Evaluating the competing risks of HIV acquisition and maternal mortality in Africa: a decision analysis

Authors


Dr MI Rodriguez, World Health Organization, 20 Avenue Appia, 1211 Geneva, Switzerland. Email rodriguezmar@who.int

Abstract

Please cite this paper as: Rodriguez M, Reeves M, Caughey A. Evaluating the competing risks of HIV acquisition and maternal mortality in Africa: a decision analysis. BJOG 2012;119:1067–1073.

Objective  To model the risk of HIV acquisition and maternal mortality for women in four African countries in the light of previous data on risk of HIV acquisition and hormonal contraceptive use.

Design  Decision analysis.

Setting  Chad, Kenya, South Africa and Uganda.

Population  Women of reproductive age, at risk of HIV, who do not desire pregnancy.

Methods  A decision analysis model was built to compare the consequences of removing progestin injectables from use, assuming an increased risk of HIV acquisition. Three scenarios were considered in four African countries: replacement of progestin injectables with no method, with combined oral contraceptives (COC) or with an intrauterine device (IUD). Health outcomes measured include: life-years, maternal mortality, HIV acquisition and unsafe abortion. Sensitivity analysis, including Monte Carlo simulation, was performed around all variables.

Main outcome measures  HIV acquisition, maternal mortality and life-years.

Results  If progestin injectables are removed from use, without a minimum of 70–100% of women switching to an IUD or COCs, up to nine additional maternal deaths will occur for every case of HIV averted. Sensitivity analysis demonstrated that this finding persisted across a broad range of variables.

Conclusions  Contraception is critical to preserving life for women in Africa. In the absence of clear evidence regarding hormonal contraception and HIV acquisition, policy decisions must not overlook the very real risk of maternal mortality.

Introduction

New data on the potential association between hormonal contraceptive use and acquisition of HIV has generated considerable attention and controversy.1 Recent evidence has suggested an increased risk of HIV seroconversion with use of the progestin injectable contraceptive.1,2 Data from a randomised trial of acyclovir to prevent HIV among HIV-discordant partners was used as a prospective cohort to examine rates of HIV acquisition among women with HIV-positive partners by hormonal contraception use. No significant association between HIV acquisition and oral contraceptive pills was noted. However, an adjusted hazard ratio of 2.19 (95% CI 1.01–4.74) was described for progestin injectables.1

This study’s findings are concerning because they represent the most extreme association seen to date between HIV acquisition and use of progestin injectable contraception. Previous data on the relationship between HIV acquisition and use of progestin injectables has been mixed, with some studies showing a protective effect and others an increased risk.3–12 Few studies showed a significant association, and interpretation of the data has been obscured by varying methodological quality. Variation in study design, in particular, control of potential confounders, makes a meta-analysis less reliable. No randomised trial or definitive data exist.

The issue is of considerable public health importance globally. The morbidity and mortality inherent to infection with HIV are well known.13 Global prevalence of HIV appears to have stabilised at 0.8% but this statistic masks dramatic variation by region.13 At the end of 2009, nearly 23 million people were living with HIV in Africa, compared with 1.5 million in North America.13 Reducing HIV prevalence among young women is critical to halting the epidemic; not only are women at significantly increased risk of HIV acquisition, but mother-to-child transmission remains a common cause of new infections.13

However, risk of HIV acquisition must be considered within the context of maternal mortality. Broadly, Africa has both the highest prevalence of HIV and the highest rates of maternal mortality.13,14 The lifetime risk of death during pregnancy, childbirth, or the puerperium in Africa is 1 in 32. This ranges from 1 in 14 in Chad to 1 in 100 women in South Africa (compared with 1 in 2900 for women in Europe).14 There is a high unmet need for family planning services in Africa.15 Only 22% of married women in Africa are using a modern method of contraception; unintended pregnancy rates are high.16 Between one-quarter and one-half of all unintended pregnancies end in abortion, and >98% of abortions in Africa are unsafe.17 For pregnancies culminating in a live birth, access to antiretroviral therapy (ART) in labour to prevent neonatal transmission varies widely. On average, only 38% of HIV-positive women in labour, in Africa, will receive ART.18

We designed a decision-analytic model to compare the competing risks of maternal mortality and HIV acquisition on life expectancy for women in four African countries. We assumed that progestin injectables are associated with an increased risk of HIV acquisition, and looked at the impact that removal of progestin injectable use would have on life expectancy. We considered maternal mortality rates, access to safe abortion, use of alternate contraceptive methods and access to HIV therapy.

Methods

A decision-analytic model was created using treeage software (TreeAge Software Inc, Williamstown, MA, USA). HIV incidence, access to ART, maternal mortality and contraceptive prevalence vary widely within Africa. We therefore examined the impact of progestin injectable use and HIV on life expectancy in four African countries: South Africa, Chad, Uganda and Kenya. Our population was women of reproductive age in these countries, who were HIV negative, and not currently planning a pregnancy.

The model begins with a woman electing a reversible method of contraception and compares a scenario where an injectable progestin contraceptive is available to one where it has been eliminated (Figure 1). A woman may choose to use nonhormonal contraception (copper intrauterine device [IUD]), hormonal contraception (progestin injectables, implant, combined oral contraceptives [COC]) or no method (including traditional methods). Although consistent condom usage is known to decrease HIV transmission, condom use is low.16 Our baseline assumption was that condom usage was the same between groups. All pathways are followed to the outcome of maternal mortality over 1 year. Our primary outcomes were life-years, HIV acquisition (maternal and neonatal) and maternal mortality. A discount rate of 3% was applied to life expectancy calculations.19 A literature search for model probabilities was performed. We used Heffron’s estimate to provide a worst case scenario for the risk of HIV acquisition with use of progestin injectables (adjusted hazard ratio 2.19, 95% CI 1.01–4.74).1 Data specific to sexually active, reproductive age women in each country were identified (Table 1). Contraceptive method use was assumed to be consistent with values reported by the United Nations.16 Contraceptive failure rates are based on typical user results over 1 year.20 Data from the World Health Organization (WHO) was obtained for the probability of maternal mortality and pregnancy outcome (including safe abortion) in African regions.14,15

Figure 1.

 The model began with a reproductive age woman seeking contraceptive care to prevent pregnancy. Probabilities of contraceptive use and non-use were considered under two strategies: current policy, and a situation where injectable contraception was unavailable. All pathways were followed to the same outcome, mortality from pregnancy or HIV. Branches truncated for clarity.

Table 1.   Model inputs
VariableChadKenyaSouth AfricaUganda
  1. *Ratio of the number of women of reproductive age married or partnered who are fecund, not using contraception and who report that they do not want any more children or wish to delay next pregnancy, divided by the number of women of reproductive age who are married or partnered.

  2. **2008 WHO estimates of maternal mortality ratio (per 100 000 live births).

  3. ***2008 WHO unsafe abortion rates (per 1000 women aged 15–44 years).

  4. ****Based on US data. Assumes typical use.

  5. *****African data. Mortality rates per 100 000 live births.

HIV
HIV incidence0.0070.0080.0100.009
Probability of receiving ARV in labour0.060.730.880.53
Probability of accessing ARV0.360.480.370.39
Contraception
Prevalence of contraceptive use (modern method)1.7%38.9%59.8%17.2%
Injectable use0.5%21.6%28.4%10.2%
COC use0.5%7.2%10.9%2.9%
Implant0.5%1.9%00.3%
IUD0%1.6%1%0.2%
Sterilisation0.2%4.8%14.3%2.4%
Condom use with high-risk intercourse7%12%24%
Unmet need for contraception*21%26%13.8%41%
Maternal Mortality-Live birth**1200530410430
Maternal Mortality-Unsafe abortion***3636936
HIV and Contraception
Risk of HIV with injectable12.19 (1.01–4.74)
Probability of maternal child transmission
 With ARV2%
 without ARV25%
Contraceptive failure rates****
Injectable6% (0–20%)
COC9% (0–40%)
IUD0.80% (0–20%)
Pregnancy outcomes*****
Live birth0.75
Induced abortion0.15
 Safe0.05
 Unsafe0.95
Spontaneous abortion0.09
Ectopic pregnancy0.01

The probability of receiving ART in labour for prevention of mother-to-child transmission in Africa ranges from 6% of births in Chad to 95% of births in Botswana. Data specific to each African country were obtained from the WHO. Therapy was assumed to decrease mother-to-child transmission of HIV from 25 to 2%.21 Access to ART for severe HIV infection (determined by 2010 WHO guidelines) varies within Africa. The probability that a woman will be able to access ART if HIV acquisition occurs was obtained from the WHO. Recent data have shown that with ART, even in low-resource settings, life expectancy is greatly improved.22,23 Based on these data, we assumed that for women diagnosed with HIV, access to ART would lead to a 25% reduction in life expectancy compared with a woman from the same country without HIV.23 For a woman diagnosed with HIV, unable to receive ART, the model assumes that her life expectancy would be reduced by 75%.

We calculated the number of cases of HIV averted as the total number of new cases of HIV (maternal and neonatal) that would be expected in each country both with injectable contraception use and without. When calculating expected cases without progestin injectable use, we considered three scenarios: the first was where all women currently using a progestin injectable switched to no method, the second and third were where all women discontinued progestin injectable use and elected to use another reversible method, either COCs or an IUD. Cases of HIV averted were compared with the difference in maternal deaths we would expect both with progestin injectable use and without.

The robustness of the model was evaluated with both univariate and multivariate sensitivity analyses. Sensitivity analysis allows an estimation of how changes in parameters such as HIV incidence, maternal mortality and contraceptive prevalence could affect results. Every variable was investigated for a threshold value. A threshold value is a point beyond which the results of the model change. Two-way sensitivity analysis was performed on all variables with threshold values and other key inputs.

The range of variation for each variable was 50–200% of the baseline estimate. We performed a Monte Carlo simulation using 10 000 trials to evaluate how simultaneous multivariable changes could affect outcomes. The Monte Carlo simulation uses sampling of the distributions of all of the model inputs in a series of trials.

Results

In all four countries, discontinuation of progestin injectables without replacement by another method, results in decreased life expectancy and increased maternal mortality (Table 2). Cases of HIV prevented ranged from three in Chad to 36 in Kenya (per 100 000 women). Averting these new cases of HIV resulted in increased maternal deaths in all countries; this varied from 23 additional maternal deaths in Chad to 330 in Kenya. This finding persisted across all values of maternal mortality, HIV incidence and contraceptive failure rates ranging from baseline to twice the initial estimate.

Table 2.   Comparison of current use of contraception with removing progestin injectable from market and all women go to using no method (per 100 000 women)
 No injectables
BaselineUse no methodDifference
Kenya
Change in life-years  −6100
 New HIV diagnosis11351099−36
 Maternal deaths263593+330
 Unsafe abortion482010 770+5950
Uganda
Change in life-years  −2600
 New HIV diagnosis868877+9
 Maternal deaths295453+158
 Unsafe abortion914911 050+1901
South Africa
Change in life-years  −3100
 New HIV diagnosis14261396−30
 Maternal deaths159405+246
 Unsafe abortion32808060+4780
Chad
Change in life-years  −500
 New HIV diagnosis940937−3
 Maternal deaths11001123+23
 Unsafe abortion11 48011 720+240

Next the impact of replacing progestin injectables with COCs was examined (Table 3). If all women were to replace use of progestin injectables with COCs, new HIV cases would be prevented at the expense of additional maternal deaths. Between three and 88 additional maternal deaths would be expected if all women using injectables transitioned to COCs. Between four and 136 new cases of HIV infection would be averted in this scenario.

Table 3.   Comparison of current use of contraception with removing progestin injectable from market and all women go to using COC (per 100 000 women)
 No injectables
BaselineAll use COCDifference
Kenya
Change in life-years  +100
 New HIV diagnosis11351009−126
 Maternal deaths263301+38
 Unsafe abortion48205704+884
Uganda
Change in life-years  +1200
 New HIV diagnosis868835−33
 Maternal deaths295383+88
 Unsafe abortion91499432+283
South Africa
Change in life-years  +800
 New HIV diagnosis14261290−136
 Maternal deaths159184−25
 Unsafe abortion32803990−710
Chad
Change in life-years  −100
 New HIV diagnosis940936−4
 Maternal deaths11001103+3
 Unsafe abortion11 48011 520+40

We then considered the effect of replacing progestin injectable use with a copper IUD (Table 4). Switching from progestin injectables to IUD would decrease new HIV diagnoses while maintaining or improving life expectancy. Cases of HIV prevented in this scenario ranged from five in Chad to 138 in South Africa. In all countries, life expectancy was maintained or increased. Sensitivity analysis demonstrated a threshold value for the probability that women would switch to an IUD (Figure 2). In all four countries, between 70 and 90% of all women would need to switch from progestin injectable use to an IUD for life-years not to be lost.

Table 4.   Comparison of current use of contraception with removing progestin injectable from market and all women go to using an IUD (per 100 000 women)
 No injectables
BaselineUse IUDDifference
Kenya
Change in life-years  +1500
 New HIV diagnosis11 1351008−127
 Maternal deaths263262−1
 Unsafe abortion48204796−24
Uganda   
Change in life-years  0
 New HIV diagnosis868856−12
 Maternal deaths2952950
 Unsafe abortion91499141−8
South Africa   
Change in life-years  +1500
 New HIV diagnosis14261288−138
 Maternal deaths159158−1
 Unsafe abortion32803260−20
Chad   
Change in life-years  0
 New HIV diagnosis940935−5
 Maternal deaths110011000
 Unsafe abortion11 48011 484+4
Figure 2.

 What proportion of women would have to switch to using an IUD for removal of injectable contraception from the market to not decrease life-years (per 100 000 women)?

Contraceptive efficacy rates are thought to vary internationally because of imperfect use.24 We performed two-way sensitivity analyses on probability of method failure with each contraceptive and HIV incidence, as well as probability of switching methods. In all countries, pregnancy rates with progestin injectables would need to exceed 20% and more than half of women would have to switch to another method for progestin injectables not to increase life-years. In Kenya, Uganda and Chad, progestin injectables increase life-years even if we assume a 3% incidence of HIV annually, and a 40% pregnancy rate with progestin injectables. In South Africa, a country with safe abortion and high contraceptive and ART use, a lower threshold is identified. If the pregnancy rate with progestin injectables exceeds 20%, at the current HIV incidence of 1.5% annually, progestin injectable use will decrease life expectancy. Below an HIV incidence of 1.5% annually, progestin injectable use increases life-years.

Monte Carlo simulations, which sample the distribution around each input of the model allowing for simultaneous consideration of uncertainty, were performed for each country. For risk of HIV acquisition with progestin injectable use, we ranged the value for the adjusted hazard ratio across its given confidence interval (1.01–4.74).1 In Kenya, Uganda and Chad, maintaining use of progestin injectables was the optimal strategy in over 90% of simulations (99.7%, 97% and 94%, respectively). In South Africa, maintaining the availability of progestin injectables was the optimal strategy in 65% of trials.

Discussion

Contraception is critical to preserving life for women in Africa. Even assuming the ‘worst case’ scenario for a significant association between progestin injectable use and HIV acquisition, all contraceptive methods result in increased life expectancy. This finding persisted across varying ranges of HIV incidence, contraceptive prevalence, access to ART and maternal mortality. If progestin injectables are removed from the market without an equally effective and acceptable replacement being provided, there will be an additional nine maternal deaths for every case of HIV averted. This highlights the importance of maintaining and increasing access to contraception for all women.

Our model made the baseline assumption that progestin injectable use is associated with an increased risk of HIV, as found in a recent epidemiological study. It is important to note that this finding is the most extreme association of HIV acquisition and progestin injectable use made to date. Previous studies have not consistently identified a significant association, and no definitive trial has been conducted. Our study therefore presents a ‘worst case scenario’. Even with assumption of a two-fold increased risk of HIV acquisition with use of progestin injectables, the use of all contraception, including injectables, results in increased life expectancy.

From a policy and programmatic perspective, it is critical to consider not only the biological and scientific validity of studies suggesting an association between progestin injectables and HIV acquisition, but the consequences for women of not using the method. Our study therefore modelled multiple outcomes for three different scenarios. We examined outcomes for removal of progestin injectables and replacement with no method, COCs or an IUD. While ideally progestin injectables would be replaced by an equally effective method, such as the IUD, this cannot be assumed. Use of IUDs is unusually low throughout Africa, probably reflecting a lack of trained providers.16 Additionally, it is possible that mistrust in the medical establishment following removal of progestin injectables would lead more women to use no method.

As a model, our study has inherent limitations. Population estimates of key variables such as contraceptive use and HIV incidence are notoriously challenging to estimate. Our model relied on inputs from the WHO and the joint United Nations Programme on HIV/AIDs. Sensitivity analysis was used to range these estimates widely, as underestimating the incidence of HIV could misrepresent the impact of HIV on life expectancy. Across a broad range of HIV incidence, our results remained the same. Additionally, our model assumes consistent condom usage and frequency of coitus between all groups. If all groups were to use condoms consistently and perfectly, both cases of HIV and of pregnancy would be overestimated by our model. A difference in coital frequency between groups would increase both the risk of pregnancy and HIV acquisition for that group. Limited data exist on contraceptive failure rates in Africa. However, our study shows that failure rates for progestin injectables can be as high as 20% in areas of high HIV incidence and still protect life expectancy.

Conclusion

Definitive data about the potential association between hormonal contraception and HIV acquisition are urgently needed. In the absence of clear evidence, policy and programmatic decisions pertaining to the use of progestin injectables among women at risk for HIV acquisition must be made cautiously. Given that the models show that no contraception results in the worst outcomes overall, efforts to define the association between injectable progestin and HIV acquisition should consider comparing the potential risk between effective methods of contraception (COCs, IUDs and implants). Barriers to the use of long-acting, effective contraception should be identified and reduced. The very real risk of maternal mortality for women in Africa must not be overlooked. Efforts to improve life expectancy should focus on increasing access to modern contraceptives, safe abortion, access to ART and both horizontal and vertical HIV prevention.

Disclosure of interests

None.

Contribution to authorship

MIR and ABC designed the model, ran the experiments, performed the analysis and wrote the paper. MR helped with model design, ran analyses and wrote the paper.

Details of ethics approval

As a decision analysis using publicly available information, this project was exempt from review.

Funding

This study was funded by the World Health Organization, Department of Reproductive Health and Research.

Acknowledgements

None.

Ancillary