Modeling the potential impact of viral load monitoring on vertical transmission of HIV in Kenya

Abstract

Introduction:

Viral load (VL) monitoring among pregnant and breastfeeding women (PBFW) can reduce vertical transmission (VT) by identifying PBFW with unsuppressed VL and promoting re-suppression. However, the impact of varying degrees of adherence to VL monitoring guidelines on the prevention of vertical transmission (PVT) is unknown.

Methods:

We developed a microsimulation model of HIV progression and PVT care for PBFW living with recently acquired HIV in Kenya. We used this model to evaluate VL monitoring in two key maternal populations: 1) newly positive (NP) pregnant women who initiate ART during antenatal care and 2) known positive (KP) women who are diagnosed and initiate ART prior to conception. For each population, we simulated three levels of adherence to Kenyan VL monitoring guidelines during pregnancy and 18 months of breastfeeding: 1) No testing (NT); 2) 50% adherence to indicated VL tests (VL-50%); 3) 100% adherence (VL-100%). We evaluated VL monitoring in each population by comparing live births, maternal deaths, and VT under VL-50% and VL-100% to NT.

Results:

Under NT, infants acquired HIV at a rate of 619 vs. 505 per 10,000 live births in the NP vs. KP populations, respectively. VL monitoring reduced VT by 1.6–2.7% in NP women vs. 9.1%−14.3% in KP women, and it reduced maternal deaths by 1.2–1.8% vs. 1.6–2.3%.

Conclusion:

Maternal VL monitoring in Kenya has considerably greater potential for achieving relative reductions in VT among KP women than among NP women. In KP women, even imperfect adherence to guidelines may achieve substantial relative reductions in VT.

INTRODUCTION

Over the last two decades, significant progress has been made in the prevention of vertical (mother-to-child) transmission of HIV (PVT), with an estimated global decrease in new child infections from 530,000 in 2000 down to 130,000 in 2022.[1] This dramatic decline in vertical transmission (VT) is primarily due to significant expansion of antiretroviral therapy (ART) coverage, with 82% of pregnant and breastfeeding women (PBFW) living with HIV receiving ART in 2022 [1]. However, the rate of PVT progress has slowed in recent years, highlighting the need for continued improvements in HIV testing, ART coverage, and retention in care: 70% of children with new infections in 2022 had mothers who did not receive ART, and an additional 22% had mothers who discontinued ART during pregnancy or breastfeeding [1].

PBFW who are on ART also need more support to maintain viral suppression, as 8% of children newly infected with HIV had mothers who were on ART but not virally suppressed [1]. This group may benefit from viral load (VL) monitoring [2], which can identify individuals with unsuppressed VL and trigger adherence counseling with the goal of achieving viral suppression or re-suppression [3]. In 2013, the WHO recommended VL testing as the preferred approach to monitor for HIV treatment failure, and in 2021 recognized the importance of maternal viral suppression with a call for more frequent VL monitoring during pregnancy and breastfeeding [4]. Since then, tremendous progress has been achieved in scaling up the availability and utilization of this intervention. However, numerous studies throughout sub-Saharan Africa have documented gaps in adhering to guidelines across the entire VL testing cascade, including high rates of delayed or missed routine and repeat VL tests [5]. VL studies among PBFW show similar trends [6–9], suggesting potential missed opportunities for increasing maternal viral suppression and reducing VT.

While modeling studies have been used to evaluate the benefits of VL monitoring among adults [10], most were conducted prior to the roll-out of dolutegravir (DTG)-based ART in Africa, and none have examined the potential reductions in VT that can be achieved by improving adherence to VL monitoring guidelines among PBFW. Prior modeling of various VL testing frequencies among PBFW examined their impact on maternal viral suppression but stopped short of simulating VT [11, 12]. A more recent study examined the impact of using a point-of-care test for VL monitoring within a package of several PVT interventions, masking the standalone impact of VL monitoring [13]. To address these gaps, we used simulation modeling to evaluate how different levels of adherence to VL monitoring guidelines among PBFW may impact PVT in two key maternal populations: pregnant women who are diagnosed and initiate DTG-based ART upon presenting to antenatal care (Newly Positive Cohort), and pregnant women living with HIV who initiate DTG-based ART prior to conception (Known Positive Cohort). We focus our analysis on Kenya, using this high HIV burden country in sub-Saharan Africa as a case study with regionally relevant results.

METHODS

Overview

We developed a microsimulation model of HIV progression and PVT care for pregnant women living with HIV in Kenya by adapting a previously described model of adult HIV progression and treatment [14]. The model is specifically parameterized for young PBFW with recently acquired HIV, a population for whom viral suppression patterns have been well-characterized [15]. Table 1 shows key model parameters (base-case values and sensitivity analysis ranges). The model updates age and key disease characteristics such as CD4 cell count, viral load, and ART regimen status in monthly time cycles. Pregnancy, modeled as 9 monthly time cycles (39 weeks gestation), is confirmed at the first antenatal clinic (ANC) visit at 5 months (22 weeks) gestation [16], and concludes in a full-term, singleton live birth. The newborn is explicitly represented in the model as a new individual who may acquire HIV during the peripartum or postpartum periods, either through a single risk at delivery or through an ongoing monthly risk during breastfeeding. Effective ART decreases VT risk. Among those on ART, VT risk is lower with earlier initiation of ART and at lower viral loads. Women may die from HIV, maternal or other-cause mortality. Infants have ongoing all-cause mortality risk until HIV acquisition, after which we cease tracking infant outcomes.

Table 1. Base case and sensitivity values for key model parameters.

See Appendix for additional details. ART=antiretroviral therapy. VL=viral load.

Parameter	Base-Case Value (Sensitivity Ranges)		References
Viral Suppression on Dolutegravir-Based ART
ART effectiveness: virologic suppression (%)a			Kinuthia et al.,[15] Dorward et al.[25]
Month 16 of ART	91.6 (84.0–94.7)
Month 33 of ART	87.6 (80.4–92.0)
Probability of re-suppression after a VL test (%)b	50 (30–70)		Sokela et al.[26]
Antenatal and HIV Care
Months gestationc at first antenatal care visit	5 (3–8)		Aksunger et al.[16]
Months duration of breastfeeding	18 (12–24)		Glaubius et al.[18]
Interrupted ART (%) from first antenatal care visit to end of breastfeeding (22 months)a, d	Known Positive	Newly Positive	Humphrey et al.,[21] Chammartin et al.[34]
	8.5 (3.4–13.6)	23.5 (9.4–37.6)
Peripartum Vertical Transmission e
Probability in month of delivery (%), mother on ART	Known Positive	Newly Positive	Mandelbrot et al.,[22] Townsend et al.[35]
VL < 50	0.0 (0.0–0.0)	0.5 (0.3–0.8)
50 < VL < 400	0.3 (0.2–0.5)	1.4 (0.7–2.1)
400 < VL < 10,000	2.2 (1.1–3.3)	2.4 (1.2–3.6)
VL >10,000	9.2 (4.6–13.8)	9.2 (4.6–13.8)
Probability in month of delivery (%), mother not on ART and HIV acquired pre-conception e			Rollins et al.[19]
CD4 < 200	37
200 < CD4 < 350	27
CD4 > 350	15
Postpartum (Breastfeeding) Vertical Transmission			Dugdale et al.[36]
Monthly probability (%), mother on ART
VL < 50	0.06 (0.05–0.07)
50 < VL < 1,000	0.39 (0.34–0.44)
VL > 1,000	0.78 (0.67–0.89)
Monthly probability (%), mother not on ART f	0.89

^aReported as a cumulative risk for ease of interpretation; modeled as a monthly risk

^bAmong those not virally suppressed, probability of re-suppression after VL testing and adherence counseling.

^cGestational age is measured in months due to the monthly time cycle of the model. Corresponding ages in weeks are: month 5 = weeks 17.4–21.7; month 3 = weeks 8.8–13; month 8 = weeks 30.4–34.7.

^dFor known-positive women, the corresponding monthly risk of ART interruption also applies to the time period between ART initiation and first ANC visit.

^ePeripartum vertical transmission parameters reflect combined in utero and intrapartum transmission.

^fSensitivity analyses were conducted by varying on-ART transmission risk relative to these values, which were kept constant.

We used this model to construct two cohorts of women living with recently acquired HIV who are aged 25 at the start of the simulation: a Newly Positive (NP) Cohort who initiate ART at the first ANC visit, and a Known Positive (KP) Cohort that has been on ART for six months at their time of conception (Figure 1). Kenyan VL monitoring guidelines recommend an initial VL test three months after ART initiation for both cohorts [17]. Subsequent routine tests occur at confirmation of pregnancy for women already on ART and every six months for all PBFW. Those who are unsuppressed at a routine VL test receive a repeat test three months later. For each cohort, we simulated three levels of adherence to these VL monitoring guidelines: (1) No Testing (NT); (2) 50% adherence to indicated routine VL tests (VL-50%); and (3) 100% adherence to indicated routine VL tests (VL-100%). We assume that imperfect adherence to VL monitoring guidelines is not due to patient behavior but rather to gaps in the healthcare system, such as limited laboratory capacity, insufficient staff, or reagent stock outs. Under VL-50%, all individuals thus have a 50% probability of receiving an indicated routine test. If a routine test is performed, we assume that adherence counseling and repeat VL testing also occur as indicated. For both VL-50% and VL-100%, we compared outcomes to those under NT to assess the impact of VL monitoring at each adherence level in each population. All simulations were programmed in Matlab R2021a and results were analyzed using R 4.2.3.

Figure 1. Monthly timeline of events in the two maternal cohorts.

All women acquire HIV at time 1, start ART at time 7, and have a 1^st VL test at month 3 of ART, at time step 10 (blue panel). Routine testing schedules for those suppressed at the prior test are either every 6 months for PBFW (green), every 12 months for non-PBFW (yellow), or at month 12 of ART for PBFW prior to their 1^st ANC visit (red). This translates to routine tests at time steps 16, 22, 28 (green) and 40 (yellow) for the Newly Positive Cohort and routine tests at time steps 18, 24, 30 and 36 (green) for the Known Positive Cohort. The Known Positive Cohort’s month 12 of ART occurs at time step 19, after their 1^st ANC visit (time step 18), at which point they have switched to PBFW guidelines. Only select time points are shown between months 23–40.

Key Features of the Two Maternal Populations

While there are numerous pregnancy and ART initiation timelines among young women living with HIV in the real world, we designed our two cohorts as demonstrative case studies that capture and contrast the characteristics most salient to maternal VL monitoring. To equalize the two cohorts with respect to duration of untreated HIV progression, HIV acquisition occurs in the first simulated month (time step 1) and ART initiation occurs six months later at time step 7 for all women (Figure 1). For the NP Cohort, conception occurs in time step 2, and ART initiation occurs at their first ANC visit, when they discover their HIV status. In the KP Cohort, conception occurs at time step 13, six months after ART initiation, and their pregnancy is confirmed at their first ANC visit in time step 18. All women breastfeed for 18 months [18], completed at time step 29 for the NP Cohort and at time step 40 for the KP Cohort. Because both cohorts acquire HIV prior to conception (“non-incident infection”), all women have the same probability of peripartum transmission if they are off ART [19].

In addition to having distinct pregnancy timelines, the two cohorts also differ in their ART engagement and VT risk. Following ART initiation, women have a monthly risk of ART interruption, and based on data from Kenyan PBFW living with HIV, this risk is higher for the NP Cohort [20, 21]. In the KP Cohort, we model the first ANC visit as an opportunity for 100% of those with interrupted ART to re-engage, after which they resume being at risk for ART interruption. Given data showing that earlier ART initiation decreases VT, we also model the KP Cohort as having lower peripartum transmission risk on ART for a given VL at delivery [22]. All other assumptions are the same for the two cohorts.

ART Effectiveness and Viral Load Monitoring

We assume ART effectiveness depends on medication adherence. Based on data from pregnant women on DTG-based ART, VL declines upon ART initiation [23], and it ultimately reaches viral suppression below 50 copies/mL or low-level viremia [24]. Effective ART also promotes CD4 cell count recovery, which in turn reduces the risk of acute clinical events and HIV-related death. Due to the scarcity of data on DTG-based ART outside of clinical trials, our ART effectiveness estimates are informed by a combination of data from Kenya and South Africa (Appendix 1.3.2; Table A5) [15, 25]. We use these data to inform an ongoing monthly risk of ART becoming ineffective due to poor adherence, at which point VL rebounds to pre-treatment levels, and CD4 cell count decline resumes.

If a routine VL test detects unsuppressed VL (VL ≥ 200 copies/mL, per Kenyan guidelines) [17], this prompts adherence counseling, which has a 50% probability of reestablishing VL declines consistent with effective ART (“re-suppression”, Table 1) [26]. We assume the VL test is laboratory-based, the most widely implemented version of VL testing, rather than point-of-care [27]. For those who re-suppress, their VL decline begins two monthly time cycles from the time the VL test is performed, to account for test result turnaround time and adherence improvement (Appendix 1.6.2) [28]. For women whose VL remains unsuppressed at the repeat test three months later (virologic failure), we assume they remain on DTG-based ART, consistent with current implementation of Kenyan guidelines (see Appendix 1.6.1) [17]. These women have future opportunities to re-suppress with continued routine VL testing and adherence counseling.

Health Outcomes

For each guideline adherence level, we projected number of live births, maternal viral suppression below 1,000 copies/mL, maternal deaths, and peripartum vs. postpartum infant infections. To standardize the cohorts and guideline adherence levels for differences in live births, we quantified the PVT impact of VL monitoring as the relative reduction (percent reduction) in infant infections per 10,000 live births compared to NT.

Sensitivity & Scenario Analyses

We investigated model sensitivity to our base-case parameter values (Table 1, see Appendix for calculations). We also modeled three variations on the VL testing cascade: a point-of-care VL test scenario with a one-month (rather than two-month) time to re-suppression, a delayed VL test scenario with a three-month time to re-suppression, and a scenario in which only 50% (instead of 100%) of indicated adherence counseling occurs. To explore the impact of there being differential ART interruption depending on NP vs KP status, we also modeled a scenario equalizing ART interruption risk to the midpoint of the two cohorts’ risks. Finally, we examined two variations on VT risk: 1) equal peripartum VT risk for mothers on ART across cohorts and 2) an “incident infection” scenario in which the NP Cohort’s risk of peripartum VT off ART was increased to the elevated VT risk observed among mothers who acquire HIV during pregnancy and are off ART at the time of delivery [19].

Ethics

Data used to inform model parameters were publicly available aggregate data, for which consent is not required.

RESULTS

Maternal Viral Suppression

In the absence of VL monitoring (NT), the proportion of women with viral suppression peaked three months after ART initiation (time step 10) and subsequently declined due to the accumulation of ineffective or interrupted ART (Figure 2). This decline was steeper in the NP Cohort due to their greater ART interruption risk. In the KP Cohort, the re-engagement in ART at the first ANC visit for those with interrupted ART (time step 18) led to a small increase in the proportion with viral suppression, after which viral suppression continued to decline. Although ART effectiveness was the same in both cohorts, their different timing of events and different ART interruption risks resulted in the KP Cohort having higher viral suppression under NT at conception, nine months postpartum, and 18 months postpartum, whereas the NP Cohort had higher viral suppression at delivery (Figure 2, Table S1).

Figure 2. Percent virally suppressed below 1000 copies/mL in the two maternal cohorts over simulation time.

Colors denote each cohort, and line type indicates the guideline adherence level. Symbols show the timing of key events. In the KP Cohort, time step 40 reflects viral suppression among two subgroups: 1) women who never had ART interruption or had it occur after the 1^st ANC visit and 2) women with ART interruption prior to time step 18 who re-engaged at the 1^st ANC visit. Y-axis between 15% and 50% not shown.

Greater guideline adherence increased the percentage of women with viral suppression at each maternal timepoint (Figure 2, Table S1). The incremental benefits of both adherence levels over NT increased over time (Figure 2) and were larger for KP than NP women, as a greater proportion of KP women had unsuppressed VL while also being on ART (Figure 3). Because maternal events for KP women occurred in later months, when suppression was lower under NT, the impact of VL monitoring was also greater in this cohort at key maternal time points (Table S1). Moreover, within the NP Cohort, there was no improvement in viral suppression at delivery because the first VL test happened only one month prior. Thus, across the two cohorts, the absolute increase in the proportion of women with viral suppression due to VL monitoring ranged from 0% at delivery in the NP Cohort for both adherence levels to 8.1% at 18 months postpartum in the KP Cohort for VL-100% (Table S1).

Figure 3. Percent virally unsuppressed above 1000 copies/mL in the two maternal cohorts under no testing, by ART status.

Panel A: at select simulation months. Panel B: at key maternal timepoints. NP = Newly Positive Cohort; KP = Known Positive Cohort. The y-axis indicates the percent of the cohort that is unsuppressed. The labels on the bars sum to 100% and indicate the breakdown of on ART vs. off ART among the unsuppressed (i.e., within each bar). The height of the blue bars thus show the percent of each cohort that is unsuppressed while on ART, under no testing.

Maternal Deaths and Live Births

Under NT, the NP Cohort had more total deaths than the KP Cohort but fewer occurring prior to delivery, resulting in more live births but fewer mothers surviving through the completion of breastfeeding. The relative reduction in maternal deaths due to VL monitoring was greater in the KP Cohort than in the NP Cohort (VL-100% = 2.3% vs. 1.8 and translated into more live births in the KP Cohort (Table S2).

Vertical Transmission

Figure 4A shows how viral suppression in the two cohorts translated into absolute numbers of peripartum and postpartum infections per 10,000 live births. Under NT, the NP Cohort had higher peripartum VT than the KP Cohort. Because being off ART substantially increases peripartum transmission risk (Table 1), the NP Cohort’s lower ART coverage among women with unsuppressed viral load at delivery (Figure 3B) offset their higher overall viral suppression at delivery (Table S1). Postpartum VT was also higher in NP women due to their lower ART coverage during breastfeeding (Figure 3B).

Figure 4. Peripartum, postpartum and total vertical transmission in the two maternal cohorts, by guideline adherence level.

Panel A: number per 10,000 live births. Panel B: percent reduction in vertical transmission due to each guideline adherence level, compared to no testing (NT). In VL-50%, each individual has a 50% probability of completing their recommended VL tests, compared to 100% completion in VL-100%.

Consistent with its differential impact on viral suppression across the two cohorts, VL monitoring led to a larger relative reduction in total VT in the KP Cohort (VL-50% = 9.1%; VL-100% = 14.3%) compared to the NP Cohort (VL-50% = 1.6%; VL-100% = 2.7%) (Figure 4B). For the NP Cohort, VL monitoring did not decrease peripartum transmission at all. While VL-50% achieved 59% of the VT reduction attained by VL-100% in the NP Cohort, it achieved 64% of the VT reduction attained by VL-100% in the KP Cohort (Table S3).

Sensitivity and Scenario Analyses

The one-way sensitivity analyses showed that the relative reduction in total VT for VL-100% in both cohorts increased with lower ART effectiveness, lower ART interruption, higher levels of re-suppression, and longer breastfeeding duration; in the NP Cohort, impact also increased with earlier first ANC visit/ART initiation (Figures S1–S4). Across the scenario analyses, the relative reduction in total VT for VL-100% increased the most under point-of-care VL testing (NP = 4.0%; KP = 16.2%) and decreased the most under a 50% reduction in adherence counseling (NP = 1.5%; KP = 8.3%; Figures S5–S6). Across all sensitivity and scenario analyses, this relative reduction ranged from 9.7% to 21.2% in the KP Cohort and from 0.9% to 4.9% in the NP Cohort and was 4–6 times higher in the KP Cohort than in the NP Cohort in the majority of analyses (Figure S7).

DISCUSSION

Given its potential to improve rates of viral suppression, VL monitoring among PBFW living with HIV is considered an important component of PVT care [2]. However, little is known about the reductions in VT that can be achieved by improving fidelity to VL monitoring guidelines. Using a simulation-based model of Kenyan PBFW, we found that relative reductions in VT associated with maternal VL monitoring are likely to be substantially greater among KP women than among NP women. While achieving 50% to 100% adherence to guidelines in this latter group may only reduce their VT by 1.6–2.7%, similar adherence levels in KP women may reduce their VT by 9.1–14.3%. Our sensitivity and scenario analyses suggest that while these percent reductions may vary due to parameter uncertainty, VL monitoring can most likely achieve over double the percent reduction in VT among KP women than it can among NP women.

Two factors increased the benefit of VL monitoring in KP women. First, the KP Cohort had multiple routine VL tests before delivery, while the NP Cohort’s single VL test prior to delivery did not occur early enough to reduce peripartum transmission. Second, the KP Cohort accumulated a larger proportion of women with unsuppressed VL who were also on ART by the time of VT exposure. This is precisely the group that stands to benefit from VL monitoring, as being engaged in care is a prerequisite for receiving a VL test, and only individuals with unsuppressed VL benefit from it. This group was larger in the KP Cohort than in the NP Cohort because of 1) greater ART engagement and 2) a longer time between ART initiation and conception, which increased the number of women whose ART was ineffective by key maternal timepoints.

The limited impact of VL monitoring in NP women raises the question of what supporting interventions could improve its PVT potential in this population. Our sensitivity analyses demonstrated that an earlier first ANC visit would increase the PVT potential of VL monitoring by reducing peripartum VT, but improving ANC access is likely to be challenging [29]. Alternatively, we found that a point-of-care VL test would also allow VL monitoring to reduce peripartum VT. For postpartum VT reduction, a retention support intervention might increase the impact of VL monitoring [30], given that its benefit was limited by ART interruption.

VL monitoring can prevent only a small fraction of all VT in Kenya, as 53% of infants with new infections are born to mothers who did not initiate ART, a population not represented in this model [1]. However, VL monitoring remains an important component of PVT care, as no single intervention will address all the existing gaps in the PVT service cascade [31]. Moreover, improvements in ART coverage and retention will increase the relevance of VL monitoring. In addition, although we focused on its VT impact, maternal VL monitoring has additional benefits to mothers and infants. Consistent with prior studies among all adults [32], we found that VL monitoring reduced maternal deaths, and maternal survival significantly decreases the risk of infant mortality [33], an effect we did not quantify as it was outside the scope of our model.

It is important to note that there are currently no well-established interventions to achieve improved adherence to maternal VL monitoring guidelines. Gaps in the VL testing cascade are likely due to multiple factors, and interventions to address them may need to be context specific and target issues across the cascade, from performing the VL tests to following up on test results. The cost-effectiveness of these interventions will be an important consideration as policy and program leaders weigh strengthening maternal VL monitoring against other PVT interventions. Future modeling studies can support innovation and decision-making by determining the cost threshold at which a novel adherence intervention would be cost-effective.

Our model-based approach allows us to examine a research question that is difficult to study empirically, given the challenges in controlling guideline adherence and the large sample sizes needed to detect differences in vertical transmission. This approach also facilitated a dual-cohort study design that revealed significant heterogeneity in intervention effect across NP and KP PBFW, a finding that would have been masked by studying PBFW as a whole. The primary limitations of our model-based approach are the uncertainty around parameter inputs and modeling assumptions. While we use the best available data to inform parameters and addressed parameter uncertainty through a wide range of sensitivity analyses, there may be relevant real-world population features that are not captured by our model. First, we assumed that ART effectiveness was the same for NP and KP women, due to a lack of data suggesting otherwise. However, given the challenges inherent in simultaneously managing a new HIV diagnosis and a new pregnancy, NP women may have poorer adherence than KP women [34], and future empirical research may thus reveal that they have lower ART effectiveness. Second, in order to efficiently evaluate and contrast the impact of VL monitoring in NP vs. KP PBFW, we used a single timeline for each cohort (Figure 1), rather than represent all possible timeline permutations of HIV and maternal events. In particular, we did not address the unique characteristics of perinatally-infected PBFW, including long-term ART experience, adherence challenges, and potentially limited ART regimen options [35]. Finally, while some experts have proposed using VL testing to triage the type of prophylaxis given to infants [2, 36], we did not examine this due to the lack of data stratifying VT risk by infant prophylaxis status. Each of these limitations also provide opportunities for further modeling research, as novel empirical data become available and our model development expands.

CONCLUSION

We present a novel evaluation of maternal VL monitoring in the context of DTG-based ART. We found that VL monitoring has greater potential for reducing VT in known-positive women already on ART at conception than newly positive women who initiate ART during pregnancy. This analysis provides a foundation for future evaluations of intervention packages to reduce and ultimately eliminate VT in this region.

Data Availability Statement:

The data that supports the findings of this study are included in the article or in the supplementary information.