You are here

Article Text

Article menu

Download PDFPDF

Research report
A longitudinal population-based analysis of relationship status and mortality in KwaZulu-Natal, South Africa 2001–2011
  1. Melanie Channon1,
  2. Victoria Hosegood2,3,
  3. Nuala McGrath3,4
  1. 1Oxford Institute of Population Ageing, University of Oxford, Oxford, UK
  2. 2Department of Social Statistics and Demography, University of Southampton, Southampton, UK
  3. 3Africa Centre for Health and Population Studies, University of KwaZulu-Natal, Somkhele, South Africa
  4. 4Department of Social Statistics and Demography and Academic Unit of Primary Care and Population Sciences, University of Southampton, Southampton, UK
  1. Correspondence to Dr Melanie Channon, Oxford Institute of Population Ageing, University of Oxford, 66 Banbury Road, Oxford, OX2 6PR, UK; melanie.channon{at}ageing.ox.ac.uk

Abstract

Background Mortality risk is lower in married than in unmarried men and women. However, little is known about the association between mortality and relationship status in South Africa where marriage rates are low, migration is common, many couples are not co-resident and HIV prevalence is high.

Method Using demographic surveillance data collected from 2001 to 2011, relationship status was categorised as conjugal (partners belong to the same household), non-conjugal (partners do not belong to the same household) or not partnered. Rates of relationship formation and dissolution were calculated by age and sex. Controlling for antiretroviral treatment (ART) introduction in 2005 as well as education, sex-specific and age-specific Cox proportional hazards models were used to investigate the association between relationship status and (1) all-cause mortality and (2) non-AIDS mortality.

Results Before 2005, individuals in conjugal relationships had a lower hazard of all-cause mortality in all age groups than not partnered men and women. Non-conjugal relationships lowered the risk of dying compared with not partnered men and women in fewer age groups. After ART introduction, the protective association of conjugal relationships was weaker but remained generally significant for men and women but not in non-conjugal relationships. In the later period, the association is reversed in young men (20–29 years) with mortality higher in conjugal and non-conjugal relationships compared with men not partnered. The analysis of non-AIDS deaths provided similar results.

Conclusions The higher degree of social connections within a shared household environment that characterises conjugal relationships affords men and women greater protection against mortality.

  • MORTALITY
  • MARITAL STATUS
  • HIV

This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/

https://doi.org/10.1136/jech-2014-205408

Statistics from Altmetric.com

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

    Background

    It is well documented that in many contexts marital status has a strong association with mortality.1–4 It has generally been found that being married is associated with a lower mortality rate and that in many cases the association is stronger for males.5–7 It has also been found that the protective effect of marriage is weaker at older ages.8 ,9 However, with a few exceptions, little attempt has been made to study the effect of relationship status as a time-varying covariate; instead, marital status has typically been treated as fixed at a certain point in time,1 ,7 ,10 and very few studies consider the association between different types of relationship status (eg, non-marital) and mortality. Furthermore, we could not find any studies on the association between mortality and marital status (or social relationships more generally) in the African context as the majority of studies have been conducted in North America or Western Europe.3 ,4

    We use longitudinal, population-based data available in the Africa Centre Demographic Information System (ACDIS) to explore whether the protective association between relationship status and mortality holds in a rural South African context, and to extend our understanding of relationship formation and dissolution in a rural South African population.

    The level and patterns of HIV incidence and prevalence, HIV treatment and mortality have been well described in this population.11–13 In South Africa, marriage among some communities has been declining for decades, particularly in KwaZulu-Natal, where, by 2008/2009, 12% of women aged 20–45 years were currently married and fewer than one-third of women aged 45–49 years were ever married.14–16 Non-marital relationships are very common but cohabitation is not as common as in other settings and a variety of different residential arrangements exist between both marital and non-marital partners.14 Thus, in this setting, we argue that there is a need to define relationships in a different manner to the standard married/non-married dichotomy. We use the concept of a conjugal relationship, which is considered to be a sexual relationship where both partners are members of the same household (whether resident or non-resident), and thus this reflects a high level of social recognition.

    In this paper, we estimate formation and dissolution rates for men and women's relationships from 2001 to 2011 in KwaZulu-Natal and compare the risk of mortality for people in a conjugal relationship, a non-conjugal relationship and those not partnered by sex in the periods before and after HIV antiretroviral treatment (ART) was available.

    Data and methods

    Data

    The data used in this paper were collected as part of the ACDIS, an ongoing demographic surveillance system (DSS) in KwaZulu-Natal.17 The ACDIS collects information on all 89 000 members of households within the site; however, individuals can have multiple household memberships and/or be a member of a household without being resident in the same geographic location as the household.18 The reason for this is that defining households only in the sense of co-residence ignores the dynamic and complex nature of households in this area; non-resident members can have relationships with other household members (eg, spouse, parent, child), and the health and well-being of non-resident members may be closely related to the health and welfare of resident members and vice versa, and in the South African context non-resident adults make frequent return visits.18 ,19 Demographic data on members of the DSS are updated through a routine household visit which varied in frequency over time between two and three times a year.17

    At each routine household visit, the current relationship status of all individuals is recorded. However, the relationship status does not indicate who the partner is or when the relationship started. When the two partners are members of the same household, the couple are classified as a conjugal relationship and additional information is collected including the identity of the partner, the start date of the relationship, the date of customary or civil marriage (if any) and the end date (if applicable).

    Given that relationship status does not remain constant over time and our analysis was conducted over a period of more than a decade, we considered relationship status as time-varying in all our models. By considering relationship status as time-varying, we are specifically studying the association between mortality risk and current relationship status, whereas previous research has generally operationalised relationship (or marital) status for a fixed time point. Over long periods, treating relationship status as static would potentially lead to misleading results. At any point in time, there are three possible mutually exclusive relationship states: not partnered, in a non-conjugal relationship (ie, partnered but not linked in the data and not members of the same household) and in a conjugal relationship.

    Methods

    All adults aged 20 years and older on 1 January 2001 were followed up until death, the ending of any membership of a household resident in the study area, or 1 December 2011 when they were censored. Person-time in a conjugal relationship was calculated using the documented start and end dates of the conjugal relationship. For person-time in a non-conjugal relationship, it was assumed that time in this state started on the date his or her non-conjugal relationship was first reported during a routine household visit. A non-conjugal relationship was assumed to have ended on the date the routine household visit that documented the person's relationship status had reverted to unpartnered.

    Relationship formation and dissolution rates were calculated as the number of formations or dissolutions divided by the number of person-years experienced at risk of that event presented per 1000 person-years. The transitions between the three relationship states are summarised in figure 1. A relationship formation was defined to be a change from not being in a relationship to being in either a non-conjugal or a conjugal relationship. Changes from non-conjugal to conjugal relationships are assumed to be a change in stage of relationship, rather than a new relationship. Individuals in the denominator were not partnered, that is, at risk of an event. Dissolution is defined as a change from being in a non-conjugal or conjugal relationship to not being partnered through separation, divorce or widowhood. Person-time at risk of dissolution was calculated as the sum of all periods where an individual was in a relationship (conjugal or non-conjugal).

    Figure 1
    Figure 1

    Illustration of possible transitions between relationship states.

    In order to model the association between relationship status and mortality, we used Cox proportional hazards models.20 The proportional hazards assumption was tested using Schoenfeld residuals.21 Relationship status was included in the models as a time-varying covariate. All of the multivariate analyses controlled for age (using linear and quadratic terms) and highest reported educational attainment (as a categorical variable). Educational attainment was split into five categories: <1 year, primary school, secondary (not matriculated), matriculated or higher and unknown/missing. Public roll-out of HIV treatment began in the study area in November 200422 following which adult mortality declined.11 Therefore, we examine the association between relationship status and mortality before and after 2005 using time-varying coefficients.23

    We also repeated the analyses for non-AIDS deaths only where individuals were censored at the time of their AIDS death (a competing risk analysis). The rationale for focusing on non-AIDS deaths is to provide a direct comparison with previous studies on marital status and mortality in low general HIV prevalence contexts. Furthermore, HIV and ART status are not available for both partners in most cases and it would not be possible to appropriately adjust for confounding by the timing of HIV status and ART initiation within couples. Previous studies have found mixed results with marriage associated with both increased24 ,25 and decreased26 HIV risk.

    Results

    In January 2001, 32% of women were not partnered (table 1), 42% were in a non-conjugal relationship and 26% were in a conjugal relationship. However, this hides a great deal of variation by age, with 70% of those aged 20–29 years in a non-conjugal relationship compared with <10% of those aged 60 and over. Those aged 60 and over were particularly likely to be not partnered, while those aged 50–59 years were most likely to be in a conjugal relationship.

    View this table:
    Table 1

    Distribution of relationship status on 1 January 2001 in the ACDIS, by age and sex

    Men were generally more likely to be in a conjugal relationship than women, especially at older ages. This is the consequence of women's longer survival and their partners dying before them, especially since in more than 85% of conjugal relationships the woman was younger than the man with a mean age difference of more than 6 years. Men were much less likely than women to be in a relationship in the 20–29 year age group, but at older ages men were more likely to be in a relationship. Men aged 20–24 years were especially unlikely to be in any form of relationship, with only 2% in a conjugal relationship and 45% not partnered. Two-thirds of men aged 25–29 years were in a relationship; however, these relationships were predominantly non-conjugal.

    Overall, 64% of couples in a conjugal relationship on 1 January 2001 were cohabiting and 58% were married. For older couples in a conjugal relationship, cohabiting was more common (78% among those aged over 60 years), while for younger couples the most common reason for not cohabiting was that the male was outside the DSS. Marriage also varied by age with around 80% of conjugal couples being married in the 60 and over age group, but less than a fifth in the age group of 20–29 years.

    Table 2 shows age-specific formation and dissolution rates of relationships. Given the sex differentials in relationship status by age highlighted in table 1, sex-specific formation and dissolution rates were calculated. Figure 2 shows point estimates for these rates with 95% CIs. Formation rates were highest for females aged 20–29 years and males aged 30–39 years. For both sexes, formation rates declined substantially in the older age groups, although for those aged 60 years and older relationship formation was much more common for males than for females. Indeed, for those aged 60 years and older, the relationship formation rates for males and females were 93 and 12/1000 person-years, respectively, and this difference was statistically significant.

    View this table:
    Table 2

    Relationship formation and dissolution patterns by age and sex in the Africa Centre Demographic Information System, 2001–2011

    Figure 2
    Figure 2

    Age-specific relationship formation and dissolution rates, by sex.

    Dissolution rates peaked for men in the youngest age group and women in the oldest age group. The high dissolution rates seen at either end of the age range are most probably related to different processes. Dissolutions in the younger age group are associated with relationship instability, whereas for women at older ages dissolution is primarily due to the death of their partner.

    Table 3 shows Cox proportional hazards models separately by age category and sex. Before 2005, being in a conjugal relationship was associated with a lower hazard of all-cause mortality for both females and males in all age groups compared to being not partnered. The protective association with being in a conjugal relationship was substantial with hazard rates for females 37–80% lower and hazard rates for males 38–66% lower. After 2005, the protective association with being in a conjugal relationship was substantially weaker but remained significant for females in all age groups, except those aged 20–29 years. Similarly, for men, being in a conjugal relationship remained significantly protective for males aged 40 years and older compared to being not partnered, but was no longer significant for men aged 30–39 years, and was associated with a significantly increased risk of mortality for males aged 20–29 years.

    View this table:
    Table 3

    Cox proportional hazards models for the effect of relationship status on all-cause mortality, by age and sex in the Africa Centre Demographic Information System, 2001–2011

    The association between being in a non-conjugal relationship and mortality is more complex. A significant protective effect of being in a non-conjugal relationship on the hazard of mortality before 2005 was observed in younger women, specifically those aged 20–29 and 30–39 years. In contrast, women aged 60 and older in a non-conjugal relationship had a significantly higher risk of mortality before 2005 than women of the same age who were not partnered. After 2005, being in a non-conjugal relationship was not associated with any significant change in the hazard of mortality compared to being not partnered.

    For men, being in a non-conjugal relationship was associated with a significantly lower hazard of mortality before 2005 compared to being not partnered, among those aged 30–39 and 40–49 years only. After 2005, being in a non-conjugal relationship remained significantly protective compared to being not partnered only for those aged 40–49 years, and was associated with a significantly increased hazard of mortality for those aged 20–29 years.

    Table 4 shows equivalent Cox models for non-AIDS deaths only. While precision in some age groups was reduced due to the smaller number of deaths, the HR estimates for men and women in conjugal and non-conjugal relationships were consistent with the size and pattern of estimates from the all-cause analysis, but not always their statistical significance. For men aged 20–29 years, the increased risk of non-AIDS mortality associated with being in a conjugal or non-conjugal relationship was attenuated and the estimates were no longer statistically significant.

    View this table:
    Table 4

    Cox proportional hazards models for the effect of relationship status on non-AIDs mortality, by age and sex in the Africa Centre Demographic Information System, 2001–2011

    All models control for highest educational attainment, the inclusion of which made little substantive difference to the association between relationship status and mortality. Education was strongly related to mortality risk, with those who were least educated having a higher mortality risk as well as those whose educational attainment was unknown or missing. The models were also run excluding the missing category, but this did not substantively change the results.

    It should be noted that we conducted a sensitivity analysis using different possible cut-off points in recognition that ART initiation and follow-up were rolled out from November 2004 to late 2007. The results did not change substantially; however, by using a cut-off point later than 2007, both changed the results and violated the proportional hazards assumption.

    Discussion

    Studies conducted elsewhere have found that marriage has a protective effect on adult mortality.2–4 However, these studies were conducted in settings with lower levels of adult mortality than this study community in northern KwaZulu-Natal, where until the introduction of ART in 2005 the severe HIV epidemic led to very high levels of adult mortality.11–13 Given that there has been a well established and substantial decline in the proportion of married Zulu-speaking men and women,14 ,16 we asserted that it was appropriate to examine differentials in mortality risks by types of relationships other than marital and non-marital. In the context of high levels of migration as well as low rates of marriage, we used population-based data collected prospectively, and categorised relationships by the extent of social connectedness based on whether or not the partners were members of the same household, and compared these individuals with those not partnered. Furthermore, we asserted that any mortality differentials by relationship status are likely to vary across the periods before and after treatment roll-out. This is the first investigation of the association between mortality and conjugal and non-conjugal relationships in an African population.

    Consistent with other studies in Africa, dissolution rates increased with age.27 Relationship formation peaks in men at a later age than in women, with men being more likely to form new relationships at older ages than women of the same age.

    Before 2005, individuals in conjugal relationships had a lower hazard of all-cause mortality in all age groups than not partnered men and women. In contrast, non-conjugal relationships were associated with a lower hazard of death in only a few age groups of women (20–29, 30–39 and 60 years and over) and men (30–39, 40–49 and 60 years and over). After ART introduction, the protective association of conjugal relationships was weaker but remained generally significant for men and women but not in non-conjugal relationships. In the later period, the association was reversed in young men (20–29 years) with mortality higher in conjugal and non-conjugal relationships compared with men not partnered. In a context where marriage rates are low and cohabitation is not universal, the findings suggest that conjugal relationships, with their higher degree of social connection through shared membership of the same household, affords men and women with greater protection against mortality.

    The results of the analyses of non-AIDS mortality for men and women are similar to the pre-2005 and post-2005 estimates for all-cause mortality. The post-2005 estimates for young men (20–29 years) in conjugal and non-conjugal relationships were attenuated in the analysis of non-AIDS mortality but remained suggestive of an increased risk of mortality compared with those not partnered. Partnering in young men (20–29 years) is common. Post-2005, the proportion of young men in a conjugal relationship was small (15% of person-time), whereas about 70% of person-time was spent in a non-conjugal relationship. Union dissolution rates are higher for men in this age group compared with other age groups; however, we would not anticipate that instability in relationships would influence the estimates of mortality risk given the time-varying approach used in this analysis. Instead, we interpret this finding as suggesting that young men who have entered conjugal and non-conjugal relationships by this age have characteristics that differ more markedly from those not partnered than at older ages. Furthermore, these characteristics, for example, socioeconomic status (SES), may also be associated with increased mortality.28 ,29 The increased risk of non-AIDS deaths in this age group of men is primarily due to accidental and intentional injuries.30

    There are some limitations to our approach. It is possible that previous relationship status impacts mortality risk for a period after the status changes; however, we would expect that effect to diminish with time and have not incorporated lagged effects in our models. Given that the identity of the partner is not available in the ACDIS for those in a non-conjugal relationship, it was necessary to consider that a change in status from a non-conjugal to conjugal relationship was with the same partner, though this could occasionally be with a different partner. Similarly, on occasion, serial reports of a non-conjugal relationship could relate to different partners. Thus, our estimated formation and dissolution rates may underestimate the underlying rates in the population. In addition, we are only able to consider primary relationships because data regarding multiple partnerships and the number of lifetime partners for individuals or their current partners are not available. In the literature, it has been debated whether the concurrency behaviour of an individual's partners increases his or her own risk of HIV infection,31–33 while the number of lifetime partners is an established risk factor for HIV.32 ,34 In this data set, the majority of partnerships cannot be linked, so we are unable to establish an individual's HIV risk due to their partner's concurrency; neither can we identify individuals in the population who have a higher risk of HIV and thus a potentially different mortality risk.32 ,35 The way in which a relationship ends may be associated with differential mortality risk, but it was not possible to explore formal divorce separately given that few customary marriages end with a formal divorce. That said, the study also has a number of strengths: the inclusion of relationship status as a time-varying rather than a fixed covariate, prospective longitudinal data collection that minimises recall bias, and the availability of detailed information allowing relationships to be categorised as conjugal and non-conjugal.

    There is much still to learn about the processes by which the timing and characteristics of a relationship impact on the health of men and women. Within our analysis, we controlled for educational attainment, which had a strong association with mortality risk, but we recognise that further work is needed to investigate socioeconomic gradients in relationship levels and patterns including age-specific formation, dissolution and duration. We anticipate that there may be socioeconomic differences in the pattern of conjugal relationships; however, these are likely to be complex given the declining marriage levels and rising age of first marriage. Appropriate measures of household-level and individual-level SES across time would be needed. Attributes of one’s partner may also determine the risk of marriage, risk of conjugal union and household SES, in addition to the risk of mortality. The effect of migration and cohabitation on mortality could also be explored and in doing so would allow for a better understanding of HIV and other health risks.

    In conclusion, we have demonstrated that being in a conjugal relationship in this context is associated with lower all-cause mortality. This study contributes to the increasing body of evidence worldwide that family relationships and partnerships are important determinants of health and health behaviours36 and suggests that greater attention on couple-focused interventions informed by knowledge about couples’ social and residential arrangements may be an effective way of improving individual and couples’ joint health outcomes.

    What is already known on this subject

    • Multiple studies have shown that being married is associated with lower mortality among both men and women. However, previous studies have concentrated on higher income contexts and have not been conducted in settings where marriage rates are low and many couples live apart.

    What this study adds

    • The concepts of conjugal and non-conjugal relationships are used in place of the married/non-married dichotomy normally used. This study shows that in South Africa being in a conjugal relationship is associated with lower mortality than for those who were not partnered, for both men and women. In some age groups, being in a non-conjugal relationship was also associated with lower mortality, meaning that not being partnered was associated with the highest risk of death.

    Acknowledgments

    The authors thank the community for their continued support and participation in ACDIS. They also thank the Africa Centre staff.

    References

    1. ↵
      1. Robards J,
      2. Evandrou M,
      3. Falkingham J, et al
      . Marital status, health and mortality. Maturitas 2012;73:295–9. doi:10.1016/j.maturitas.2012.08.007
      OpenUrlCrossRefPubMedWeb of Science
    2. ↵
      1. Hu Y,
      2. Goldman N
      . Mortality differentials by marital status: an international comparison. Demography 1990;27:233–50. doi:10.2307/2061451
      OpenUrlCrossRefPubMedWeb of Science
    3. ↵
      1. Holt-Lunstad J,
      2. Smith TB,
      3. Layton JB
      . Social relationships and mortality risk: a meta-analytic review. PLoS Med 2010;7:e1000316. doi:10.1371/journal.pmed.1000316
      OpenUrlCrossRefPubMed
    4. ↵
      1. Manzoli L,
      2. Villari P,
      3. M Pirone G, et al
      . Marital status and mortality in the elderly: a systematic review and meta-analysis. Soc Sci Med 2007;64:77–94. doi:10.1016/j.socscimed.2006.08.031
      OpenUrlCrossRefPubMedWeb of Science
    5. ↵
      1. Gove WR
      . Sex, marital status, and mortality. AJS 1973;79:45–67. doi:10.1086/225505
      OpenUrlCrossRefPubMed
    6. ↵
      1. Rendall MS,
      2. Weden MM,
      3. Favreault MM, et al
      . The protective effect of marriage for survival: a review and update. Demography 2011;48:481–506. doi:10.1007/s13524-011-0032-5
      OpenUrlCrossRefPubMedWeb of Science
    7. ↵
      1. Ploubidis GB,
      2. Silverwood RJ,
      3. DeStavola B, et al
      . Life course partnership status and biomarkers in mid-life: evidence from the 1958 British birth cohort. Am J Public Health 2015;105:1596–603. doi:10.2105/AJPH.2015.302644
      OpenUrlCrossRefPubMed
    8. ↵
      1. Johnson NJ,
      2. Backlund E,
      3. Sorlie PD, et al
      . Marital status and mortality: the national longitudinal mortality study. Ann Epidemiol 2000;10:224–38. doi:10.1016/S1047-2797(99)00052-6
      OpenUrlCrossRefPubMedWeb of Science
    9. ↵
      1. Kallan J
      . Effects of sociodemographic variables on adult mortality in the United States: comparisons by sex, age, and cause of death. Soc Biol 1997;44:136–47. doi:10.1080/19485565.1997.9988940
      OpenUrlPubMedWeb of Science
    10. ↵
      1. Berntsen KN,
      2. Kravdal O
      . The relationship between mortality and time since divorce, widowhood or remarriage in Norway. Soc Sci Med 2012;75:2267–74. doi:10.1016/j.socscimed.2012.08.028
      OpenUrlCrossRefPubMed
    11. ↵
      1. Bor J,
      2. Herbst AJ,
      3. Newell ML, et al
      . Increases in adult life expectancy in rural South Africa: valuing the scale-up of HIV treatment. Science 2013;339:961–5. doi:10.1126/science.1230413
      OpenUrlAbstract/FREE Full Text
    12. ↵
      1. Zaidi J,
      2. Grapsa E,
      3. Tanser F, et al
      . Dramatic increase in HIV prevalence after scale-up of antiretroviral treatment. AIDS 2013;27:2301–5. doi:10.1097/QAD.0b013e328362e832
      OpenUrlCrossRefPubMedWeb of Science
    13. ↵
      1. Tanser F,
      2. Bärnighausen T,
      3. Grapsa E, et al
      . High coverage of ART associated with decline in risk of HIV acquisition in rural KwaZulu-Natal, South Africa. Science 2013;339:966–71. doi:10.1126/science.1228160
      OpenUrlAbstract/FREE Full Text
    14. ↵
      1. Hosegood V,
      2. McGrath N,
      3. Moultrie T
      . Dispensing with marriage: marital trends in rural KwaZulu-Natal, South Africa, 2000–2006. Demogr Res 2009;20:279–312. doi:10.4054/DemRes.2009.20.13
      OpenUrlCrossRefPubMed
    15. ↵
      1. Posel D,
      2. Rudwick S,
      3. Casale D
      . Is marriage a dying institution in South Africa? Exploring changes in marriage in the context of ilobolo payments. Agenda 2011;25:102–11.
      OpenUrl
    16. ↵
      1. Hosegood V
      . Women, marriage and domestic arrangements in rural KwaZulu-Natal, South Africa. Acta Juridica: Marriage, land and custom, 2013:143–68.
    17. ↵
      1. Tanser F,
      2. Hosegood V,
      3. Bärnighausen T, et al
      . Cohort Profile: Africa Centre Demographic Information System (ACDIS) and population-based HIV survey. Int J Epidemiol 2008;37:956–62. doi:10.1093/ije/dym211
      OpenUrlFREE Full Text
    18. ↵
      1. Hosegood V,
      2. Timaeus I
      . Household composition and dynamics in KwaZulu-Natal, South Africa: mirroring social reality in longitudinal data collection. In: Van der Walle E, ed. African households: an exploration of census data. New York: M.E. Sharpe Inc, 2005:58–77.
    19. ↵
      1. Hosegood V,
      2. Benzler J,
      3. Solarsh GC
      . Population mobility and household dynamics in rural South Africa: implications for demographic and health research. S Afr J Demography 2006;10:43–68.
      OpenUrl
    20. ↵
      1. Cox DR
      . Regression models and life-tables (with discussion). J R Stat Soc, Ser B 1972;30:248–75.
      OpenUrl
    21. ↵
      1. Schoenfeld D
      . Partial residuals for the proportional hazards regression model. Biometrika 1982;69:239–41. doi:10.1093/biomet/69.1.239
      OpenUrlAbstract/FREE Full Text
    22. ↵
      1. Houlihan CF,
      2. Bland RM,
      3. Mutevedzi PC, et al
      . Cohort profile: hlabisa HIV treatment and care programme. Int J Epidemiol 2011;40:318–26. doi:10.1093/ije/dyp402
      OpenUrlFREE Full Text
    23. ↵
      1. Tian L,
      2. Zucker D,
      3. Wei L
      . On the Cox model with time-varying regression coefficients. J Am Stat Assoc 2005;100:172–83. doi:10.1198/016214504000000845
      OpenUrlCrossRefWeb of Science
    24. ↵
      1. Zuma K,
      2. Gouws E,
      3. Williams B, et al
      . Risk factors for HIV infection among women in Carletonville, South Africa: migration, demography and sexually transmitted diseases. Int J STD AIDS 2003;14:814–7. doi:10.1258/095646203322556147
      OpenUrlAbstract/FREE Full Text
    25. ↵
      1. Clark S
      . Early marriage and HIV risks in Sub-Saharan Africa. Stud Fam Plann 2004;35:149–60. doi:10.1111/j.1728-4465.2004.00019.x
      OpenUrlCrossRefPubMedWeb of Science
    26. ↵
      1. Bongaarts J
      . Late marriage and the HIV epidemic in sub-Saharan Africa. Popul Stud (Camb) 2007;61:73–83. doi:10.1080/00324720601048343
      OpenUrlCrossRefPubMed
    27. ↵
      1. Porter L,
      2. Hao L,
      3. Bishai D, et al
      . HIV status and union dissolution in sub-Saharan Africa: the case of Rakai, Uganda. Demography 2004;41:465–82. doi:10.1353/dem.2004.0025
      OpenUrlCrossRefPubMedWeb of Science
    28. ↵
      1. Shisana O,
      2. Zungu-Dirwayi N,
      3. Toefy Y, et al
      . Marital status and risk of HIV infection in South Africa: original article. S Afr Med J 2004;94:537–43.
      OpenUrlPubMed
    29. ↵
      1. Johnson L,
      2. Dorrington R,
      3. Bradshaw D, et al
      . Sexual behaviour patterns in South Africa and their association with the spread of HIV: insights from a mathematical model. Demogr Res 2009;21:289–340. doi:10.4054/DemRes.2009.21.11
      OpenUrlCrossRefWeb of Science
    30. ↵
      1. Garrib A,
      2. Herbst AJ,
      3. Hosegood V, et al
      . Injury mortality in rural South Africa 2000–2007: rates and associated factors. Trop Med Int Health 2011;16:439–46. doi:10.1111/j.1365-3156.2011.02730.x
      OpenUrlCrossRefPubMed
    31. ↵
      1. Mah TL,
      2. Halperin DT
      . Concurrent sexual partnerships and the HIV epidemics in Africa: evidence to move forward. AIDS Behav 2010;14:11–6. doi:10.1007/s10461-008-9433-x
      OpenUrlCrossRefPubMedWeb of Science
    32. ↵
      1. Tanser F,
      2. Bärnighausen T,
      3. Hund L, et al
      . Effect of concurrent sexual partnerships on rate of new HIV infections in high-prevalence rural South Africa: a cohort study. Lancet 2011;378:247–55. doi:10.1016/S0140-6736(11)60779-4
      OpenUrlCrossRefPubMedWeb of Science
    33. ↵
      1. Padian NS,
      2. Manian S
      . The concurrency debate: time to put it to rest. Lancet 2011;378:203–4. doi:10.1016/S0140-6736(11)60974-4
      OpenUrlCrossRefPubMed
    34. ↵
      1. Quigley M,
      2. Munguti K,
      3. Grosskurth H, et al
      . Sexual behaviour patterns and other risk factors for HIV infection in rural Tanzania: a case–control study. AIDS 1997;11:237–48. doi:10.1097/00002030-199702000-00015
      OpenUrlCrossRefPubMedWeb of Science
    35. ↵
      1. Eaton JW,
      2. Hallett TB,
      3. Garnett GP
      . Concurrent sexual partnerships and primary HIV infection: a critical interaction. AIDS Behav 2011;15:687–92. doi:10.1007/s10461-010-9787-8
      OpenUrlCrossRefPubMedWeb of Science
    36. ↵
      1. Lawlor DA,
      2. Lawlor DA,
      3. Mishra GD
      . Family matters: designing, analysing and understanding family based studies in life course epidemiology. Oxford University Press, 2009.

    Footnotes

    • Contributors The analysis was conducted by MC, with input from NM. The paper was jointly written by MC, NM and VH. All authors approved the final draft.

    • Funding NM and MC were supported by a Wellcome Trust fellowship awarded to NM (grant number WT083495 MA). VH involvement in the Africa Centre research activities is supported by the ESRC (ES/J021 202/1) and the Wellcome Trust (082384/Z/07/Z). The Africa Centre receives core funding from the Wellcome Trust, including for the surveillance (grant 082384/Z/07/Z).

    • Competing interests None declared.

    • Ethics approval Ethics permission for the demographic and HIV surveillance was obtained from the University of KwaZulu-Natal research ethics committee.

    • Provenance and peer review Not commissioned; externally peer reviewed.