In most countries in sub-Saharan Africa, the HIV epidemic is monitored by conducting sentinel surveillance among pregnant women attending antenatal clinics (ANC) . In this part of Africa, heterosexual intercourse is the main mode of HIV transmission, and the prevalence of infection among attendees of ANC is usually assumed to be representative of that among the general female population and is thus used for planning and evaluating public-health interventions. Nonetheless, comparative studies conducted in Africa have shown that the HIV prevalence among pregnant women tends to represent an underestimate with respect to the prevalence among the general female population of reproductive age [2–9], especially in areas where the epidemic is mature and there is a low rate of contraception use. This is mainly because HIV-positive women are less represented in ANC than HIV-negative women, as a result of the lower fertility among HIV-positive women [7,8,10], which is associated with social, behavioural and biological factors [11,12].
To address this issue, two methods that adjust for differences in fertility by HIV serostatus have been proposed for estimating the HIV prevalence among the general female population using data from pregnant women [13–16]. The first method was described and applied to data from an ANC in the Gulu district of North Uganda , but its performance was not fully evaluated because of the lack of comparative data from population-based surveys. The second method was applied separately to data from pregnant women living in the Masaka district (Southwest Uganda) and in the Mwanza region (Tanzania), and the results were encouraging when the adjusted estimates were compared to the observed prevalence among the general female populations living in the same study areas and among whom the use of modern contraceptives is low [16,18]. On the other hand, the same method did not provide accurate results when applied to data from ANC attendees living in three districts of the Manicaland province in Zimbabwe, an area where the rate of contraceptive use among the population is high .
The objective of the present study was to evaluate these two methods by comparing the adjusted HIV prevalence rates among ANC attendees to those observed among the general female population, using published data from studies conducted in sub-Saharan countries where the epidemic is mature and the use of modern contraceptives among the population is not highly diffused [4,6,9].
Materials and methods
The first method consists of adjusting the age-specific HIV prevalence rates observed among ANC attendees accounting for differences in fertility risk between HIV-positive women and HIV-negative women, and then directly standardizing the overall prevalence using the age structure of the general female population as a reference [14,17]. In detail, the adjusted HIV prevalence for the general female population of reproductive age (PGFP) was calculated according to the following formula:
where, for each age class, FGFP indicates the proportion of women among the general female population; PANC indicates the HIV prevalence observed among ANC attendees; FR*HIV+ and FR*HIV− indicate the fertility risk among HIV-positive women and HIV-negative women in the standard reference population, respectively.
To apply this method, we used, as standard reference, the age-specific ratios of fertility risk among HIV-positive women to that among HIV-negative women (equivalent to using the age-specific relative odds of HIV infection in pregnant women compared with all women ) estimated in a large population-based study conducted in the Masaka district (Southwest Uganda) (Table 1). In this study, over 1500 resident women of reproductive age were tested for HIV infection and classified according to whether or not they had given birth in the previous year at each of the annual survey rounds, which were conducted from 1990 to 1996 .
The second method adjusts separately for two aspects of fertility, that is, infertility and subfertility . However, for the present study, we adjusted only for infertility because some of the data necessary for evaluating subfertility were lacking (i.e., last birth-interval among multipara pregnant women). This method requires standard reference data on both the relative HIV prevalence and the population distribution by fertility risk category (see categories listed in Table 2) and parity (i.e., mothers versus women who have never borne a child). Specifically, the relative HIV prevalence rates were used for extrapolating HIV prevalence from ANC attendees (assumed to be representative of the group of sexually active fecund women) to the other fertility risk categories of the general female population, thus accounting for differences in infertility by HIV serostatus, whereas the population distribution was used as reference for directly standardizing by fertility risk category and parity the overall adjusted HIV prevalence for the general female population. In detail, the adjusted HIV prevalence for the general female population of reproductive age (PGFP) was calculated according to the following formula:
where, for each stratum of parity by fertility risk category, FGFP and RP*ANC indicate the proportion of women among the general female population and the standard relative HIV prevalence with respect to that among ANC attendees, respectively, and, for each parity stratum, PANC indicates the HIV prevalence observed among ANC attendees.
The data on the standard relative HIV prevalence by fertility risk category and parity were those given by Zaba et al.  and estimated for populations with a low rate of contraception use analysing pooled data collected in Kisesa, Tanzania (1994–1998) and Masaka, Uganda (1995–1996) from over 5000 women of reproductive age [8,16,20]. The data on the population distribution by fertility risk category and parity were those referred to urban Zambia (Table 2), where the use of modern contraceptive methods is not highly diffused (17.0%), and estimated by analysing data from a large demographic and health survey conducted in 1996 .
The adjustment methods were applied to data from towns in Uganda, Tanzania and Zambia, and to data from rural areas of the latter two countries. A detailed description of the data sets used for these analyses is reported in Table 3. For the first method, the adjusted HIV prevalence rates for the years 1994 in Chelston (Lusaka, Zambia) and Mposhi (Zambia) were compared to the prevalence observed among the corresponding general female populations in 1995–1996. For the second method, we adjusted only data from Chelston for the years 1994 and 1998 and those from Ndola (Zambia), whereas, given that the second adjustment method was already applied to the data set from rural areas of the Mwanza region (Tanzania), we reported just the adjusted estimate previously presented by Changalucha et al. . We did not apply the second adjustment method to the other data sets because information was not available on the HIV prevalence by parity among ANC attendees.
For each study site, we first compared the HIV prevalence observed among ANC attendees to that observed among the same-age general female population to determine the extent to which the prevalence among ANC attendees underestimated the prevalence among the general female population of reproductive age. We then applied the adjustment methods to the prevalence observed among ANC attendees and measured the difference between these adjusted prevalence rates and that among the general female population, in order to determine whether the approximation had improved. The differences in the estimates were evaluated using the relative error, which was calculated as the percentage difference with respect to the HIV prevalence observed among the general female population.
The standardized HIV prevalence among ANC attendees underestimates the prevalence observed among the same-age general female population by 31.2%, 23.8%, 31.9% and 28.7% in Fort Portal (Uganda), Mwanza Municipality (Tanzania), rural Mwanza (Tanzania) and Mposhi (Zambia), respectively (Table 3). In the other Zambian sites, the HIV prevalence among ANC attendees underestimates the prevalence observed among Chelston's general female population by 20.7% and 21.4% for 1994 and 1996, respectively, whereas for 1998 there is an underestimation of 8.0%. For Ndola, the prevalence among ANC attendees underestimates the prevalence observed among the general female population by 23.7%.
First method: data adjusted for differences in age-specific fertility by HIV serostatus
The age-specific ratios of fertility risk provided by the Masaka study, which we used for applying the first adjustment method, show a fertility risk of 1.6–2.8 times higher among HIV-negative women than among HIV-positive women for all age groups, with the exception of women aged 15–19 years (Table 1). For this age group, the fertility risk among HIV-positive women was 1.3 times higher than that among HIV-negative women. As a consequence, when adjusting the data observed among ANC attendees, the difference in prevalence between the 15–19-year age group and the 20–24-year age group becomes much wider in all sites.
The adjusted estimates underestimate the HIV prevalence observed among the general female population in Fort Portal and Mposhi by just 0.4% and 0.6%, respectively, and overestimate by 3.3% and 3.7% that observed in Chelston in 1994 and 1996, respectively, and by 6.5%, 10.6% and 12.8% that observed in Ndola, Mwanza municipality and rural Mwanza, respectively (Table 3). By contrast, the adjusted prevalence for the year 1998 in Chelston overestimates the observed prevalence among the general female population by 22.9%, thus providing a much less accurate prediction of the true prevalence than did the non-adjusted prevalence (relative error = −8.0%).
Second method: data adjusted for differences in HIV prevalence by fertility risk category and parity
The relative HIV prevalence rates estimated from pooled data collected in the Masaka and Kisesa studies, which we used for applying the second adjustment method, show that the sexually active fecund women had a lower HIV prevalence than all other sexually active women; although this was true for both women who have never borne a child and mothers, the differences were more pronounced among the former (Table 2). Among mothers, all categories of sexually active women had a lower prevalence than women who were not currently sexually active, whereas among women who have never borne a child, those who were currently sexually active had a higher prevalence, mainly because in this group, women who are not currently sexually active are younger and mainly represented by women who had never had sex.
The proportion of women who have never borne a child among the general female population in urban Zambia (30.3%) was similar to the proportion of primipara observed among the ANC attendees in the two study areas where the method was applied (26.6% and 32.2% in Chelston in 1994 and 1998, respectively, and 27.2% in Ndola). This suggests that most of the differences between the HIV prevalence rates among the ANC attendees and the adjusted rates (Table 3) are due to the assumed higher overall prevalence among those fertility risk categories not represented in ANC. When using the second method, the adjusted estimates for the Zambian urban sites were 1.26–1.36 times higher than the non-adjusted estimates among the ANC attendees. However, considering data by parity, the adjusted estimates for mothers in the general population were 1.52 times higher than the non-adjusted estimates among multipara ANC attendees, whereas the adjusted estimates for women who have never borne a child in the general population were 0.71 times lower than the non-adjusted estimates among primipara ANC attendees (Table 2).
When comparing the results of the two methods, in Chelston, the second method provided less accurate estimates, which represented overestimates of 7.0% and 25.1%, respectively, in 1994 and 1998 (Table 3). By contrast, in Ndola, the adjusted estimate provided by the second method was more accurate, representing an underestimate of 3.8%. When the second adjustment method was previously applied to data from rural Mwanza, using the population distribution by fertility risk category and parity estimated at the Tanzanian national level as standard reference , it provided a more accurate adjusted estimate (underestimate of 2.1%) than that provided by the first adjustment method (overestimate of 12.8%).
As expected, the HIV prevalence rates among ANC attendees underestimate the prevalence among the general female population in all of the considered sites, confirming findings from other studies conducted in sub-Saharan Africa [2,4,7]. The first adjustment method removed this difference almost totally in Fort Portal and Mposhi (98%), whereas it reduced the absolute value of this difference by about 83% when applied to data from Chelston for the years 1994 and 1996, and by 72% and about 60% when applied to data from Ndola and both urban and rural Mwanza, respectively. However, when the method was applied to data from Chelston for the year 1998, the adjusted prevalence actually overestimated the prevalence among the general female population by 22.9%, representing a less accurate estimate than the non-adjusted data. This could, in part, be due to the unusually high prevalence observed in 1998 among the ANC attendees aged 30–39 years. In fact, in 1994 and 1996, the prevalence for this age group was lower and very close to that observed among the ANC attendees aged 15–19 years. This relationship between HIV prevalence and age, which is mainly due to both AIDS-related mortality and the increasing difference in fertility by HIV serostatus with age in mature epidemics, is also suggested by data from other ANC-based sentinel sites in sub-Saharan Africa [4,18,22,23]. The high prevalence observed among the ANC attendees aged 30–39 years in Chelston in 1998 could, in part, be due to the fact that, for this age-group, the proportion of highly educated women (i.e., more than 7 years of school attendance) was much higher in 1998 than in 1994 and 1996, a finding that was not observed for the other age groups (data not shown). As the Chelston surveys showed that the level of education was positively associated with HIV infection in this age group, a selection bias toward an overestimation could have been introduced .
Regarding the actual use of the first method, it can be easily applied, in that the only data required are the age of antenatal clinic attendees and the relative age distribution of the general female population, which could be approximated using census data or official population projections. Regarding the standard data on age-specific differences in fertility risk by HIV serostatus, there arises the issue of whether or not data from one geographical area can be considered as representative of other areas, especially for women aged 15–19 years, among whom this difference could strongly depend on age at first sexual intercourse. When considering data from other areas of Africa, the pattern of age-specific differences in fertility by HIV serostatus are broadly similar [10,24], suggesting that the age-specific ratios of fertility might not differ greatly by geographical area and that the data on fertility risks from the Masaka cohort study could be used in other areas of sub-Saharan Africa where the HIV epidemic is mature and the use of modern contraceptives is not highly diffused. However, this issue is the main weak point of the method, and additional studies, possibly based on large samples and the same study designs, will be needed to evaluate the geographical homogeneity of these ratios.
In addition to age, other factors potentially associated with both HIV infection and fertility, such as educational level and marital status, could be taken into consideration in order to bridge the residual gap observed between the adjusted estimates and the HIV prevalence among the general female population. The striking diversity in HIV prevalence trends by level of education found in the Chelston surveys is of relevance in this regard . However, in the literature, the data on fertility risks by HIV serostatus are not stratified by these factors, and stratified data would be more sensitive to geographical variations, thus limiting the applicability of the adjustment method.
Furthermore, in cases where pregnant women have a low rate of attendance at ANC, all of the factors potentially associated with both HIV infection and ANC attendance could introduce a bias in the adjusted estimates based on ANC data. However, in many countries of sub-Saharan Africa, more than 90% of pregnant women have been reported to attend ANC [21,25,26], reducing to a minimum the possible selection bias due to assuming ANC attendees as representative of all pregnant women.
Since the first method takes into account the relative risk of fertility in order to control for differences by HIV serostatus in the probability of giving birth, applying the method to data from pregnant women instead of women who have given birth could result in an overestimation because HIV-positive pregnant women usually have a higher rate of abortion than HIV-negative pregnant women . However, given that most abortions are likely to occur in the early months of pregnancy and that pregnant women in sub-Saharan Africa usually visit a clinic for the first time relatively late in their pregnancy (median time in Zambia, Uganda and Tanzania more than 5.5 months into the pregnancy) [21,25,26], using data from ANC attendees instead of women who have given birth is expected to produce a minimum bias.
The second method of adjustment produced accurate estimates of the HIV prevalence among the general female population when applied to data from Ndola and rural Mwanza, although, in the latter case, the national population distribution by fertility risk category and parity, which was used as standard reference, may not be strictly representative of the rural areas of Tanzania . By contrast, the adjustment method produced less accurate estimates when applied to data from Chelston, particularly for the year 1998. As discussed for the first method, the unusually high prevalence observed among the ANC attendees aged 30–39 years in 1998 could partially explain this result. In fact, a biased high prevalence among the oldest ANC attendees would reflect a biased high prevalence among multipara pregnant women, thus leading to the HIV prevalence being overestimated among mothers, who represent most of the general female population in urban Zambia (69.7%) .
This second method was originally designed to also adjust for differences in subfertility by HIV serostatus by weighting data observed among multipara ANC attendees by their last birth interval, measured in years . This would reduce to a minimum the bias created by assuming that multipara pregnant women are representative of all fecund sexually active mothers in the general female population. However, data on the last birth interval were not available, and thus we did not adjust for subfertility. Nonetheless, adjusting for subfertility would most likely increase the estimated prevalence because it entails taking into consideration the fact that HIV-positive women with reduced fertility, yet still capable of bearing children, are under-represented in ANC. Thus, adjusting for both subfertility and infertility would produce an even greater overestimation than that obtained when we adjusted for infertility alone in Chelston. Furthermore, when the method was previously applied to data from Uganda and Tanzania , subfertility was found to account for a relatively small part of the difference between the non-adjusted and the adjusted prevalence.
Although the second adjustment method could be improved by applying it separately by age group, this would be quite difficult because of the lack of robust age-specific data on relative HIV prevalence by parity and fertility risk category. Finally, as for the first method, this method could be affected by a selection bias due to differences in ANC attendance between HIV-positive and HIV-negative pregnant women, as well as by the possibility that the standard HIV prevalence ratios estimated in Kisesa and Masaka are not highly representative of other geographical areas.
In conclusion, the results of this study suggest that data provided by HIV sentinel surveillance, which is usually conducted among ANC attendees  in the attempt to estimate the HIV prevalence among the general female population, could be more accurately interpreted using these methods of adjustment in settings where the HIV epidemic is mature and the use of modern contraceptives is not highly diffused. However, additional analyses based on data from other sub-Saharan countries are needed to evaluate further the performance of these adjustment methods and, in particular, to evaluate the geographical homogeneity of the data assumed as standard references in these analyses.
The authors are grateful to M. Kanieff for editorial assistance, to D. Pino for linguistic revision, and to M. G. Dente, P. Tancredi, A. Ranghiasci, A. Di Vincenzo and E. Costabile for their helpful support. The authors would also like to thank all of the researchers who indirectly contributed to this manuscript by conducting the studies on which it is based.
Sponsorship: Supported by ISS Uganda AIDS Project no. 667 and the Norwegian Agency for Development Co-operation (NORAD).
1.United States Bureau of Census. HIV/AIDS Surveillance Data Base.
Washington DC: U.S. Bureau of the Census, 2000.
2.Kwesigabo G, Killewo JZ, Sandstrom A. Sentinel surveillance and cross sectional survey on HIV infection prevalence: a comparative study.East Afr Med J
3.Fylkesnes K, Ndhlovu Z, Kasumba K, Musonda RM, Sichone M. Studying dynamics of the HIV epidemic: population-based data compared with sentinel surveillance in Zambia.AIDS
4.Glynn JR, Buve A, Carael M, Musonda RM, Kahindo M, Macauley I, et al.Factors influencing the difference in HIV prevalence between antenatal clinic and general population in sub-Saharan Africa.AIDS
5.Fylkesnes K, Musonda RM, Sichone M, Ndhlovu Z, Tembo F, Monze M. Declining HIV prevalence and risk behaviours in Zambia: evidence from surveillance and population-based- surveys.AIDS
6.Kilian AHD, Gregson S, Ndyanabangi B, Walusaga K, Kipp W, Sahlmuller G, et al.Reductions in risk behaviour provide the most consistent explanation for declining HIV-1 prevalence in Uganda.AIDS
7.Gray RH, Wawer MJ, Serwadda D, Sewankambo N, Li C, Wabmire-Mangen F, et al.Population-based study of fertility in women with HIV-1 infection in Uganda.Lancet
8.Carpenter LM, Nakiyingi JS, Ruberantuari A, Malamba SS, Kamali A, Whitwhort JAG. Estimates of the impact of HIV infection on fertility in a rural Ugandan population.Health Trans Rev
9.Kigadye RM, Klokke A, Nicoll A, Nyamuryekung'e KM, Borgdorff M, Barongo L, et al.Sentinel surveillance for HIV-1 among pregnant women in a developing country: 3 years’ experience and comparison with a population serosurvey.AIDS
10.Zaba B, Gregson S. Measuring the impact of HIV on fertility in Africa.AIDS
11.Ross A, Morgan D, Lubega R, Carpenter LM, Mayanja B, Whitworth JAG. Reduced fertility associated with HIV: the contribution of pre-existing subfertility.AIDS
12.Brocklehurst P, French R. The association between maternal HIV infection and perinatal outcome: a systematic review of the literature and meta-analysis.Br J Obstet Gynaecol
13.Lessner L. Projection of AIDS incidence in women in New York State.Am J Public Health
14.Boisson E, Nicoll A, Zaba B, Rodrigues LC. Interpreting HIV seroprevalence data from pregnant women.J Acquir Immune Defic Synd Hum Retrovirol
15.Nicoll A, Stephenson J, Griffioen A, Cliffe S, Rogers P, Boisson E. The relationship of HIV prevalence in pregnant women to that in women of reproductive age: a validate method for adjustment.AIDS
16.Zaba BW, Carpenter LM, Boerma TJ, Gregson S, Nakiyingi J, Urassa M. Adjusting ante-natal clinic data for improved estimates of HIV prevalence among women in sub-Saharan Africa.AIDS
17.Fabiani M, Accorsi S, Lukwiya M, Rosolen T, Ayella EO, Onek PA, et al.Trend in HIV-1 prevalence in an antenatal clinic in North Uganda and adjusted rates for the general female population.AIDS
18.Changalucha J, Grosskurth H, Mwita W, Todd J, Ross D, Mayoud P, et al.Comparison of HIV prevalences in community-based and antenatal clinic surveys in rural Mwanza, Tanzania.AIDS
19.Gregson S, Terceira N, Kakowa M, Mason PR, Anderson RM, Chandiwana SK et al.Study of bias in antenatal clinic HIV-1 surveillance data in a high contraceptive prevalence population in sub-Saharan Africa.AIDS
20.Boerma JT, Urassa M, Senkoro K, Klokke A, Ng'weshemi JZL. Spread of HIV infection in a rural area in Tanzania.AIDS
20.Central Statistical Office [Zambia], Ministry of Health, Macro International Inc. Zambia Demographic and Health Survey 1996
. Lusaka and Maryland: Central Statistical Office [Zambia], Ministry of Health, Macro International Inc.; 1997.
22.Fabiani M, Ayella EO, Blè C, Accorsi S, Dente MG, Onek PA, et al.Increasing HIV-1 prevalence among pregnant women living in rural areas of the Gulu District (North Uganda).AIDS
23.STD/AIDS Control Programme: HIV/AIDS surveillance report – June 2000.
Kampala: Ministry of Health; 2000.
24.Desgrees du Lou A, Msellati P, La Ruche G, Welffens-Ekra C, Ramon R, Dabis F. Estimation of HIV prevalence in the population of Abidjan by adjustment of the prevalence observed in antenatal centres.AIDS
25.Statistics Department and Macro International [Uganda]. Uganda Demographic and Health Survey, 1995.
Calverton, Maryland: Statistics Department and Macro International; 1996.
26.Bureau of Statistics and Macro International Tanzania. Tanzania Demographic and Health Survey, 1996.
Calverton, Maryland: Bureau of Statistics and Macro International; 1997.