High HIV incidence or poor test performance?

Abstract

We read with much interest the article by Moodley et al. [1] who reported an HIV incidence of 10.7 per 100 woman-years among pregnant women in three provinces of South Africa [95% confidence interval (CI) = 8.2–13.1]. This high incidence is particularly worrisome considering its implications for programmes to prevent mother-to-child transmission (PMTCT) of HIV [2]. With improvements in coverage of PMTCT programmes and implementation of more efficacious prophylactic regimens, a considerable proportion of HIV infections in infants can be attributed to incident maternal HIV. A Botswana study [3] estimated that this proportion was as high as 40%. Before drawing such definitive conclusions, however, critical methodological issues around HIV testing need to be addressed. Variability in the performance of test kits, category of staff and testing algorithms make it hard to make direct comparisons between incidence studies. Concerns about testing validity present quite different policy implications from those described above. Using different testing methodologies, reports of HIV incidence among pregnant women in South Africa have varied from 0% (0/524) in the Western Cape [4] and 5.2% per year in a national sample [5] to 10.7 per 100 woman-years in the study by Moodley et al. [1]. Differences in the sensitivity of tests and of testing algorithms in field conditions could, at least in part, explain these differences. This assertion is supported by our experiences at PMTCT sites in three provinces of South Africa. Rapid kits for HIV testing often vary across sites and change frequently, depending on national tender processes and available stock. Training of midwives and counsellors in the performance of these tests is not systematic and often not repeated each time a kit changes. Given these circumstances, it was not surprising that Moodley et al. [1] reported previously on the poor sensitivity of four HIV rapid tests in field settings in a PMTCT programme in KwaZulu Natal. Sensitivity of kits in that study ranged from 92.5% to 97.3%, with a serial testing algorithm as recommended by the National Department of Health [6]. Using these figures and the HIV prevalence among pregnant women in South Africa (28.0%, 95% CI = 26.9–29.1) [7], it is possible to estimate the number of women who had a false-negative test and were then enrolled in the HIV incidence study of Moodley et al. [1]. Of the 2377 women enrolled, between 25 and 70 may have had false-negative initial tests, comprising a substantial proportion of the 72 incident cases that they report. The authors do not specify which rapid HIV tests or the algorithm used in the field to determine inclusion in the HIV negative cohort nor whether quality control procedures were in place at these sites. Interestingly, women in their cohort were re-tested later in pregnancy using the Determine HIV-1/2 rapid HIV test (Abbott Laboratories, Abbott Park, Illinois, USA), which demonstrated the highest sensitivity in the previous study by Moodley and et al. [8], and has been previously validated in Africa [9]. Use of a more sensitive test during re-testing may also partly explain the high incidence. To improve comparisons between studies and more clearly separate new HIV infections from low negative predictive values of HIV tests or testing algorithms, future reports should fully describe the test kits, algorithms and quality control procedures used. More importantly, additional research is needed to determine whether parallel HIV testing (in which a negative HIV diagnosis is made after two concordant negative tests) is superior to serial algorithms in field conditions. Robust quality control procedures in all sites offering HIV rapid tests remain critical. Moreover, we argue that it is premature to implement re-testing for pregnant women before obtaining further clarity about the quality of actual testing procedures. Acknowledgements Vivian Black, Regina Osih, Helen Rees and Matthew Chersich contributed to the conception and writing of the paper. There are no conflicts of interest.