HIV viral load patterns and risk factors among women in prevention of mother-to-child transmission (PMTCT) programs to inform differentiated service delivery (DSD)

Abstract

Background:

Differentiated service delivery (DSD) approaches decrease frequency of clinic visits for individuals who are stable on antiretroviral therapy (ART). It is unclear how to optimize DSD models for postpartum women living with HIV (PWLH). We evaluated longitudinal HIV viral load (VL) and cofactors, and modelled DSD eligibility with virologic failure (VF) among PWLH in PMTCT programs.

Methods:

This analysis used programmatic data from participants in the Mobile WAChX trial (NCT02400671). Women were assessed for DSD-eligibility using the WHO criteria among general people living with HIV (receiving ART for ≥6 months and having at least one suppressed VL [<1,000 copies/mL] within the past 6 months). Longitudinal VL patterns were summarized using group-based trajectory modelling (GBTM). VF was defined as having a subsequent VL ≥1,000 copies/mL after being assessed as DSD-eligible. Predictors of VF were determined using log-binomial models among DSD-eligible PWLH.

Results:

Among 761 women with 3,359 VL results (median 5 VL per woman), a three-trajectory model optimally summarized longitudinal VL, with most (80.8%) women having sustained low probability of unsuppressed VL. Among women who met DSD criteria at 6 months postpartum, most (83.8%) maintained viral suppression until 24 months. Residence in Western Kenya, depression, reported interpersonal abuse, unintended pregnancy, nevirapine-based ART, low-level viremia (VL 200-1,000 copies/mL), and drug resistance were associated with VF among DSD-eligible PWLH.

Conclusions:

Most postpartum women maintained viral suppression from early postpartum to 24 months and may be suitable for DSD referral. Women with depression, drug resistance and detectable VL need enhanced services.

Background

Differentiated service delivery (DSD) approaches adapt the frequency of clinic-based visits for individuals who are clinically stable on antiretroviral therapy (ART) ¹. This patient-centered approach enables healthcare to focus on individuals in need of more intensive clinical care and benefits patients, healthcare providers and health systems ^2,3. In 2017, Kenya implemented DSD for individuals on ART for at least 6 months with suppressed HIV viral load (VL <1,000 copies/mL) ⁴. DSD allows for 6-monthly clinic visits and annual VL testing. Current Kenyan DSD guidelines recommend that pregnant and breastfeeding women have specific visits aligned with prevention of mother-to-child HIV transmission (PMTCT) programs ^5-7 and include VL testing 6-monthly until cessation of breastfeeding ⁸, therefore as unstable and do not qualify for DSD ^5-7.

The World Health Organization (WHO) has recommended that the DSD definition of being stable apply to all populations ⁹. However, it is unclear how to implement DSD models for mothers with HIV, given their need for routine antenatal and postnatal care visits and HIV-exposed infant follow-up visits. In addition, retention in care and adherence have been challenging for PMTCT programs ^10-14. Studies in sub-Saharan Africa have reported facility-level reasons for women not engaging with clinic-based care, including lack of time and difficulty with transport to clinic ¹⁵, inadequate care ^12,16,17, high staff turnover and long waiting time ^18,19. Women stable with HIV in late postpartum may benefit from DSD approaches through multi-month ART dispensing, community-based ART delivery, and longer intervals between HIV clinical visits. A randomized clinical trial (RCT) in South Africa reported improved viral suppression up to 24 months among women with HIV referred to community-based adherence clubs at their first postnatal visits compared to women who received standard of care ²⁰. DSD approaches require an understanding of long-term viral suppression trajectories and individuals who may be at risk for poorer outcomes with DSD. To date, there is limited evidence to inform tailored DSD models for women in PMTCT programs. In this study, we aimed to describe longitudinal HIV VL patterns and cofactors among PWLH in PMTCT programs in Kenya, and assess the frequency potentially eligible to receive DSD and subsequent risk of virologic failure (VF).

Methods

Study design and population

This nested observational study used data from a completed 3-arm RCT (Mobile WAChX, ClinicalTrial.gov number NCT02400671 ^21,22). The parent study assessed the short messaging service (SMS) among women attending PMTCT programs followed-up through 2 years postpartum (study period: November 22, 2015 to May 4, 2017) ²². Briefly, the trial enrolled pregnant women with HIV from six maternal-child health (MCH) clinics in Nairobi and Western Kenya who were aged ≥14 years and had daily access to a mobile phone. The intervention did not have an impact on maternal VL or retention as reported previously ²². The RCT was approved by the Kenyatta National Hospital/University of Nairobi Ethical Review Committee and UW Institutional Review Board.

Data Collection

At enrollment, a standardized survey was administered to collect data on demographics, depression (using Patient Health Questionnaire 9 [PHQ-9] ²³), abuse history (using Abuse Assessment Screen [AAS] ²⁴), disclosure of HIV status, and ART knowledge based on the Information–Motivation and Behavioral Skills (IMB) model adapted from 15 items from the LifeWindows ART adherence questionnaire ²⁵. Pre-pregnancy ART initiation was determined based on last menstrual period date and ART starting date. Data on ART use history was abstracted from the mother's MCH booklet. HIV VL testing results were obtained from the routine VL monitoring system of the Kenya National AIDS & STI Control Program (NASCOP). If programmatic VL results were not available, VL testing was conducted by the study on maternal plasma samples. ART resistance was assessed using an oligonucleotide ligation assay (OLA) on maternal plasma samples and was defined as detection of any resistance mutations by OLA at an abundance of ≥10% ²². Programmatic data of clinic visits was obtained from clinic paper records and electronic medical records ²⁶. Pharmacy refill data was abstracted from records of ART doses administered during HIV care visits ²².

Study outcomes

Being DSD-eligible was defined according to WHO criteria as receiving ART for at least 6 months, with at least one suppressed VL (<1,000 copies/mL) within the past 6 months. Low-level detection (LLD) was defined as having HIV VL <1,000 copies/mL but ≥200 copies/mL. VF was defined as having HIV VL ≥1,000 copies/mL after being assessed as DSD-eligible. VL data in this analysis were restricted to data from ≥4 months since ART initiation, in order to include only data after it was biologically plausible for VL to be suppressed by ART; VF results within 30 days of a previous VF were excluded. ART adherence was calculated as the proportion of days between visits that were covered by ART doses dispensed based on self-report and was classified as adherent (100%) vs. non-adherent (<100%) ²². Any missed daily pill in the past refill period was defined as non-adherence. Any appointment not made within 14 days of the scheduled visit was defined as a missed visit ²⁷.

Statistical analysis

Chi-square tests were used to measure the difference in the distribution of characteristics for outcomes. Logistic group-based trajectory modeling (GBTM) was performed to identify discrete trajectories of HIV VL patterns. GBTM is an application of finite mixture modeling that assumes the population is composed of distinct groups, each with a different underlying trajectory and every individual in the group approximately follows the same patterns of behavior of outcome over time ²⁸. GBTM has been used to evaluate virologic outcomes in high-income countries ^29-35. One study in Zambia reported six engagement trajectories in medication possession ratio and retention in care among new ART starters ³⁶, but overall evidence on viral trajectories in resource-limited settings is limited due to lack of sufficient longitudinal VL data.

In this study, the number of trajectory groups was hypothesized a priori to be between 1 and 5. The number of trajectory groups and the appropriate function of trajectory polynomials of time window were tested based on observed plots and the Bayesian information criterion (BIC); the model with the lowest BIC value representing the statistically optimal number of latent groups ³⁷. Since not all women had equivalent observation times and VL testing results, posterior probabilities were based on comparisons between trajectories and observed data in study follow-up. Women were then assigned to the trajectory group with which they had the highest probability of membership based on their estimated posterior probabilities.

The proportion of women eligible for DSD was evaluated at different postpartum timepoints, and outcomes were assessed among women who had at least one subsequent VL result. Log-binomial regression with robust standard errors was used to identify predictors of VF by 24 months postpartum among women who were DSD-eligible estimated at 6 months. All analyses were conducted using RStudio Version 1.2.5042 (RStudio, Inc).

Results

Overall, 761 women from the Mobile WAChX trial were included in this study. At enrollment, median age was 27 years (interquartile range [IQR] 23-31) and median gestational age was 24.2 weeks (IQR 18.1-29.1). Sixty percent (456/761) of women started ART before the current pregnancy, and 82.6% (629/761) reported no health problems at pregnancy. By 24 months postpartum, a total of 3,359 VL testing results were available for analysis, and women had a median of 5 VL testing results (IQR 3-6) during study follow-up, with the median duration between two tests of 5.3 months (IQR 3.6-6.9) (Figure 1). Among 761 women, 502 (66.0%) had VL results at delivery, and the cumulative proportion with VL available was 87.8% and 96.5% by 6 weeks and 6 months postpartum, respectively (Figure 1). By 6 months postpartum, women had a median of 3 VLs (IQR 2-4), and 78.1% of women had at least 2 VLs (Figure 1). The cumulative proportion of women ever having VL ≥1,000 copies/mL by 24 months postpartum was 26.4% (201/761). Among 698 women with at least two VLs available by 24 months postpartum, 614 had viral suppression at their first VL after enrollment, of whom 519 (84.5%) maintained viral suppression (<1,000 copies/mL). Among 84 women whose first VL after enrollment was unsuppressed, 8 (9.5%) had persistent non-suppression and 33 (39.3%) became virally suppressed and maintained suppression.

Figure 1. VL tests through 24 months postpartum (N=761 women).

^a Time window including 1 month after

^b Time window including 3 months after

*Among women who had at least 2 VLs

A three-trajectory model was selected to present longitudinal patterns of VL from enrollment to 24 months postpartum: (1) “sustained low probability of viral non-suppression” (80.8% of participants); (2) “high probability of delayed viral suppression” (5.9% of participants); and (3) “sustained high probability of viral non-suppression” (13.3% of participants) (Figure 2). The “high probability of delayed viral suppression” and “sustained high probability of viral non-suppression” groups were best represented by cubic trajectories, while the “sustained low probability of viral non-suppression” group was best represented by a non-time varying constant function. Most women (86.7%) were likely to achieve viral suppression by early postpartum and remain virally suppressed afterwards. In a sensitivity analysis among 531 women with at least four VL data points available with at least a 3-month window between tests, the number of trajectories and proportions remained similar (data not shown). The primary trial intervention did not have significant effects on the VL trajectories in this analysis.

Figure 2. VL trajectories for PBFW by probability of HIV VL ≥1,000 copies/mL.

Longitudinal trajectories of unsuppressed (≥1,000 copies/mL) during 24-month study follow-up. Blue = ‘Sustained low probability of viral non-suppression; red = ‘High probability of delayed viral suppressed; green = ‘Sustained high probability of viral non-suppression’. Dots represent the observed proportion of individuals with unsuppressed VL in a given interval among individuals assigned to distinct VL trajectory groups. The smoothed lines depict the predicted probability of unsuppressed VL for each trajectory group from the GBTM. The grey dashed lines represent 95% Confidence intervals for the predicted probability of unsuppressed VL at each time point. The y axis represents the probability of unsuppressed VL and the x axis represents the number of time points (every 3 months since delivery).

During pregnancy, 507 women were already on ART for at least 6 months, and 84.6% (429/507) women met WHO eligibility criteria to receive DSD (on ART ≥6 months and at least 1 suppressed in the past 6 months) (Figure 3a). At 6 weeks and 3 months postpartum, the proportion eligible for DSD increased, mainly because women had more recent VL testing data available (Figure 3a, 3b). Among women who met the DSD criteria assessed at early postpartum periods, most women maintained viral suppression until the 2-year postpartum endpoint (Figure 3c). Among women who did not maintain suppression, the median time from viral suppression to the date of VF was 9.1 months (IQR 4.6-15.2). At late postpartum periods (18 months and 24 months), fewer women had subsequent VL testing results available for assessing the maintenance status, but among those who had subsequent data, the proportion of those who maintained viral suppression increased (Figure 3b, Figure 3c). Starting ART pre-conception or in pregnancy did not significantly influence the proportion who developed VF (data not shown). In a sensitivity analysis among 371 women with VL data available in pregnancy and at least 1 VL available after 18 months postpartum, the proportions who developed VF remained similar (data not shown).

Figure 3. Proportion of women who would meet DSD criteria and likelihood of remaining suppressed, applying standard adult DSD guidelines and PMTCT monitoring window.

Black/solid = on ART >6mo; Blue/dashed = with treatment success; Green/dotted = having subsequent data; Brown/long dashed = maintained suppressed. Bars = number of women within each category in a time interval; lines = proportion of each category among previous category. x axis depicts time window; the left y axis depicts numbers; the right y axis depicts proportions. Pregnancy time window including 1 month before and after; Postpartum window including 3 months before and after.

Among 580 women eligible for DSD when evaluated at 6 months postpartum, residence in Western Kenya was associated with higher risks of VF (prevalence ratio [PR] 2.48, 95% confidence interval [95%CI] 1.42-4.35; p=0.002) (Table 1). In site-adjusted regression models, women with at least moderate depressive symptoms (PHQ-9 score ≥5) (adjusted prevalence ratio [aPR] 2.19, 95%CI 1.36-3.54; p=0.001) and those with reported abuse in the past year (aPR 1.96, 95%CI 1.08-3.56; p=0.027) had increased risk of VF (Table 1). Women reporting their last pregnancy was intended had a lower risk of VF (aPR 0.63, 95%CI 0.41-0.96; p=0.030). Women who received nevirapine (NVP) as first-line regimen had higher risk of VF than women who received efavirenz (EFV) (aPR 2.03, 95%CI 1.25-3.31; p=0.005) as did women enrolled with a drug resistance mutation (aPR 8.73, 95%CI 4.30-17.7; p<0.001), and women who had LLD within the past 6 months (aPR 2.16, 95%CI 1.36-3.44; p=0.001) (Table 1). Women’s age, education level, marital status, IMB score, self-reported ART adherence and ART initiation time were not associated with later VF.

Table 1.

Risk factors of VF among women who would meet the DSD criteria at 6 months postpartum

Characteristic	Overall, N = 580	VF, N=94	Maintained VS, N=486	Site-adjusted PRe (95%CI); p-value	Multi-adjusted aPRe (95%CI); p-value
Western (ref. Nairobi)	400 (69.0%)	78 (83.0%)	322 (66.3%)	2.48 (1.42-4.35); 0.002	2.32 (1.20-4.47); 0.012
Young age (15-24 years)	187 (32.2%)	34 (36.2%)	153 (31.5%)	1.22 (0.78-1.91); 0.376
Primary school completed	448 (77.2%)	73 (77.7%)	375 (77.2%)	1.16 (0.69-1.96); 0.568
Married	504 (86.9%)	78 (83.0%)	426 (87.7%)	0.75 (0.41-1.34); 0.326
Employed	313 (54.2%)	50 (53.2%)	263 (54.3%)	0.83 (0.54-1.29); 0.416
At least mild depression symptoms a	135 (23.3%)	35 (37.2%)	100 (20.6%)	2.19 (1.36-3.54); 0.001	1.95 (1.13-3.35); 0.016
Abuse last year	64 (11.0%)	16 (17.0%)	48 (9.9%)	1.96 (1.08-3.56); 0.027	2.17 (1.08-4.33); 0.029
Time to clinic >1h	43 (7.5%)	5 (5.4%)	38 (7.9%)	0.61 (0.23-1.63); 0.326
Primigravida	71 (12.2%)	12 (12.8%)	59 (12.1%)	1.27 (0.65-2.46); 0.483
Pregnancy intended	310 (53.7%)	40 (42.6%)	270 (55.9%)	0.63 (0.41-0.96); 0.030	0.84 (0.50-1.39); 0.493
Start ART before pregnancy	408 (71.2%)	69 (73.4%)	339 (70.8%)	1.07 (0.65-1.77); 0.787
Know status before pregnancy	423 (73.1%)	69 (73.4%)	354 (73.0%)	0.93 (0.57-1.54); 0.789
Disclosed status to anyone	515 (89.6%)	85 (91.4%)	430 (89.2%)	1.21 (0.56-2.64); 0.631
Ever missed a dose in the past 30 days b	61 (11.6%)	8 (9.4%)	53 (12.0%)	0.74 (0.34-1.63); 0.454
Ever missed a dose in the past 90 days b	89 (17.3%)	16 (19.5%)	73 (16.9%)	1.23 (0.66-2.3); 0.517
ZDV-based regimen (ref: TDF)	44 (8.4%)	10 (11.6%)	34 (7.8%)	1.51 (0.7-3.24); 0.294
NVP-based ART regimen (ref: EFV)	132 (26.0%)	33 (39.8%)	99 (23.3%)	2.03 (1.25-3.31); 0.005	1.71 (0.96-3.02); 0.066
Mutations	41 (7.1%)	22 (23.4%)	19 (3.9%)	8.73 (4.3-17.7); <0.001	8.00 (3.66-17.47); <0.001
Low IMB scorec	254 (46.4%)	45 (51.1%)	209 (45.4%)	1.13 (0.7-1.83); 0.616
Ever LLDd within the last 6 months	118 (20.3%)	32 (34.0%)	86 (17.7%)	2.16 (1.36-3.44); 0.001	1.51 (0.85-2.68); 0.156

^aevaluated by Patient Health Questionnaire 9 (PHQ9), a score >5 indicating at least moderate depressive symptoms

^bSelf-reported at baseline behavioral survey

^cInformation-Motivation-Behavioral (IMB) score evaluated by 15 items from LifeWindows ART adherence questionnaire

^dLLD: having HIV VL ≥200 and <1000 copies/mL

^ePR estimated by Log-binomial regression adjusting for site

In a multivariate model adjusting for variables with univariate p-value <0.1, residence in Western Kenya, depression, reported abuse and drug resistance remained significantly associated with VF among women who met DSD criteria (Table 1). In secondary analyses among 432 women eligible for DSD at 6 weeks postpartum, all associations remained the same, while among 591 women eligible for DSD at 12 months postpartum, associations with pregnancy intention and NVP-based ART use were no longer significant (data not shown).

Overall, 733 (96.3%) women had pharmacy refill data available, with a median of 18 refill cycles (IQR 12-22) per woman. Most (98.5%) women ever missed at least one dose within a refill cycle over time. When evaluating the effect of ART adherence on subsequent VF among DSD-eligible women, women who had a history of non-adherence by 6 weeks postpartum (missed at least one dose in the past 6 months) had a higher risk of VF (18.7% vs. 11.9%, p=0.029). At 6 months and 12 months, the frequency of VF among women who did not adhere were higher than among women who adhered to ART, but the differences were not significant (Table 2). Overall, 743 (97.6%) women had clinic visit data available from MCH records, with a median of 17 scheduled visits (IQR 12-22) per woman. Most (83.4%) women had one missed visit (late for >14 days for at least 1 scheduled visit) at some timepoint. Among DSD-eligible women, missed visits were not associated with VF, regardless of time of summarizing missed visits (Table 2).

Table 2.

ART adherence and retention in care among women who met the DSD criteria at different postpartum time of DSD assessment

ART adherence*				Retention in care**
	N	Proportion of VF	p-value		N	Proportion of VF	p-value
At 6 weeks postpartum a	407		0.2	At 6 weeks postpartum a	422		0.2
Not adherent		18.7% (65/348)		Not retained		13.7% (26/190)
Adherent		11.9% (7/59)		Retained		20.3% (47/232)
At 6 months postpartum a	563		0.4	At 6 months postpartum a	569		0.4
Not adherent		17.0% (91/536)		Not retained		15.4% (51/332)
Adherent		7.4% (2/27)		Retained		18.1% (43/237)
At 12 months postpartum a	579		0.8	At 12 months postpartum a	576		0.15
Not adherent		10.7% (58/544)		Not retained		9.1% (30/328)
Adherent		11.4% (4/35)		Retained		12.9% (32/248)

^*ART adherence and DSD-eligibility were evaluated among women who had pharmacy refill visits within the past 6 months and had VL data after the evaluation window; not adhere: ever missed a daily dose within the past 6 months

^**Retention and DSD-eligibility were evaluated among women who had clinic visits scheduled within the past 6 months and had VL data after the evaluation window; not retained: ever missed a visit within the past 6 months

^aTime window including 3 months before and 3 months after

Discussion

In this study, we evaluated longitudinal VL patterns among women living with HIV receiving PMTCT services in Kenya from pregnancy to 24 months postpartum. We found that most women met standard DSD eligibility criteria by 6 months postpartum (being on ART ≥6 months and at least one viral suppression within the past 6 months), and among women who had follow-up VL data, we found 84.5% of women were persistently suppressed; of women whose first VL since enrolment was unsuppressed, 39.3% later suppressed and maintained suppression until the end of follow-up. The frequency of suppression we observed was similar to a cohort study in South Africa, which reported that 16.6% of pregnant women with HIV ever had VL ≥1,000 copies/mL by 24 months postpartum ³⁸. Studies in Ethiopia and Malawi with shorter follow-up periods have also reported similar frequencies of viral non-suppression (10.3% in pregnancy and 12.1% at 6 months postpartum, respectively) ^39,40. Our study cohort did not receive currently recommended dolutegravir-based regimens, which have greater efficacy for viral suppression than efavirenz-based regimens ^41,42. Thus, we may have underestimated potential rates of DSD-eligibility and viral suppression.

Prior clinical trials among individuals in the general population stable on ART in sub-Saharan Africa have shown that those receiving community-based ART services had comparable VL or significantly lower risks of unsuppressed VL than those receiving clinic-based standard of care ^43-46. While many countries have endorsed DSD models in national guidelines ⁴⁷, postpartum women regardless of VL level are often not considered as stable ^48,49 and excluded from DSD models, even though the global community seeks to find ways to offer DSD to all people stable on ART ⁹. Breastfeeding period can extend two to three years in resource-limited settings, and fertility rates are such that women can spend much of their adult life pregnant or breastfeeding. Thus, many mother-infant pairs will have prolonged time of restricted access to DSD models. There is a need to align the criteria determining individuals as “stable” with what has already been established by WHO for non-pregnant adults ⁹ while still recognizing the continued need for routine antenatal and postnatal care.

The three distinct group based VL trajectory patterns identified using GBTM yielded comparable patterns to those described in a study of women living with HIV in Canada ²⁹. Achieving and maintaining viral suppression is the key to minimize vertical HIV transmission and to optimize women's health outcomes ⁵⁰. Our study provided supportive evidence for inclusion of PWLH in DSD models. We focused on those who were clinically stable during early postpartum periods for DSD in later postpartum, reasoning that women and children may need more intensive services than DSD in the peripartum period.

We found depression and reported abuse in the last year significantly predicted subsequent VF among women eligible for DSD at 6 months postpartum. These results suggest the importance of emotional stressors as a determinant of VF. In our study, 23% of women experienced at least mild depressive symptoms, similar to rates of antenatal depression among African women with HIV in a systematic review ⁵¹. Depression has been reported as risk factors of poor adherence ^52-54 and high HIV VL ^53,55. While our study did not find significant effects of self-reported ART adherence and retention on VF, it is likely that depression influenced ART adherence and self-reported adherence overestimates actual adherence ⁵⁶. Given the high prevalence of depression, and association with VF, standard depression screening during pregnancy and postpartum and referral for mental health services may be required to ensure sustained viral suppression.

We found that women with LLD (HIV VL ≥200 copies/mL) had significantly higher risks of subsequent VF even when they met DSD eligibility criteria with suppressed VL <1,000 copies/mL. There have been concerns regarding non-sustained viral suppression during a long follow-up throughout postpartum ⁵⁰. A study in Malawi reported increased risk of non-suppression at 24 months among women with prior detectable VL (40-1,000 copies/mL) at 1-6 months postpartum ⁵⁷, and non-durable suppression has been linked to increased risks of vertical transmission in the same cohort ⁵⁸. A randomized clinical trial in Uganda and Kenya among ART-experienced patients with viremia reported that a DSD intervention with streamlined care was associated with higher rates of 3-year viral suppression (67% vs. 47%) ⁵⁹. Since risks of vertical transmission increase with maternal detectable VL ^58,60, maintaining undetectable VL is essential for elimination of vertical transmission ⁶¹. On a structural level, it is important to anticipate and avoid potentially detrimental impacts of less frequent DSD visits on the uptake of standard MCH services and timely HIV VL testing, as well as coverage of early infant diagnosis and vaccination. Models for ART-stable PWLH who have LLD VL need to include components to track closely until a final suppressed or undetected status can be established.

We found HIV drug resistance resulted in >8-fold increased risk for VF, and women taking a NVP-based regimen had a higher risk of VF than women taking an EFV-based regimen. In this cohort, women received fixed-dose ART in line with contemporaneous WHO guidelines. Currently, dolutegravir is the first-line regimen of choice ⁶² and has a high barrier to resistance ⁶³. Monitoring women for drug resistance in PMTCT programs may continue to be useful to identify those who need regimen switches and who may not be good candidates for DSD. We did not find a difference in the risks of VF between women who started ART pre-conception versus those who started ART in pregnancy, perhaps due to generally high levels of ART coverage in this cohort.

Our study had several limitations. First, the WHO criteria for DSD also includes evaluation of current illness ⁹. In this study, women were asked about health conditions; however, we had limited systematic data on other illnesses. We relied on VL testing data, which may overestimate the proportion of those who were classified as stable in the cohort, but in a supplementary analysis excluding women who indicated any health problems at baseline, the effect of depression and drug resistance on VF persisted. Secondly, there have been variations in VL cutoff to define VF in PMTCT programs: a cluster trial with a 3-year follow-up in 32 communities in Kenya and Uganda used a threshold of 500 copies/mL ⁶⁴, and 400 copies/mL has also been used in other studies ^65-68. We found a similar proportion of women who would be classified as DSD-eligible using lower cutoffs (88.4% with <500 copies/mL; 87.7% with <400 copies/mL; 86.2% with <200 copies/mL; 80.0% with <50 copies/mL), but fewer (55.8%) women with the lowest detection level at <20 copies/mL. Thirdly, our models did not accommodate time-varying socioeconomic factors, and the explanatory variables were measured at baseline only. Our definitions of pill count-based adherence and missed visit-based retention may not reflect the true temporal associations between poor engagement and VF. Fourth, our cohort was from 2015-2017 and women received efavirenz-based ART regimens which are less effective than currently recommended dolutegravir-based regimens, likely leading to underestimated rates of DSD eligibility and viral suppression in current PMTCT programs. Finally, the primary study pre-dated DSD guidelines, so DSD was not provided to the study participants, and the study could not assess the impact of DSD on outcomes. Qualitative studies have reported that women in PMTCT programs may be reluctant to use fast-track queues, even when services are in place ^69,70, which suggests that women may prefer more clinical interactions, particularly if postpartum with maternal or infant health concerns. It is important to understand women’s preference on DSD, as well as clinical facilitators and barriers affecting DSD model implementation.

Conclusion

This study has important implications for understanding DSD eligibility among women in PMTCT programs. Most women had sustained low probability of VF, and among women who met the general DSD-eligible criteria, most maintained viral suppression until the 24 months postpartum. Integration of patient-centered DSD strategies, choice in which DSD model is both clinically indicated but also preferred by women, for those stable on ART may be feasible and beneficial within the framework of perinatal care. Women with depression or reported abuse need enhanced mental health support to prevent VF. Drug resistance testing may be useful to exclude women from DSD. Lower VL threshold may exclude many women from DSD. Implementation science studies focused on DSD models in MCH settings are needed for postpartum women stable on ART.