Long-term effects of unintended pregnancy on antiretroviral therapy outcomes among South African women living with HIV

Abstract

Objective:

Unintended pregnancies are common among women living with HIV, but there are no data on their long-term impact on treatment outcomes. In a cohort of women initiating antiretroviral therapy (ART) during pregnancy, we examined the association between the intendedness of the current pregnancy, measured antenatally, and elevated viral load (VL) up to five years postpartum.

Design:

Prospective study with enrolment at entry into antenatal care and follow-up at study visits separate from routine care.

Methods:

At enrolment women completed the London Measure of Unplanned Pregnancy. Mixed effects models examined the impact of the intendedness of the pregnancy (planned versus each of unplanned or ambivalent, respectively) on VL ≥50 copies/mL across postpartum study visits.

Results:

Overall, 459 women were followed for a median of 43 months postpartum, contributing 2535 VL measures (median per woman: 6). Ambivalent and unplanned pregnancy were commonly reported (20% and 60%, respectively), and the proportion of women with elevated VL increased over time (16% at 6 weeks to 43% by 36–60 months postpartum). Compared to those reporting a planned pregnancy, elevated VL was more common among women reporting an unplanned pregnancy [odds ratio (OR): 2.87; 95% confidence interval (CI): 1.46–5.64], with a trend towards a higher odds among those reporting ambivalence (OR: 2.19; 95% CI: 0.97–4.82); associations persisted after adjustment for a wide range of demographic, clinical and psychosocial factors.

Conclusions:

These novel data suggest that unplanned pregnancy may be a prevalent and persistent predictor of poor ART outcomes among women initiating ART during pregnancy.

Introduction

Unintended pregnancies are a major public health concern, with global estimates suggesting that 44% of pregnancies were unintended in 2010–2014 [1]. In low- and middle-income country (LMIC) settings, unintended pregnancies are similarly prevalent: population-level surveys have reported that 65% of pregnancies during 2012 in South Africa [2] and 46% during 2001–2013 in Uganda were unplanned [3]; 69% of women in Swaziland have reported that their most recent pregnancy was unintended [4]. Outside the context of HIV, there are data to suggest that unintended pregnancy may be associated with a range of maternal health concerns [5] such as depression [6–9] or risk behaviours such as delayed initiation of antenatal care [5,10,11].

Despite the frequency of unintended pregnancy there have been few considerations of long-term effects among women living with HIV. In South Africa, fewer than 30% of pregnant women living with HIV report that their pregnancy was planned [12,13], and experiencing an unintended pregnancy after being diagnosed HIV-positive is common in the United States [14,15]. An unintended pregnancy may further heighten the psychosocial and economic vulnerabilities of pregnant women living with HIV [16,17], and case-control studies have suggested that unintended pregnancies may be associated with mother-to-child HIV transmission [18,19]. However, there are few data on the long-term effects on maternal and child outcomes.

In particular there has been no consideration of the long-term impact of an unintended pregnancy on maternal HIV treatment outcomes, including adherence to antiretroviral therapy (ART). The World Health Organization’s policy of universal use of lifelong ART in all pregnant and breastfeeding women (“Option B+”) [20] has contributed to major increases in ART uptake but adherence to ART remains a major concern, particularly during the postpartum period [21–23]. Existing understandings of the drivers of suboptimal adherence in women living with HIV remain limited, including the potential role of unintended pregnancy. We have previously shown that unintended pregnancy is associated with elevated viral load (VL) at entry into antenatal care among women already on ART [24]. Here, we explored factors associated with the intendedness of the pregnancy among women who initiated ART during pregnancy; and examined the association between unintended pregnancy and elevated VL through 36–60 months postpartum.

Methods

Study design

We recruited consecutive pregnant women living with HIV from a public sector antenatal clinic in Gugulethu, Cape Town, for the MCH-ART study (ClinicalTrials.gov NCT01933477). The design and primary results of the study have been previously described [25,26]. In this setting, women receive integrated PMTCT services and antenatal care within the broader maternal and child health care platform during pregnancy. Women who were eligible to initiate ART were followed through delivery. Women who then opted to breastfeed were randomized immediately postpartum to different models for delivering HIV care and were followed at repeated study measurement visits up to 18 months postpartum. Women were randomised to either the local standard of care (referral out of the maternal and child health clinic to general adult ART services at their first postpartum visit), or to the MCH-ART intervention of integrated concurrent and co-located maternal ART and paediatric care in the maternal and child health clinic through the end of breastfeeding. The trial found that there were significant improvements in the combined endpoint of viral suppression and retention in HIV care through 12 months postpartum among women randomized to the intervention arm; retention in the MCH-ART trial did not differ across allocation [26]. We later extended follow-up to include one additional study visit between 36–60 months postpartum.

Participants

This secondary analysis includes all women who initiated ART during pregnancy and were followed as part of the randomized trial. ART eligibility was determined based on local guidelines: eligibility was based on CD4 cell count or clinical disease staging until June 2013, after which all pregnant women were ART-eligible under Option B+ guidelines. All women provided written informed consent prior to enrolment, and the study was approved by the University of Cape Town’s Faculty of Health Sciences Human Research Ethics Committee and by Columbia University Medical Center’s Institutional Review Board.

Measures

Enrolment into the study coincided with women’s entry into antenatal care. Women attended a maximum of two additional antenatal and one early postpartum study visit. Further study visits were then completed at 6 weeks and at 3, 6, 9, 12, 18 and 36–60 months postpartum, for a maximum of 7 postpartum study visits. All study visits were completed separately from any routine HIV or antenatal/postpartum care.

Study measures based on self-report were administered in isiXhosa, the predominant local language, by trained study interviewers. Measures were translated from English into isiXhosa and were back-translated using standard procedures to ensure accuracy [27]. A composite poverty score was calculated based on employment, housing type and access to household resources [28]. At all study visits, women underwent phlebotomy for batched HIV VL testing (Abbott RealTime HIV-1) conducted by the South African National Health Laboratory Services (NHLS). Infants underwent phlebotomy for HIV PCR testing at the 12 month study visit, conducted by the NHLS using the Roche Cobas AmpliPrep/Cobas TaqMan (CAP/CTM) HIV-1 assay (Roche diagnostics, New Jersey, USA).

The intendedness of the current pregnancy was assessed at enrolment, coinciding with entry into antenatal care, using the 6-item London Measure of Unplanned Pregnancy (LMUP) tool [29]. Each item was scored 0, 1 or 2, and all items were summed to create a score between 0 and 12, with higher scores indicating higher levels of intendedness; total scores were then categorised into unplanned (a score of 0–3), ambivalent (4–9) or planned pregnancy (10–12) [29]. The performance of the LMUP has been evaluated in both high-income and LMIC settings [30]. We have previously shown that the LMUP is a valid and reliable tool (Cronbach’s alpha, 0.84) in pregnant women living with and without HIV in South Africa [13].

The Edinburgh Postnatal Depression Scale (EPDS) [31] was administered at participants’ second antenatal study visit and at 6 weeks and 12, 18 and 36–60 months postpartum. A score of ≥13 was used to indicate elevated depressive symptoms [31]. The Alcohol Use Disorders Identification Test (AUDIT) [32] was administered at participants’ second antenatal study visit to assess alcohol use in the year prior to pregnancy recognition, and at 6, 12 and 36–60 months postpartum to assess postpartum alcohol use. In analysis, the AUDIT-C (AUDIT-Consumption) scoring system was used, where a score ≥3 on the first 3 items of the tool was used to indicate risky alcohol use [33,34]. The World Health Organization Violence Against Women tool was used to assess intimate partner violence [35]. Violence in the year prior to pregnancy recognition was assessed at participants’ second antenatal study visit, and postpartum violence was assessed at 12 and 36–60 months postpartum. In analysis, we categorised women as reporting any versus no violence.

Data analysis

Data were analysed using Stata 14 (StataCorp Inc, College Station, Texas, USA). We used mixed effects models to examine the impact of the intendedness of the pregnancy on elevated VL ≥50 copies/mL during the postpartum period. Throughout, we compared women reporting each of ambivalence or an unplanned pregnancy, respectively, to those reporting a planned pregnancy; and also examined the effect of increasing intendedness of the pregnancy using continuous LMUP scores. We conducted sensitivity analyses using (i) VL ≥1000 copies/mL and (ii) restricted to women who were virally suppressed <50 copies/mL at delivery. We explored the impact of missing data by assuming that women who did not attend a study visit had (i) suppressed and then (ii) elevated VL at that visit.

Given the potential for confounding of the putative association between the intendedness of the pregnancy and elevated VL we present results from unadjusted models and then: (A) adjusting for demographic and clinical characteristics only; (B) additionally adjusting for psychosocial factors measured at participants’ second antenatal study visit; and (C) additionally adjusting for postpartum psychosocial factors. To account for differences in viral suppression at 12 months postpartum observed in the primary trial analysis, all adjusted models were adjusted for women’s allocation in the MCH-ART trial. For postpartum psychosocial factors, we categorised women as scoring above the threshold value at any study visit at which these were assessed versus scoring below threshold at all visits. Possible modifiers of the association between the intendedness of the pregnancy and postpartum VL were examined in stratified analyses. Using standard formulae and assuming that unplanned pregnancy is an independent cause of elevated VL for these purposes, we calculated the population attributable fraction for the proportion of elevated VL that may be due to unplanned pregnancy. Finally, we used product-limit methods and the log-rank test to compare vertical transmission across LMUP categories.

Results

Intendedness of the pregnancy

A total of 471 women who had initiated ART during pregnancy and opted to breastfeed were followed postpartum. Seven women had not completed the LMUP and were excluded from analysis; a further five women attended no follow-up visits and were excluded. The remaining 459 women were enrolled while pregnant between March 2013-April 2014. Approximately two-thirds of women reported that they were not using contraception during the month prior to the pregnancy; had not discussed getting pregnant with their partner; and did not want a baby or intend to get pregnant (Table 1). However, 49% of women who reported not wanting a baby and not intending to get pregnant, respectively, reported that the timing of the pregnancy was okay, although not quite right. Using the LMUP tool, 20%, 20% and 60% of women reported that their current pregnancy was intended, ambivalent and unplanned, respectively.

Table 1.

Responses to London Measure of Unplanned Pregnancy tool

Variable	n (%)
Number of women	459
In the month that I became pregnant…	315 (69)
I/we were not using contraception
I/we were using contraception, but not on every occasion	88 (19)
I/we always used contraception, but knew that the method had failed at least once	49 (11)
I/we always used contraception	7 (2)
In terms of becoming a mother, I feel that my pregnancy happened at the…
Right time	146 (32)
Ok, but not quite right time	173 (38)
Wrong time	140 (31)
Just before I became pregnant…
I intended to get pregnant	129 (28)
My intentions kept changing	34 (7)
I did not intend to get pregnant	296 (64)
Just before I became pregnant…
I wanted to have a baby	131 (29)
I had mixed feelings about having a baby	36 (8)
I did not want to have a baby	292 (64)
Before I became pregnant…
My partner and I had agreed that we would like me to be pregnant	124 (27)
My partner and I had discussed having children together, but hadn’t agreed for me to get pregnant	29 (6)
We never discussed having children together	306 (67)
Actions to improve health in preparation for pregnancy1
None	414 (90)
One	8 (9)
Two or more	4 (0.9)
London Measure of Unplanned Pregnancy score – median [IQR]	3 [2, 9]
Unplanned	275 (60)
Ambivalent	91 (20)
Planned	93 (20)

¹Actions could include taking folic acid; stopping/cutting down on smoking or drinking alcohol; eating more healthily; seeking medical advice; or other actions.

Demographic, clinical and psychosocial characteristics

Overall the median age was 27.9 years, and women had entered antenatal care at a median gestation of 21 weeks (Table 2). Women reported low levels of educational attainment and employment; 83% reported one or more previous pregnancies; and 58% were diagnosed HIV-positive during their current pregnancy. Psychosocial risk factors were commonly reported; 61% had disclosed their HIV status to their male partner by delivery; and 75% were virally suppressed <50 copies/mL at delivery. Women reporting lower levels of intendedness related to their current pregnancy were significantly less likely to be married and/or cohabiting; had entered antenatal care at a later gestation with consequent ART initiation at a later gestation; and were somewhat less likely to have VL <50 copies/mL at delivery. In addition, women reporting lower levels of intendedness were more likely to report risky alcohol use prior to pregnancy and during the postpartum period and were less likely to disclose to their male partner by delivery. No differences in intendedness of the pregnancy were observed by age; timing of HIV diagnosis (before versus during the pregnancy); or allocation in the MCH-ART trial.

Table 2.

Demographic, clinical and psychosocial characteristics by the intendedness of the pregnancy

Variable1	Total – n (%)	Unplanned – n (%)	Ambivalent – n (%)	Planned – n (%)	P-value
Number of women	459	275	91	93	0.999
Median [IQR] age	27.9 [24.5, 32.6]	27.7 [24.3, 32.3]	27.8 [24.9, 31.9]	28.9 [24.6, 33.0]	0.600
Completed secondary/any tertiary education	113 (25)	66 (24)	19 (21)	28 (30)	0.324
Employed	175 (38)	99 (36)	41 (45)	35 (38)	0.303
Married and/or cohabiting	189 (41)	87 (32)	35 (38)	67 (72)	<0.001
Timing of HIV diagnosis
Before this pregnancy	194 (42)	117 (43)	39 (43)	38 (41)
During this pregnancy	265 (58)	158 (57)	52 (57)	55 (59)	0.953
Previous antiretroviral use	125 (27)	80 (29)	26 (29)	19 (20)	0.255
First pregnancy	78 (17)	48 (17)	11 (12)	19 (20)	0.305
Median [IQR] gestation in weeks at entry into antenatal care	21 [16, 26]	22 [17, 28]	20 [14, 24]	19 [15, 24]	0.003
Median [IQR] CD4 cell count at entry into antenatal care	353 [247, 527]	346 [245, 534]	400 [273, 565]	345 [223, 493]	0.135
Median [IQR] HIV viral load at entry into antenatal care (log10 copies/mL)	4.0 [3.4, 4.6]	4.0 [3.4, 4.5]	4.0 [3.3, 4.6]	4.1 [3.3, 4.7]	0.847
Gestation at ART initiation
<14 weeks	67 (15)	32 (12)	19 (21)	16 (17)
14–27 weeks	279 (61)	162 (59)	52 (57)	65 (70)
≥28 weeks	113 (25)	81 (29)	20 (22)	12 (13)	0.007
Antenatal depression: EPDS score ≥13	45 (10)	31 (11)	8 (9)	6 (6)	0.374
Risky alcohol use prior to pregnancy: AUDIT-C score ≥3	115 (25)	78 (28)	23 (26)	14 (15)	0.036
Any intimate partner violence prior to pregnancy	102 (22)	66 (24)	18 (20)	18 (19)	0.512
HIV viral load at delivery
<50 copies/mL	346 (75)	198 (72)	73 (80)	75 (81)
50–999 copies/mL	83 (18)	52 (19)	15 (16)	16 (17)
≥1,000 copies/mL	30 (7)	25 (9)	3 (3)	2 (2)	0.090
Disclosed to male partner by delivery	282 (61)	164 (60)	50 (55)	68 (73)	0.025
Postpartum depression: EPDS score ≥13 at any timepoint	59 (13)	33 (12)	17 (19)	9 (10)	0.151
Risky postpartum alcohol use: AUDIT-C score ≥3 at any timepoint (n=439)	99 (23)	66 (25)	22 (24)	11 (13)	0.047
Any postpartum intimate partner violence reported at any timepoint (n=424)	49 (12)	31 (12)	8 (9)	10 (12)	0.764

¹EPDS: Edinburgh Postnatal Depression Scale; AUDIT-C: Alcohol Use Disorders Identification Test – Consumption.

Impact of the intendedness of the pregnancy on postpartum VL

Women were followed for a median of 42.6 months postpartum, contributing a total of 2535 VL measures [median per woman: 6; inter-quartile range (IQR): 5–7]. The proportion of women with VL ≥50 copies/mL increased from 16% at 6 weeks postpartum to 43% by 36–60 months postpartum. Overall, 56% of women had VL ≥50 copies/mL at one or more postpartum study visit. At most visits, a dose-response association was observed: elevated VL was more common among women who had reported ambivalence compared to a planned pregnancy, with the prevalence of elevated VL even higher among those who had reported that their pregnancy was unplanned (Figure 1). VL ≥50 copies/mL at one or more postpartum study visit was observed among 60% of women who reported that their pregnancy was unplanned, 62% of women who reported ambivalence, and 41% of women who reported that their pregnancy was planned.

Figure 1.

Proportion with elevated viral load (VL) ≥50 copies/mL at postpartum timepoints¹ by intendedness of the pregnancy:

In unadjusted analyses, lower levels of intendedness were associated with elevated VL at postpartum study visits: compared to women who had reported a planned pregnancy, those who had reported an unintended pregnancy had almost 3 times the odds [odds ratio (OR): 2.87; 95% confidence interval (CI): 1.46–5.64] of elevated VL across study visits, with a trend towards a higher odds among those reporting ambivalence (OR: 2.19; 95% CI: 0.97–4.92; Table 3). Elevated VL was also more common among women who were younger; were neither married nor cohabiting; had entered antenatal care at a later gestation and with higher VL; and reported risky pre-conception alcohol use or intimate partner violence. In addition, the odds of elevated VL increased with increasing duration on ART, and elevated VL was more common among women reporting risky alcohol use or intimate partner violence during the postpartum period.

Table 3.

Associations between the intendedness of the pregnancy and elevated viral load (VL) ≥50 copies/mL postpartum from unadjusted models, and after adjustment for (A) demographic and clinical factors; and additional adjustment for (B) baseline and (C) postpartum psychosocial factors

Variable1	Unadjusted models		Adjusted model A		Adjusted model B		Adjusted model C
	OR [95% CI]2	P-value	aOR [95% CI]3	P-value	aOR [95% CI]	P-value	aOR [95% CI]	P-value
LMUP category (versus Planned)
Ambivalent	2.19 [0.97, 4.92]	0.058	1.72 [0.73, 4.05]	0.217	1.72 [0.73, 4.05]	0.214	1.56 [0.65, 3.76]	0.322
Unplanned	2.87 [1.46, 5.64]	0.002	1.97 [0.94, 4.14]	0.074	1.86 [0.89, 3.89]	0.099	1.73 [0.81, 3.69]	0.156
Age	0.91 [0.87, 0.95]	<0.001	0.91 [0.86, 0.95]	<0.001	0.90 [0.86, 0.95]	<0.001	0.90 [0.85, 0.94]	<0.001
Poverty score	1.02 [0.81, 1.27]	0.893	1.09 [0.86, 1.37]	0.484	1.08 [0.86, 1.37]	0.506	1.08 [0.84, 1.38]	0.534
Married and/or cohabiting (versus neither married nor cohabiting)	0.45 [0.27, 0.75]	0.002	0.69 [0.39, 1.24]	0.216	0.71 [0.39, 1.28]	0.259	0.78 [0.42, 1.44]	0.431
Diagnosed HIV-positive during this pregnancy (versus before)	0.82 [0.49, 1.36]	0.444
Previous antiretroviral use (versus no previous use)	1.68 [0.96, 2.94]	0.068	2.03 [1.12, 3.66]	0.019	2.02 [1.12, 3.64]	0.020	2.00 [1.10, 3.64]	0.023
Reports one or more previous pregnancies (versus first pregnancy)	0.70 [0.36, 1.37]	0.297
Log10 VL at entry into antenatal care	1.63 [1.24, 2.14]	<0.001	1.76 [1.32, 2.35]	<0.001	1.78 [1.33, 2.38]	<0.001	1.73 [1.28, 2.33]	<0.001
Gestation at ART initiation in weeks	1.06 [1.03, 1.10]	<0.001	1.08 [1.04, 1.12]	<0.001	1.07 [1.03, 1.11]	<0.001	1.07 [1.03, 1.11]	0.001
Months on ART at VL assessment	1.06 [1.05, 1.07]	<0.001	1.06 [1.05, 1.07]	<0.001	1.06 [1.05, 1.07]	<0.001	1.06 [1.05, 1.07]	<0.001
Disclosed to male partner by delivery (versus did not disclose)	0.72 [0.43, 1.21]	0.216
Antenatal depression	1.37 [0.58, 3.22]	0.472			0.64 [0.26, 1.57]	0.329	0.65 [0.26, 1.64]	0.363
Risky pre-conception alcohol use	1.80 [1.02, 3.20]	0.043			1.06 [0.57, 1.98]	0.846	0.95 [0.49, 1.87]	0.891
Any pre-conception intimate partner violence	2.93 [1.63, 5.28]	<0.001			2.81 [1.49, 5.31]	0.001	2.17 [1.12, 4.20]	0.022
Postpartum depression	1.98 [0.96, 4.11]	0.065					1.37 [0.61, 3.09]	0.446
Risky postpartum alcohol use	2.66 [1.48, 4.81]	0.001					1.73 [0.86, 3.47]	0.124
Any postpartum intimate partner violence	2.66 [1.23, 5.78]	0.013					2.23 [0.95, 5.19]	0.064

¹LMUP: London Measure of Unplanned Pregnancy; depression defined as Edinburgh Postnatal Depression Scale (EPDS) score ≥13; risky alcohol use defined as Alcohol Use Disorders Identification Test – Consumption (AUDIT-C) score ≥3

²OR: odds ratio; 95% CI: 95% confidence interval

³aOR: adjusted odds ratio; All adjusted models adjusted for all covariates shown and for design effect.

The effect of the intendedness of the pregnancy on elevated VL persisted after adjustment for demographic and clinical characteristics (adjusted model A); additional adjustment for pre-conception and antenatal psychosocial factors (B); and after additional adjustment for postpartum psychosocial factors (C), although the associations did not reach statistical significance in all instances. Results were similar when elevated VL was defined as ≥1000 copies/mL and when models were restricted to women who were virally suppressed <50 copies/mL at delivery (data not shown) but were slightly attenuated when assuming that missing VL measures were (i) suppressed or (ii) elevated (Supplemental Table 1).

Results were consistent when examining the effect of continuous LMUP scores: increasing intendedness of the pregnancy was strongly associated with a reduced odds of elevated VL in unadjusted models, with associations only slightly attenuated in adjusted models (Supplemental Table 2). In a graphical portrayal of LMUP scores and the predicted probability of experiencing elevated VL at each score (Figure 2), the magnitude of these results is clearly seen. First, the distribution of LMUP scores is bimodal, but the majority of women fall within the range of an unplanned pregnancy. Second, the highest probability of elevated VL falls in the large group of women who reported that their pregnancy was unintended, and the probability decreases in a dose-response relationship as the intendedness of the pregnancy increases.

Figure 2.

Distribution of London Measure of Unplanned Pregnancy (LMUP) scores¹, with predicted probability of elevated viral load (VL) ≥50 copies/mL at postpartum study visits:

Impact of the intendedness of the pregnancy across subgroups

The association between continuous LMUP scores and elevated VL ≥50 copies/mL was examined across subgroups (Supplemental Figure 1). Overall, increasing intendedness of the pregnancy reduced the relative odds of elevated VL in most subgroups explored. However, this effect appeared to be stronger among women who had completed less than secondary education and who reported more disadvantage. In addition, the effect was stronger among women who were married and/or cohabiting and who reported no previous pregnancies. Although HIV-status disclosure to a male partner was not associated with elevated VL in the total sample (Table 3), the effect of the intendedness of the pregnancy on elevated VL appeared to be stronger among women who had disclosed to their male partner by delivery.

The role of HIV-status disclosure to a male partner was further explored in stratified analyses (Supplemental Table 2). Among women who had not disclosed to their male partner by delivery, the intendedness of the pregnancy was not associated with elevated VL in either unadjusted or adjusted models. Among women who had disclosed by delivery, increasing intendedness of the pregnancy was strongly associated with a reduced odds of elevated VL (OR: 0.85; 95% CI: 0.77, 0.94); this effect persisted after adjustment for demographic and clinical factors but was slightly attenuated after additional adjustment for antenatal and postpartum psychosocial factors.

Impact of the intendedness of the pregnancy on vertical transmission

A total of 466 infants were born to the 459 women included in analysis, with 5 infants testing HIV-positive in the first 12 months postpartum. The rate of transmission at 12 months postpartum was 1.2% overall, with no difference across LMUP categories (p=0.999).

Discussion

This analysis presents novel data suggesting that unplanned pregnancy may be a common and persistent risk factor for poor ART outcomes among women initiating ART during pregnancy. We observed a dose-response association between lower levels of pregnancy intendedness and elevated VL, with elevated VL most common among women who had reported an unplanned pregnancy. This association persisted up to 36–60 months postpartum and was observed independent of demographic, clinical and psychosocial factors. To our knowledge, these data are the first to demonstrate a long-term effect of the intendedness of a pregnancy on VL and, given the high frequency of unintended pregnancy in women living with HIV in many LMIC settings, make an important new contribution to our understanding of the drivers of suboptimal adherence during the postpartum period.

Unintended pregnancies have been previously linked to other adverse health behaviours including delays in seeking antenatal care [5,11,36–38], substance use during pregnancy [36,37], and lower levels of breastfeeding [5]. Further, the psychosocial consequences may be profound: an unintended pregnancy can cause severe disruption to pre-existing life plans [36] and may lead to common mental disorders such as depression [5,7,36,38], as well as relationship dissolution or intimate partner violence [5,36]. Our finding that unintended pregnancy affects maternal ART outcomes extends these findings to the biomedical outcomes of HIV treatment, viral suppression. Preventing unintended pregnancies is a central aspect of the approach to reduce vertical HIV transmission [39]; further, our data suggest that an unintended pregnancy may compromise the benefits of ART for maternal health in the long term.

Although the impact of pregnancy intention persisted despite adjustment for multiple other risk factors, direct causality is difficult to determine [5]. Unintended pregnancies occur in complex social contexts and relationships [37], including being more common among unmarried women [36], and may heighten existing vulnerabilities [16,17]. It is likely that an unintended pregnancy is one several risk factors that contribute to women’s vulnerability, rather than a standalone factor that independently predicts elevated VL. Indeed, both the intendedness of a pregnancy and subsequent health outcomes may be determined by the same underlying factors [5]. In this study, several factors were associated with both lower levels of pregnancy intendedness and elevated VL: each was more common among women who were neither married nor cohabiting; had entered antenatal care at a later gestation; and who reported risky alcohol use. Here we posit that an unintended pregnancy heightens women’s vulnerability as part of a constellation of risk factors. In addition, we posit that unintended pregnancy may be a marker of some degree of ambivalence both towards pregnancy and towards one’s own health, with these ambivalences reflected in women’s adherence to ART. Additional socio-behavioural research is needed to explore these possibilities and the mechanisms of these novel findings.

If unplanned pregnancy can be viewed as an independent cause of elevated VL for the purpose of considering population-level impacts, then it is possible to calculate a population attributable fraction for the proportion of elevated VL in the cohort that may be due to unplanned pregnancy. Using standard formulae, we estimate that approximately 16% of elevated VL may be associated with unplanned pregnancy in this population, compared to either ambivalence or a planned pregnancy; if we include ambivalence to pregnancy in the calculations this estimate increases to 28%. Given that unintended pregnancies are preventable (for example through effective contraception), an alternate way to conceptualise this is to consider the possibility that only women with planned pregnancies entered the cohort. In this case, we estimate that 64% and 85% of all episodes of elevated VL would be prevented by the prevention of unplanned, or unplanned and ambivalent, pregnancies, respectively. Regardless of the formula used, these estimates speak to the population-level implications of the associations observed here.

A strength of this study is the inclusion of longitudinal data with long-term follow-up and the use of a robust biological endpoint. However, our results must be interpreted in light of several limitations. Unintended pregnancy is a complex construct to measure [37,38] but the LMUP has been validated widely, including in this population [13], and is robust enough to be recommended as an outcome measure for preconception care [40]. We assessed the intendedness of the pregnancy at entry into antenatal care, but the desirability of the pregnancy may change during the course of pregnancy [5]. Indeed, we did not assess how quickly women adapted in psychosocial terms to the pregnancy but note that 49% of women who reported not wanting a baby and not intending to get pregnant, respectively, reported that the timing of the pregnancy was okay. Preconception measurement of pregnancy intentions would be ideal but is challenging. As these data arise from a peri-urban setting in South Africa in the context of a research study, our findings should be generalized to other settings with caution, but we note that high levels of both unplanned pregnancy and elevated VL have been observed in many high-burden settings [3,4,21–23]. Finally, this research focussed on women initiating ART in pregnancy, and the question of how unplanned pregnancy effects outcomes among women who initiated ART before conception needs further investigation.

Despite these limitations, these results clearly warrant further attention. Given the high levels of unintended pregnancy and elevated postpartum VL observed here and in other settings, we argue that unintended pregnancy needs to be escalated as a global public health concern for maternal and child health in the context of HIV. Moreover, these data shed light on a unique factor in this patient population that may shape adherence behaviours and suggest that postpartum women who report antenatally that they did not intend their pregnancy may require specific attention from counselling and support interventions. Further, we argue that pregnancy planning needs to be incorporated into routine care for all women living with HIV: preventing unintended pregnancies may reduce the prevalence of elevated VL in this population and maximise maternal health.