Background: Antenatal clinic (ANC) surveillance is the primary source of HIV prevalence estimates in low-resource settings. In younger women, prevalence approximates incidence. Sexual behavior monitoring to explain HIV distribution and trends is seldom attempted in ANC surveys. We explore the use of marital history in ANC surveillance as a proxy for sexual behavior.
Methods: Five ANC clinics in a rural African district participated in surveillance from 1999 to 2004. Unlinked anonymous HIV testing and marital history interviews (including age at first sex and socioeconomic variables) were conducted. Data on women aged <25 years were analyzed.
Results: Inferred sexual exposure before marriage and after first marriage increased the adjusted odds of infection with HIV by more than 0.1 for each year of exposure. Increasing years within a first marriage did not increase HIV risk. After adjusting for age, women in more recent birth cohorts were less likely to be infected.
Conclusions: Marital status is useful behavioral information and can be collected in ANC surveys. Exposure in an ongoing first marriage did not increase the odds of infection with HIV in this age group. HIV prevalence decreased over time in young women. ANC surveillance programs should develop proxy sexual behavior questions, particularly in younger women.
From the *Department of Epidemiology and Public Health, London School of Hygiene and Tropical Medicine, London, United Kingdom; and the †Karonga Prevention Study, Malawi, Africa.
Received for publication October 4, 2007; accepted March 3, 2008.
Correspondence to: Amelia Catharine Crampin, MBChB, MSc, Department of Epidemiology and Public Health, London School of Hygiene and Tropical Medicine, Keppel, London, WC1E 7HT United Kingdom (e-mail: email@example.com).
After rapid increases in the 1980s and early 1990s, HIV prevalence has recently stabilized or declined in several high-prevalence countries in sub-Saharan Africa.1,2 There is a need to document and explain trends in HIV prevalence3-5 and to relate the changes to behavior, particularly when trends are inconsistent between countries and population groups, and to establish whether changes in prevalence are accompanied by behavior change5. “Second-generation” surveillance aims to explain and describe the changes in HIV prevalence3,6 to inform policy directly of changes in risk behavior.
To date, antenatal clinic (ANC) surveillance has been the main source of HIV prevalence trend data. It has not been used for sexual behavior monitoring, because detailed inquiry has been considered time-consuming and intrusive in the context of unlinked anonymous surveillance within an essential clinical service,7 despite the fact that it is important to examine sexual behavior in the same population from which HIV prevalence data are collected. Much effort has recently gone into designing improved sampling frames for anonymous ANC surveillance to ensure wide coverage and to minimize self-selection in multiple-provider settings or when services such as prevention of mother-to-child transmission (PMTCT) programs offer named testing. It is therefore important to maximize the usefulness of data collected.
Marital history may be a good proxy for sexual behavior insofar as it allows identification of periods of varying exposure to HIV, for example, relatively low-risk periods (stable first marriage) and higher risk periods (eg, premarital exposure, marriages that break up, remarriages).
Among women aged <25 years, HIV prevalence is a reasonable proxy for incidence, because infections are relatively recent and HIV-related mortality is relatively low. Marital stability in this age group is also a sensitive indicator of community behavior change, because an appreciable proportion of the group is replenished each year by newly sexually active entrants.
This article explores the use of ANC surveillance to infer behaviors relevant to HIV transmission in rural northern Malawi, where the Karonga Prevention Study (KPS) has been conducting epidemiologic studies of HIV since the 1980s.8 Since 1999, these studies have included ANC sentinel surveillance in 5 health facilities in Karonga district, using a variant of Malawi's national data collection protocol.9 Data presented here demonstrate the utility of ANC surveillance in addressing the relation between sexual behavior patterns and observed trends in HIV prevalence.
Between 1999 and 2004, ANC unlinked anonymous surveillance was conducted in 5 main health facilities: 1 district hospital (urban), 2 rural hospitals (1 semiurban and 1 rural), and 2 health centers (rural) of Karonga district, northern Malawi. From 2003, data collection ceased in clinics on an individual basis because of the introduction of PMTCT programs, which complicated anonymous data collection.
Questionnaires were anonymized, and dual HIV testing was carried out using enzyme-linked immunosorbent assay (ELISA) and particle agglutination assay.9 Data collection included standard questions, age at start of current marriage, age at start of first marriage (AAFM), and (from 2002 onward) age at first sex (AAFS). “Marriage” was defined by the respondent; in this population, it usually refers to a union after a traditional ceremony or a community-acknowledged state of cohabitation. “First sex” was defined as first penetrative sex, and the interviewers (who were all female, with study-specific training and completion of primary education) were trained to ensure that this was correctly established in the vernacular. Interviews were carried out in private.
Approval for the study was granted by the Malawi National Health Sciences Research Committee and the Ethics Committee of the London School of Hygiene and Tropical Medicine.
Data were double-entered and verified in FoxPro (Microsoft Corp, Redmond, WA) and managed and analyzed in Stata (version 8.1; Stata Corportion, College Station, TX). Women were grouped for analysis by area of current residence, having been coded by village groups to avoid inadvertently identifying individuals.
Years spent in each of 3 separate “sexual exposure” categories were calculated for each woman. Duration of premarital exposure was calculated as [current age - AAFS] for those never married and [AAFM - AAFS] for those ever married, representing a period when a woman may have multiple unstable partnerships. Duration of first marriage exposure was calculated as [current age - AAFM] for those still in their first marriage and as [age at end of first marriage - AAFM] for those whose first marriage had ended and represents a period when sexual exposure is likely to be confined to a single partner and to be related to child-bearing. Marital status questions identified women whose first marriage had ended. Age at end of first marriage was not recorded, however, so this was estimated by allowing 2 years of marriage for every child born in that marriage or, when there were no children, allocation to the marriage of two thirds of the interval between the start of the first marriage and the start of a second marriage and the remaining one third to an intermarriage period or four fifths of the interval between the start of the first marriage and current day for those who had not remarried (based on observations from large demographic surveillance studies of the KPS, which link children and spousal relationships).10 Time after first marriage included all other periods since sexual debut. This is another period that may include multiple sexual partners and may involve economic dependence on sexual partners, once away from the parental home. For the calculation of rates, time after the first marriage was divided further into remarried and ex-married periods according to whether the time spent after the first marriage was spent married or single.
Information on AAFS was incomplete, because the question was not included in the first version of the ANC questionnaire (1999 to 2001). Upper bounds for the start of sexual activity have been established for those with missing or inconsistent information, however, because exposure to sexual activity must have occurred at least half a year before the ANC visit for women expecting their first birth [current age - 0.5], and at least 9 months than age at first birth (AAFB) for those who had previously given birth [AAFB - 0.75]. For married women, sexual activity can be presumed to start at or before first marriage, and time spent in premarital sexual activity was estimated for missing data using the predicted median premarital exposure based on AAFS to AAFM using a Poisson regression method. Thus,
The HIV prevalence in ANC women aged 25 years and older in each residence area was used as a proxy for background HIV prevalence in the population from which the young women drew their partners.4,11
Logistic regression of HIV status was undertaken against continuous measures of exposure (eg, time spent at risk and background HIV prevalence) and standard categoric and binary variables (eg, current marital status and parity; behavioral and health indicators such as past family planning use; past experience of child mortality; and sociodemographic variables such as education, area of residence, and age of father of the child).
A classification of “low parity for age” was devised to include women in the top quartile of the age range for a given parity; for the study group of interest, this comprised those aged 20 years or older who were expecting their first birth.
Less than 1% of women seen at the ANC clinics refused the interview or venipuncture. ANC data (interview and serology) were available on 2874 women younger than 25 years of age over the 6-year period. This excluded 13 women resident outside the district, 1 woman with a missing birth year, and 44 ever married women with missing age at marriage. Age at marriage data were internally consistent with other age-related events, such as birth of children within marriage, for all but 5 women, who were excluded. AAFS was not available for 1679 women interviewed between 1999 and 2001 and was missing (or inconsistent) for 547 of 1195 women interviewed in 2002 to 2004. Those not reporting AAFS when questioned directly were mainly (60%) married women, older than 20 years of age, with recall difficulty. Table 1 shows the age distribution and mean years of sexual exposure at each age, comparing the mean duration of sexual exposure among those for whom AAFS was imputed and those for whom it was measured directly. Data for earlier and later periods are presented separately to reveal period factors affecting imputed ages at first sex. Only 21 women aged 18 years or younger reporting in 2002 to 2004 had imputed AAFS.
Over the time period examined (1999 to 2004), the age distribution of those younger than 25 years of age at interview did not change (mean = 20.0 years), nor did mean ages at first marriage (17.5 years) or first birth (18.1 years).
The distribution of mean time spent in each sexual exposure category for women aged younger than 25 years at interview showed a plausible pattern with age (Fig. 1). Premarital exposure was similar across all ages, with a mean of just >1 year; exposure after end of first marriage only becomes important after 20 years of age. Unadjusted HIV prevalence is shown in Figure 2 by duration in each sexual exposure category. There is a progressive increase in HIV risk associated with longer durations (more than 4 years) of premarital exposure and longer duration since a first marriage ended. Increasing duration of sexual exposure in a first marriage was not associated with increased risk of HIV infection in this age group.
To refine the estimates of the effect of marriage status, other risk factors were examined for women younger than 25 years of age. Crude HIV prevalence and odds of HIV infection are shown in Table 2. Increasing age, area of residence, current marital status, birth cohort, parity for age, experience of child death, age of sexual partner, evidence of syphilis, previous use of contraception, previous residence outside the district, and secondary education were all significantly associated with HIV infection.
Variables were entered in a full multivariate model if the odds ratios (ORs) of at least 1 exposure category reached statistical significance or if the variables were part of exhaustive classifications in which the other categories had statistically significant ORs. When linear trends could be verified using appropriate tests, categoric variables were replaced with their continuous equivalents (eg, age, years of exposure in different marital status categories, year of birth, year of interview). Age, interview year, and year of birth cannot all be included in the full model because of colinearity, and year of birth provided a stronger measure of trend (birth cohort effect) than interview year (calendar time).
After allowing for duration of exposure since sexual debut, classified into exclusive components (premarital, first marriage, and after first marriage), age lost its explanatory power as a measure of exposure. Marital status at the time of interview had no additional explanatory value when duration of exposure is controlled for in this way. Low parity and previous child deaths were excluded from the multivariate model in Table 3, because these may be consequences rather than risk factors of HIV infection.
In the crude and multivariate adjusted models, area of residence had a strong effect on the odds of HIV infection. This was quantified by explicitly including measures of the background prevalence (assumed to reflect the likelihood that sexual partners had HIV). The multivariate model showed that the OR for HIV infection in an individual woman younger than 25 years of age increased by 0.04 with each increase of 1% in HIV prevalence among ANC attendees aged 25 years and older resident in the same area, after adjusting for all other factors. The adjusted odds of HIV infection decreased by 0.09 for each 1-year increase in calendar year of birth. Additional years of premarital exposure or exposure after first marriage each added independently 0.2 and 0.3 per year, respectively, to the odds of infection. After adjustment for other factors, the increased odds of infection associated with each year in a first marriage seen in the crude model were lost. Sensitivity analyses demonstrated that the odds of HIV infection for sexual exposure in and out of marriage were robust to the assumptions made about time of end of first marriage (not shown).
For each additional year of the unborn child's father's age, the adjusted odds increased by 0.04 and the effects of previous residence outside the district and secondary school education remained after adjustment for other factors.
To examine the effects of the imputed age at first sex, the model was run using those with reported AAFS alone. Although the regression results in this limited sample were not all statistically significant, the ORs for the smaller sample generally were within the confidence intervals for the corresponding values in the overall sample. In the full model, adding “imputation of AAFS” as an explanatory variable did not change results for the other variables and was not a statistically significant risk factor by itself.
These data and analyses demonstrate a novel use of ANC surveillance data to reveal changes in HIV infection risk associated with marital history in a rural African community at a time when HIV prevalence is stable and may be starting to decline. Inclusion of marital history questions did not seem to affect participation rates adversely (<1% refusals), although some women may have avoided clinics during recruitment periods. The study design did not allow assessment of the impact of different types of questioning on participation rates.
The data show no evidence of continued risk of HIV acquisition for women younger than 25 years of age in first marriages but that extended premarital exposure, subsequent extramarital exposure, or exposure during subsequent marriages are all associated with increased risk. This is consistent with the findings in urban Kenya and Zambia that less than half of HIV in young married women is acquired within the marriage,12 although analyses of those data concluded that early marriage puts young women at higher risk than intermittent contact associated with premarital sex.13 Studies finding that never married women have lower HIV prevalence than their married age mates may not discriminate between virgins and sexually active single women.14
The lack of ongoing risk in a first marriage may be, in part, an artifact of the cross-sectional study design, because introduction of HIV over time may result in subfertility or mortality, removing individuals from the sample. It is also consistent with the relative sexual exclusivity of a stable first marriage: unfaithfulness, domestic violence associated with alcohol use, and other factors associated with HIV risk also precipitate marital breakup.
The insights these data give into sexual behavior and associated risk of HIV demonstrate the usefulness of simple interviews during ANC surveillance. As with all ANC-based surveillance, results are only generalizable to women who conceive. AIDS is likely to be responsible for much of the widowhood in women younger than 25 years of age, thereby lowering conception risks for women with a high probability of being infected. Relative age of a current sexual partner, use of contraception, and low parity for age are also associated with increased risk. Other community-based data from this project reveal that although more than 99% of women attend an ANC when carrying an established pregnancy, approximately one third of all women of child-bearing age have not attended an ANC in the 4 years before being surveyed.15 Nearly 90% of the ANC nonattenders have never had a child (with the remainder having unusually long birth intervals), and although a third of these are virgins, there are many women with low fertility or whose lifestyles preclude stable relationships and child-bearing, who are at relatively high risk of HIV and need to be studied independently.10,16
There are data quality limitations for marital history and premarital sex questions in ANC interviews, and selective reporting of behavior favored by health education messages may be common in this setting. Health education intensifies over time and younger women may feel less inclined to report actual rather than “acceptable” behavior (eg, reporting informal relationships resulting in pregnancy as marriage). The lack of a question on the date when the first marriage ended made it necessary to impute time spent in and after the terminated first marriage in divorcees or widows. Nevertheless, given those caveats, this study presents clear and plausible findings on recent behavior among young sexually active women in this community.
These data show a downward trend of HIV prevalence over time in women younger than 25 years of age, which remains significant when other factors are controlled. The introduction of PMTCT programs to the district did not influence this trend as a real effect or artifactually, because introduction was low key and piecemeal and data collection was not conducted in any clinics with established PMTCT services. We looked at 2 ways of quantifying this trend: first, looking at change in the OR of infection in relation to calendar year of interview, and, second, looking at the effect of the women's birth cohort (year of birth). The fact that the latter exerts a stronger effect suggests that younger birth cohorts adopt safer sexual behavior more quickly, even after allowing for their reduced length of exposure to marital and extramarital sex.
The unmeasured favorable behavior change could be fewer sexual partners, less frequent contacts with high-risk partners, or increased condom use. Direct questions on these behaviors were introduced only in the latter part of the study; thus, insufficient data have been collected to allow for a more complete analysis. Although these questions were completed, interviewers reported an element of reluctance or concealment in answering these direct enquiries, as, indeed, there may have been when reporting AAFS, although this was not pronounced.
The difficulty in elucidating reliable responses to specific sexual behavior questions in this high-throughput clinical setting emphasizes the value of informative but less controversial questions such as marital history and age at sexual debut, which may be less prone to social desirability biases and easier to check. Quality of data, particularly on the latter question, should be improved with appropriate training of interviewers in language and attitudes, by ensuring privacy during interviews, and by emphasizing confidentiality and anonymity to the client.
In this population, premarital sex is the norm and pregnancy is a common precipitant for marriage.17 This is reflected in the contribution of up to 10 years of premarital sexual exposure in the group studied, carrying a year-on-year risk of HIV that was as high as exposure of women in vulnerable inter- or postmarriage periods of their lives. Premarital sex for teenage girls may occur with age mates (eg, school friends) or with older men. In the teenage women sampled here (who may or may not have been married by the time of presentation at the ANC), 40% of pregnancies were reported to be by a man from an age cohort at least 5 years older than the women's age cohort. The corresponding figure is 29% for women in their early twenties.
Qualitative work in this population suggests that women 16 years of age or younger have premarital sex only on isolated occasions: unexpected and irregular (consequently, unprotected).17 Women in their twenties are more likely to have had regular boyfriends before marriage.
Although there seems to have been a decline in HIV prevalence between the first 3 and last 3 years of this study, the significance of this overall trend cannot be inferred directly, because ANC surveillance was not continuous in all participating clinics. Areas of residence differ in their HIV infection prevalences, and some were not represented in the latter years.
The association of HIV with low parity for age may be causal in either direction: long-standing HIV infection is a biologic cause of infertility and a social cause of infertility (attributable to widowhood), and other sexually transmitted infections increase the risk of infertility and HIV acquisition. Primary or secondary infertility can destabilize marriage14 and lead to periods of risky exposure. The association of HIV with low parity in this young age group was not significant when adjusted for other factors.
We have shown that it is possible to collect useful data on sexual behavior in an ANC setting. Questions on AAFS may be answered less well in older than in younger women, but it is the age at sexual debut in the youngest groups that gives cogent information on trends, and the questions are generally well answered in this group. The same data collection tools, with the addition of dates of termination of marriages to avoid the need for imputation, can be used in PMTCT program settings to investigate whether women who choose to attend a specific facility in preference to another have different behavioral characteristics, which may explain any observed HIV prevalence differences by facility type. The Malawi National AIDS Commission used a variant of marital status questions developed by the KPS in its last round of ANC surveillance, and it is to be hoped that other ANC surveillance programs follow suit, with appropriate quality controls in place.
This article is dedicated to the late Masiya Kondowe. The KPS is supported by the Wellcome Trust and the British Leprosy Relief Association. J. R. Glynn was supported by the UK Department of Health. The authors thank the National Health Sciences Research Committee of Malawi and the National AIDS Commission of Malawi for their continued support.
1. Asamoah-Odei E, Garcia Calleja JM, Boerma JT. HIV prevalence and trends in sub-Saharan Africa: no decline and large subregional differences. Lancet
2. UNAIDS/WHO. AIDS Epidemic update. December 2007. UNAIDS, Geneva. In: AIDS Epidemic Updates
3. UNAIDS/WHO Working Group of Global HIV/AIDS and STI Surveillance. Guidelines for Second Generation HIV Surveillance: The Next Decade
. Geneva, Switzerland: UNAIDS; 2000.
4. Zaba B, Slaymaker E, Urassa M, et al. The role of behavioral data in HIV surveillance. AIDS
. 2005;19 (Suppl 2):S39-S52.
5. Garnett GP, Garcia-Calleja JM, Rehle T, et al. Behavioural data as an adjunct to HIV surveillance data. Sex Transm Infect
. 2006;82(Suppl 1):i57-i62.
6. Diaz T, De Cock K, Brown T, et al. New strategies for HIV surveillance in resource-constrained settings: an overview. AIDS
. 2005;19(Suppl 2):S1-S8.
7. Allen DM, Lee NC, Schuiz SL, Pappaisanou M, et al. Determining HIV seroprevalence among women in women's health clinics. Public Health Reports. 1990;105:133-134.
8. Glynn JR, Pönnighaus JM, Crampin AC, et al. The development of the HIV epidemic in Karonga District, Malawi. AIDS
9. Crampin AC, Glynn JR, Ngwira BM, et al. Trends and measurement of HIV prevalence in northern Malawi. AIDS
10. Jahn A, Crampin AC, Glynn JR, et al. Evaluation of a village-informant driven demographic surveillance system in Karonga, northern Malawi. Demogr Res
11. Auvert B, Buvé A, Ferry B, et al. Ecological and individual level analysis of risk factors for HIV infection in four urban populations in sub-Saharan Africa with different levels of HIV infection. AIDS
. 2001;15(Suppl 4):S15-S30.
12. Glynn JR, Caraël M, Buvé A, et al. HIV risk in relation to marriage in areas with high prevalence of HIV infection. J Acquir Immune Defic Syndr
13. Clark S. Early marriage and HIV risks in sub-Saharan Africa. Stud Fam Plann
14. Nilses C, Lindmark G, Munjanja S, et al. A community based study of HIV in women in rural Gutu District, Zimbabwe 1992 to 1993. Cent Afr J Med
15. Jahn A, Ngwira L, Chishambo C, et al. HIV prevalence and risks among women who do not attend antenatal clinic. Presented at: National HIV/AIDS Research Dissemination Conference; 2004; Lilongwe.
16. Hemmings JM, Banda D, Crampin AC, et al. Infertility and women's life courses in northern Malawi. Presented at: International Union for the Scientific Study of Population Conference; 2005; Tours.
17. Hemmings JM. Infertility and Women's Life Courses in Northern Malawi
[PhD thesis]. London, UK: London School of Hygiene and Tropical Medicine, University of London; 2007.
This article has been cited
Keywords:© 2008 Lippincott Williams & Wilkins, Inc.
antenatal clinic; HIV; Malawi; sexual behavior; surveillance