Overall the pregnancy rate was at the low end of the range observed in other HIV prevention trials, on par with the one other vaccine trial reporting on pregnancy incidence in sub-Saharan Africa 
, and was half of that seen among African women enrolled in non-trial cohorts 
. Women in this trial were largely able to adhere to their initial agreement to avoid becoming pregnant during the vaccination period. Although the difference in pregnancy rates during and after the vaccination period was not statistically significant, there was a trend to higher rates after the vaccination period. We expect that the true difference is even greater since pregnancy testing in the post-vaccination period was less rigorous and likely under estimated the true pregnancy rate.
The proportion of elective abortions appeared to be elevated for pregnancies conceived during the vaccination period, compared with those conceived later, although we cannot conclude this with certainty because the study was not powered to evaluate this finding. We do not have information on women’s motivations, but a range of explanations for these elective terminations are possible. These may include: more immediate access to the health care system during trial participation; that women perceived it was unsafe to conceive a pregnancy during the vaccination period as they were advised to be on contraception during this time; or may simply reflect that pregnancies in the post-vaccination period were more often intended given women’s agreement to avoid them during the vaccination period. We lack detail on the circumstances of the elective abortions, but this procedure is legal in South Africa. Regarding pregnancy outcomes, those observed here were on par with another trial 
and from a representative community sample of South African women 
Encouraging was that predictive factors were identified, over-and-above the requirement that women use at least two forms of contraception as was required in this trial. All of the additional factors can be readily measured during screening for a large clinical trial, some are modifiable at trial entry, and preventive measures were associated with at least a halving in pregnancy risk. It is likely that the convenience, immediacy, certainty, and quality-of-care associated with on-site injectable contraceptive access is critical to its preventive role, as this finding is buttressed by another trial showing that women obtaining contraception off-site were at increased pregnancy risk 
. The pregnancy rates among injectable contraception users was higher than that typically seen for “perfect,” established users. It may reflect imperfect measurement as these findings were based on self-report for an unspecified timeframe of “current use” not clinical records, and thus may not have adequately captured the exact duration of use. Alternatively, women may not have been established users.
We also found that women who entered the trial using, and who continued to use, injectable contraceptives, but not oral contraceptive pills, were less likely to get pregnant. Despite being counter-intuitive, this finding has also been observed elsewhere: in a South African microbicide trial, pregnancy rates were 11.5/100 w-yrs among oral contraceptive users and <2.0/100 w-yrs among women on injectables 
. Even though both injectables and the pill have high method-effectiveness, women’s ability to consistently and/or correctly use the pill may diminish its method-effectiveness, and result in lower use-effectiveness, which takes into account user error. It is not biologically possible for oral contraceptives to increase pregnancy risk, and there are several possible explanations for observing a positive association between oral contraceptives and pregnancy. First, some women reporting oral contraceptive use, may not have been using a legitimate contraceptive, as there are pills in the marketplace that may not be contraceptive, but are advertised or understood as such. Second, as some form of hormonal contraception was required during the trial, those not using oral contraceptives were therefore using injectables, and the contrast between these two groups may simply reflect the greater use-effectiveness of injectables. Alternatively, when faced with the hormonal-method requirement, those opting for oral contraceptives may reflect a group of women: ambivalent about their pregnancy desires; not fully committed to avoiding pregnancy who thus opted for a self-dosing method, or; committed to avoiding pregnancy but who had difficulty adhering to the self-dosing regimen required by oral contraceptives.
Regarding condoms, how use is measured is important for distinguishing between casual and consistent users, as we found that only the latter group was at lowered pregnancy risk. Condom use measured via a single, “yes/no” question about “current” use, was not predictive, while consistent condom use, derived from a series of questions enquiring about condom use by each partner type during a specified timeframe, showed a trend toward halving pregnancy risk in the adjusted analysis. Careful measurements of condom use to determine consistent use has also been shown as protective by Reid et al., and use at last sex before study entry (but as not as one’s main method) by Halpern et al. in trial settings. Together these findings suggest that condoms may be a viable method for pregnancy prevention, so long as they are used consistently or recently. In contrast, if only a single, non-specific condom use measure is employed and women say “yes” to current condom use without further corroboration, (either in response to additional questioning or by failing to demonstrate familiarity with condoms when using a model) these women should be flagged as those in need of more intensive pregnancy counselling and support.
Pregnancy was also a function of HIV status, and number, of partners. Using women with a single partner of unknown HIV status as a reference, we were able to examine the unique impact of HIV status alone (by comparing pregnancy rates of women with only one partner with unknown status vs. negative HIV status) and of multiple partners (by comparing the impact of 1 versus 2 partners among women who didn’t know their partners’ HIV status). Regarding the former, knowing a sex partner’s HIV status may be emerging as a partnership factor of interest. Not knowing a partner’s HIV status may be a marker of a newer relationship where HIV status has not yet been discussed and so women were taking greater precautions against becoming pregnant. Alternatively, it may reflect a relationship where HIV status cannot be discussed, and thus women were hesitant to cement the relationship further with a pregnancy. We were unable to test these hypotheses as we lacked data on relationship duration and disclosure. This finding deserves further research. Regarding multiple partners, this may be a marker for women with more risky behaviour in general, as multiple partnering is also a risk factor for HIV. It is unclear whether the multiple partnering seen here is due to commercial transactions, as almost no women reported engaging in commercial sex when specifically asked (data not shown as it was extremely uncommon).
While almost twenty percent of the sample reported heavy drinking, it was heavy drinking along with either marijuana use, or concurrent drinking and sex that were associated with a two fold risk of pregnancy. Heavy drinking and recreational drug use are well-established risk factor for HIV in South Africa 
. The distinct nature of our finding requires corroboration, but preliminary implications for clinical trialists are that women reporting multiple risk factors, but not heavy drinking alone, should be flagged for increased pregnancy prevention counselling, and may be especially suitable candidates for trials given their increased risk for HIV, and their need for the risk reduction packages offered within trials. However, the larger public health problem of alcohol use and pregnancy remains.
Strengths and Limitations
Strengths of this analysis are that it is the first report from a vaccine trial on risk factors for pregnancy and the longitudinal design with high retention enhanced ability to make causal inferences. Factors examined were all collected within the context of a typical clinical trial, and didn’t require specialized interviewing techniques, yet were highly predictive of pregnancy risk. This bodes well for future trials as women can be readily screened for pregnancy using a few questions. Limitations of this analysis were the variable pregnancy prevention messages and pregnancy outcome ascertainment once the trial was interrupted. Nonetheless, we were still able to make valid comparisons within the dataset as the direction of the bias was known, but given the lack of systematic pregnancy testing in the post-vaccination period, overall pregnancy rates may be underestimated.
It is possible to efficiently screen women for pregnancy risk, and concrete steps such as providing injectable hormonal contraception for free on-site, and supporting consistent condom users, can reduce pregnancy risk among South African women in HIV trials. Additionally, among women with a single partner, differential knowledge of male partners’ HIV status impacts pregnancy rates and is a new finding that deserves further research to illuminate the underlying reason for its association with pregnancy. Given long-standing calls to better integrate family planning and HIV/STI risk reduction counselling, clinical trialists and health counsellors should make it a point to enquire about the number of, and male partner’s HIV status as a potential modifier of pregnancy and HIV risk. Together, these few simple steps may help to maximize the safety of the mother and children conceived during HIV prevention trials.