy. However, zidovudine could be more effective at blocking some mechanisms of transmission than others and thus affect the age at which viral markers are first detectable. Kuhn et al. (1997) explored these and other issues in a recent paper, defining presumed intrauterine and presumed intrapartum infection as a positive or negative PCR within 3 days of birth, respectively. Their findings were inconclusive, possibly due to the small sample size (48 infected infants).

This is an interesting approach and there is a need for collaborative analyses. Using an arbitrary cut-off, as in the paper of Kuhn et al., is wasteful of

information, and the paucity of data warrants use of one of the more efficient methods described in Chapter 3.

The most difficult challenges in terms of study design and statistical analysis concern studies conducted in breastfeeding populations. For example, the primary endpoint in an African trial to assess the effectiveness of antiseptic cleansing of the birth canal was infection status as assessed by PGR at six weeks of age (Biggar et al. 1996). Rates of PCR positivity were almost

identical in the intervention and control arms and it was inferred that antiseptic cleansing is not effective in reducing vertical transmission. However, if a difference had been found the usefulness of this intervention in public health terms would have been uncertain because the effect of subsequent

transmission through breastfeeding had not been measured. The frailty models of Chapter 4 suggest that breastfeeding will tend to nullify the effect of an

intervention which reduces the risk of transmission during or shortly after delivery. It has been strongly argued, therefore, that follow-up should continue in all intervention studies throughout the entire breastfeeding period (Ekpini et al. 1997).

In this situation the statistician would have data available on antibody tests and clinical status, but the new method for estimating the vertical transmission rate that was developed in Chapter 2 would not be directly applicable. Firstly, a negative antibody test result implies that a child has ultimately escaped infection only if he has already been weaned and is not in the window period between infection and antibodies appearing in peripheral blood. Secondly, the method was developed in a context where the clinical diagnosis of AIDS can reasonably be assumed to be 100% specific for HIV infection. In settings with high

underlying infant morbidity and mortality rates, for example parts of Africa, this assumption would not be valid. Further work is required to develop appropriate methods of estimation for these circumstances. This need is pressing in view of

the large number of studies of "affordable" interventions that are being carried out in less-developed countries (Cohen 1995). To my knowledge, none of these studies are using a control group of children born to HIV-uninfected mothers, precluding use of the "indirect" approach described in Section 1.5 to estimate the absolute risk of transmission within each intervention arm.

Finally, there remains the important issue of the extent and timing of

transmission through breastfeeding in itself. The analysis in Chapter 4 of data from the Sao Paulo study appears to indicate that this is an important mode of transmission in the first few months after delivery but that the risk of infection is small at older ages. This conclusion must be tentative since the accuracy of the information on duration of breastfeeding, which was recalled retrospectively, is open to question. However, a recent prospective study in Soweto, South Africa also found a strong effect of ever breastfeeding on transmission risk without a clear dose-response relationship with duration (Glenda Gray, personal communication).

An alternative approach to studying transmission through breastfeeding is to test serial blood samples by PCR or antibody tests. Serological tests can provide information only on late post-natal transmission as few uninfected children have cleared maternal antibodies before 6 months of age. However, as HIV DMA can be detected in almost all infected, non-breastfed children by one month of age, PCR can be used to demonstrate earlier post-natal

transmission. Buterys et al. (1995) estimated a transmission rate of 5.8 per 100 child-years after age 20 months. Ekpini et al. (1997) estimated that among children who escape infection during the first 6 months of life, 20% of those who are breastfed until age 24 months will acquire infection by that age. These findings suggest, unlike the studies in Sao Paulo and Soweto, that substantial transmission through breastfeeding is not limited to early infancy.

Appendix 2.1 FORTRAN 77 program to Implement EM algorithm