To compare time to achieve viral load <400 copies/ml and <1000 copies/ml in HIV-infected antiretroviral (ARV) -naive versus ARV-experienced pregnant women on highly active antiretroviral therapy (HAART).
To compare time to achieve viral load <400 copies/ml and <1000 copies/ml in HIV-infected antiretroviral (ARV) -naive versus ARV-experienced pregnant women on highly active antiretroviral therapy (HAART).
Retrospective cohort study.
Three university medical centers, USA.
HIV-infected pregnant women initiated or restarted on HAART during pregnancy.
We calculated time to viral load <400 copies/ml and <1000 copies/ml in HIV-infected pregnant women on HAART who reported at least 50% adherence, stratifying based on previous ARV exposure history.
Time to HIV viral load <400 copies/ml and <1000 copies/ml.
We evaluated 138 HIV-infected pregnant women, comprising 76 ARV-naive and 62 ARV-experienced. Ninety-three percent of ARV-naive women achieved a viral load < 400 copies/ml during pregnancy compared with 92% of ARV-experienced women (P = 0.82). The median number of days to achieve a viral load < 400 copies/ml in the ARV-naive cohort was 25.0 (range 3.5–133; interquartile range 16–34) days compared with 27.0 (range 8–162.5; interquartile range 18.5–54.3) days in the ARV-experienced cohort (P = 0.02). In a multiple predictor analysis, women with higher adherence (adjusted relative hazard [aRH] per 10% increase in adherence 1.29, 95% confidence interval [CI] 1.08–1.54, P = 0.01) and receiving a non-nucleotide reverse transcriptase inhibitor (NNRTI) -based regimen (aRH 2.48, 95% CI 1.33–4.63, P = 0.01) were more likely to achieve viral load <400 copies/ml earlier. Increased baseline HIV log10 viral load was associated with a later time of achieving viral load <400 copies/ml (aRH 0.60, 95% CI 0.39–0.92, P = 0.02). In a corresponding model of time to achieve viral load <1000 copies/ml, adherence (aRH per 10% increase in adherence 1.79, 95% CI 1.34–2.39, P < 0.001), receipt of NNRTI (aRH 2.95, 95% CI 1.23–7.06, P = 0.02), and CD4 cell count (aRH per 50 count increase in CD4 1.12, 95% CI 1.03–1.22, P = 0.01) were associated with an earlier time to achieve viral load below this threshold. Increasing baseline HIV log10 viral load was associated with a longer time of achieving viral load <1000 copies/ml (aRH 0.54, 95% CI 0.34–0.86, P = 0.01). In multiple predictor models, previous ARV exposure was not significantly associated with time to achieve viral load below thresholds of <400 copies/ml and <1000 copies/ml.
Pregnant women with ≥50% adherence, whether ARV-naive or ARV-experienced, on average achieve a viral load <400 copies/ml within a median of 26 days and a viral load of <1000 copies/ml within a median of 14 days of HAART initiation. Increased adherence, receipt of NNRTI-based regimen and lower baseline HIV log10 viral load were all statistically significant predictors of earlier time to achieve viral load <400 copies/ml and <1000 copies/ml. Increased CD4 count was statistically significant as a predictor of earlier time to achieve viral load <1000 copies/ml.
Over the last several decades, the risk of mother to child transmission (MTCT) of HIV has dramatically declined in the setting of highly active antiretroviral therapy (HAART) use in pregnancy.[1-3] Specifically, maternal viral load has been shown to be the strongest predictor of perinatal transmission, with a viral load of <50 copies/ml at time of delivery associated with <1% vertical transmission risk.[2, 3] Additionally, risk of MTCT has been demonstrated to decrease for each additional week a woman is on HAART during pregnancy. In the absence of antiretroviral (ARV) therapy, previous observational studies among nonbreastfed cohorts demonstrated that approximately two-thirds of perinatal transmission occurred intrapartum.[5, 6] However, in the HAART era, approximately 80% of MTCT now occurs in the antepartum period before the onset of labour. As such, it is now important to identify interventions that address this residual antepartum risk of transmission, as suppression of viral load earlier in pregnancy may further reduce risk of perinatal transmission during this period. The duration of time needed to achieve a viral load below the specific level of detection in HIV-infected pregnant women on HAART remains unclear, and recommendations regarding the optimal timing of ARV therapy initiation during pregnancy may vary.[8-11]
Whereas HIV viral dynamics have been well described in non-pregnant cohorts, physiological and pharmacological changes in pregnancy may limit the generalisability of these studies to a pregnant population.[12-14] Few studies have examined virological response in HIV-infected pregnant women receiving HAART,[15-21] and many of these studies have included exclusively or predominantly ARV-naive cohorts. Additionally, recent data show that an increasing proportion of pregnant women are ARV-experienced before pregnancy, though not more likely to actually be taking ARVs at the time of conception. In light of the conflicting results of the few studies that have assessed differences in time to achieve viral load suppression in pregnant women who are ARV-naive versus ARV-experienced, further examination of potential confounders in the association between ARV exposure history and time to undetectable viral load during pregnancy is needed.[17, 20, 21] The potential difference in viral decay based on previous ARV exposure and the increasing proportion of ARV-experienced pregnant women justify further analysis of viral decay based on previous ARV exposure.
We therefore sought to compare time to <400 copies/ml and <1000 copies/ml HIV RNA virus in ARV-naive versus ARV-experienced pregnant women on HAART and to describe factors associated with achieving HIV viral suppression.
We conducted a retrospective cohort study of HIV-infected pregnant women receiving prenatal care at the University of California, San Francisco (UCSF) and San Francisco General Hospital Bay Area Perinatal AIDS Center (SFGH BAPAC) from August 1997 to April 2009 and Baylor College of Medicine from April 2006 to April 2009. This study was approved by the investigational review boards at all three institutions.
To be included in this study, women needed to be engaged in prenatal care at one of the three institutions with a documented positive HIV antibody and have a pre-HAART initiation HIV detectable viral load above 400 copies/ml. Study participants also needed at least one additional viral load measurement after HAART initiation. Women could be either ARV-naive or ARV-experienced and reinitiating ARV medication during the index pregnancy. Status of ARV use before the index pregnancy was determined by patient report, as well as extensive review of the medical records by the clinician. Women were considered ARV-experienced if they had previously taken HAART or abbreviated monotherapy or dual ARV therapy for the purpose of either their own health or for the prevention of MTCT. ARV-experienced women needed to have last taken a protease inhibitor (PI), non-nucleoside reverse transcriptase inhibitors (NNRTI), or a nucleoside reverse transcriptase inhibitor (NRTI) -based regimen at least 2 months before HAART reinitiation during the index pregnancy. To be included in this study, women also needed to initiate HAART, defined as an ARV regimen containing at least three ARV medications, during the index pregnancy. Subjects who entered the study with a viral load <1000 copies/ml at onset of prenatal care were only included in analyses of time to achieve viral load <400 copies/ml and were excluded from the secondary analyses of time to achieve viral load < 1000 copies/ml. We excluded pregnant women with a viral load below the level of detection (<50, <75, <400, or <500 copies/ml assay depending on the assay used by each institution and at different time-points) before HAART initiation during the index pregnancy, as well as those women with poor adherence (defined as <50% by self-report).
The primary outcome was defined as the number of days to reach an HIV RNA viral load < 400 copies/ml. This level was chosen because it is one of the criteria used for MTCT purposes and was the most applicable viral load detection level for our subject cohort given that this level was most inclusive of the different viral load assay thresholds (<50, <75, <400 copies/ml) used during our study period. The second outcome we used was number of days to achieve a viral load of<1000 copies/ml, given that standard care is to perform a caesarean delivery for women with viral load > 1000 copies/ml at term.[11, 23]
The definition of viral suppression depended on which HIV viral load assays were used. At the UCSF and SFGH laboratories, the lower limit of HIV RNA detection was 50 copies/ml (Chiron Quantiplex HIV-1 RNA 3.0 assay [bDNA]; Emeryville, CA) from 1997 until October 2002 and 75 copies/ml (Chiron Quantiplex HIV-1 RNA 3.0 assay [bDNA]; Emeryville, CA until 2004 and Siemens Versant HIV-1 RNA 3.0 Assay [bDNA]; Deerfield, IL after 2004) for the remainder of the study period. Three women seen at SFGH obtained HIV viral load measurements at an outside laboratory with a different lower limit of detection (lower limit viral load for one woman was 500 copies/ml and for two women was 400 copies/ml). At the Baylor laboratory, the lower limit detection through December 2006 was 400 copies/ml (Roche Amplicor HIV-1 Monitor Test, version 1.5 Standard; Pleasanton, CA, USA). Starting in January 2007 and for the remainder of the study period, Baylor clinicians had the option to order either Roche Amplicor HIV-1 Monitor Test, version 1.5 Standard (400 copies/ml) or Roche Amplicor HIV-1 Monitor Test, version 1.5 UltraSensitive (50 copies/ml).
Baseline viral load was defined as the HIV viral load result obtained closest to the time before HAART initiation during the index pregnancy, but no more than 4 months before initiation of HAART or 1 day after initiation. Ninety-nine percent of participants had baseline viral load measured within 3 months before or 1 day after HAART initiation during the index pregnancy, whereas only 1% had baseline viral loads measured between 3 and 4 months before HAART initiation. Viral load closest to delivery was defined as viral load obtained within 3 months before delivery or less than 4 days after delivery. Ninety percent of participants had viral load measured within 1 month before or 1–4 days after delivery and 7% between 1 and 2 months before delivery, whereas only 3% had final viral loads measured between 2 and 3 months before delivery.
We examined a range of potential additional predictors as part of this study to generate a model of time to reach levels below the specified viral load thresholds. These consisted of maternal HIV viral load at time of HAART initiation, maternal CD4 cell count at time of HAART initiation, type of ARV prescribed, gestational age at the time of HAART initiation, class(es) of previous ARV medication, HIV genotypic resistance profile with intermediate or high-level ARV resistance as defined by the Stanford University HIV Drug Resistance Database, sociodemographic characteristics, obstetric characteristics, concurrent maternal infections occurring during index pregnancy, study sites, and self-reported ARV adherence. Medication adherence was calculated as a percentage of missed doses compared with total prescribed doses based on patient self-report at each prenatal visit. Co-infection was defined as concurrent active infection during HAART treatment period including hepatitis B or C virus acute or chronic infection with viraemia, herpes zoster, bacterial pneumonia, tuberculosis, sexually transmitted infections (Neisseria gonorrhoeae, Chlamydia trachomatis, trichomoniasis, syphilis, genital herpes simplex virus), pyelonephritis, skin or soft tissue infection, upper respiratory tract infection, osteomyelitis, meningitis, or other opportunistic infection. Multiparity was defined as one or more births (>20 weeks of gestation) before the index pregnancy.
We based the sample size calculation on detecting a minimum hazard ratio of 1.75 comparing ARV-naive with ARV-experienced women in a log-rank-test analysis. A sample size of 51 women in each group (ARV-naive versus ARV-experienced) was required to have 80% power to detect a difference of this magnitude at the 5% significance level. This was based on the assumptions that the follow-up period during pregnancy would be approximately 300 days and that study participants may have started HAART at any time during pregnancy and continued throughout pregnancy until delivery.
We calculated descriptive statistics using mean and standard deviation, as well as median and range as appropriate to summarise variables. Time to reach specified viral load levels was calculated in the subset of our cohort who achieved the primary and secondary outcomes. Because of interval censoring, these summary statistics of time to reach specified viral load levels were calculated based on assuming that events occurred at the mid-points of censoring intervals. Bivariate associations among categorical data were compared using the chi-square and Fisher's exact tests, and continuous data were compared using the Student's t-test when a normal distribution assumption appears reasonable and by the Mann–Whitney test if distributions were skewed or contained obvious outliers. We conducted all non-survival analyses using stata software, version 9.1 (Statacorp, College Station, TX, USA). Statistical significance was defined by P < 0.05 throughout.
Survival analysis methods formed the basis of primary statistical modelling in this study, with time to reach viral load below threshold level considered as the ‘event time’. Modelling was conducted for the entire sample of women including those who did not achieve viral loads below specified levels who were considered as right censored in the survival analysis. In addition, because of the nature of the data, event times were only known to occur at some point within an interval of time, necessitating the use of interval-censored methods. Furthermore, we compared the frequency of viral load measurements in ARV-naive versus ARV-experienced women to ensure that the two groups did not differ substantially because similar frequency is required for valid interval censored survival comparisons between groups.
Time to HIV viral load < 400 copies/ml and <1000 copies/ml in ARV-naive versus ARV-experienced pregnant women on HAART were first compared using asymptotic log-rank two-sample (permutation) tests; this group comparison test is the ‘default’ non-parametric interval censored survival analysis test (ictest function) of the interval package in R (www.r-project.org). All other survival analyses were based on accelerated failure time models taking into account interval censoring (performed using the base ‘survival’ package in R). The logs of the event times were modelled via a Weibull distribution and robust standard errors were used throughout to guard against too narrow confidence intervals that might arise from modelling assumptions.
Our multiple predictor model-building strategy followed conventional regression method ideas, attempting to minimise the number of models explored and thereby reduce the risk of spurious results. First, models were fitted with one single predictor at a time. Based on the single-predictor models a ‘full’ multiple predictor model was generated that included all variables (additively) that either had an association with the outcome of interest defined by P < 0.10 (in the single predictor models) or were thought to be clinically essential to include in a multiple predictor model irrespective of level of statistical significance. The full multiple predictor model was then reduced by backward elimination, removing the predictor with the largest P-value (other than the clinically essential variables) until all nonclinically essential predictors had P-values >0.1 (in at least one of the <400 copies/ml or viral load <1000 copies/ml models). Hence, for the final multiple predictor viral load < 400 copies/ml and viral load <1000 copies/ml models, we included ARV exposure history, baseline HIV log10 viral load, baseline CD4 cell count, adherence, study site and ARV regimen type.
During the specified study period for each site, 285 HIV-infected pregnant women underwent prenatal care at the three hospitals. We excluded women on HAART with undetectable viral load (as determined by specific assay used) at initiation of pregnancy (n = 69), with insufficient ARV washout period (n = 35), on mono-therapy or dual-therapy regimens (n = 18), with adherence of < 50% (n = 13), and lost to follow-up with no viral load measurements after initiation or restart of HAART (n = 12). Overall, 138 pregnant women fulfilled study inclusion criteria. Baseline demographic and laboratory characteristics are listed in Table 1. In the two hospitals (UCSF and SFGH) combined, of the total 89 pregnancies there were seven pregnancies before 2000; 63 from January 2000 to March 2006; and 19 from April 2006 to April 2009. In the third hospital, Baylor, there were 49 pregnancies from April 2006 to April 2009. Seventy-six (55%) women were ARV-naive and 62 (45%) were ARV-experienced. Indications for previous therapy in the 62 ARV-experienced women included MTCT (38.7%), maternal health (30.1%) and both MTCT and maternal health (30.1%). Of the ARV-experienced participants, 94% had at least a >3-month period elapsed before HAART reinitiation during the index pregnancy. Of the ARV-experienced women who had a history of an NNRTI-based regimen, 67% had a greater than 6-month and 33% had at least a 2-month period elapsed before HAART reinitiation during the index pregnancy. The ARV-naive and ARV-experienced women did not differ significantly by age, ethnicity, insurance status, HAART regimen, participant number at study site, or whether they received directly observed therapy. The ARV-naive women had significantly higher baseline CD4 cell counts (394 versus 304 cells/μl, P = 0.01), more advanced gestational age at time of HAART initiation (23 versus 20 weeks, P = 0.01), more advanced gestational age at the time of delivery (38 versus 37 weeks, P = 0.03), and were more likely to be at least 90% adherent to their HAART regimen (88% versus 71%, P = 0.02). The ARV-experienced women were statistically significantly more likely to be multiparous (86% versus 57%, P < 0.001), experience a concurrent infection during the pregnancy (45% versus 24%, P = 0.01), and have a clinically significant genotypic resistance profile (27% versus 5%, P < 0.001). There was also a potential effect toward ARV-naive women having a lower median HIV viral load at the time of HAART initiation compared with the ARV-experienced women, though this was not statistically significant (11 209 versus 18 482 copies/ml, P = 0.08).
|Characteristic||ARV-naive (n = 76)||ARV-experienced (n = 62)||P-valuea|
|Age, mean (SD)||27.6 (7.04)||28.1 (6.12)||0.70†|
|Ethnicity, n (%)||0.14‡|
|White||12 (15.8)||15 (24.2)|
|Black||27 (35.5)||29 (46.8)|
|Latina||27 (35.5)||13 (21.0)|
|Otherb||10 (13.2)||5 (8.1)|
|State/Emergency insurance, n (%)||72 (94.7)||58 (93.6)||1.00‡|
|Gestational agec (weeks), mean (SD)||22.9 (6.8)||20.1 (6.0)||0.01†|
|<14 weeks||9 (11.8)||9 (14.5)|
|14–27 weeks 6 days||53 (69.7)||47 (75.8)|
|≥28 weeks||14 (18.4)||6 (9.7)|
|Multiparous, n (%)||43 (56.6)||53 (85.5)||<0.001‡|
|Gestational age at delivery (weeks), mean (SD)||38.1 (2.49)||37.1 (3.11)||0.03†|
|First ARV regimen type in pregnancyd, n (%)||0.78‡|
|PI||60 (78.9)||51 (82.3)|
|NNRTI||10 (13.2)||6 (9.7)|
|Triple NRTI||6 (7.9)||4 (6.5)|
|Concurrent PI and NNRTI||0 (0)||1 (1.6)|
|All ARV regimen types in pregnancye, n (%)||0.86‡|
|PI only||60 (79.0)||49 (79.0)|
|NNRTI only||6 (7.9)||6 (9.7)|
|Other combinations||10 (13.2)||7 (11.3)|
|HIV viral loadc (copies/ml), median (range; IQR)||11 209 (82–283 800; 3398–40 369.5)||18 482 (99–556 164; 6643–51 168)||0.08§|
|≥100 000||7 (9.2)||9 (14.5)|
|10 000–99 999||33 (43.4)||33 (53.2)|
|1000–9999||27 (35.5)||17 (24.4)|
|<1000||9 (11.8)||3 (4.8)|
|HIV viral loadc (log10 copies/ml), median (range; IQR)||4.0 (1.9–5.5; 3.5–4.6)||4.3 (2.0–5.7; 3.8–4.7)||0.23‡|
|≥5||7 (9.2)||9 (14.5)|
|4.99–4.0||33 (43.4)||33 (53.2)|
|3.99–3.0||27 (35.5)||17 (24.4)|
|<3.0||9 (11.8)||3 (4.8)|
|CD4 cell countc (cells/μl), median (range; IQR)||394 (21–998; 260.5–587.5)||303.5 (5–1476; 185–416)||0.01§|
|<200||10 (13.2)||17 (27.4)|
|200–499||38 (50.0)||35 (56.5)|
|≥500||28 (36.8)||10 (16.1)|
|Co-infection after ARV start, n (%)||18 (23.7)||28 (45.2)||0.01‡|
|Co-infection after ARV start and before viral load < 400 copies/ml, n (%)||11 (14.7)||13 (21.0)||0.33‡|
|Study site, n (%)||1.00¶|
|UCSF/SFGH||49 (64.5)||40 (64.5)|
|Baylor||27 (35.5)||22 (35.5)|
|Genotypic resistance performed f , n (%)||51 (67.1)||51 (82.3)||0.05‡|
|Clinically significant mutation f , n (%)||4 (5.3)||17 (27.4)||<0.001‡|
|Self-reported adherenceg, n (%)||0.02‡|
|90–100%||67 (88.2)||44 (71.0)|
|50–89%||9 (11.8)||18 (29.0)|
|Directly observed therapy, n (%)||7 (9.2)||11 (17.7)||0.20‡|
The majority (79%) of women in both groups received a PI-based regimen during the pregnancy. Of the 109 (79%) women who received an exclusive PI-based therapy during their index pregnancy, 60 of these were ARV-naive and 49 were ARV-experienced. Furthermore, of these women, 43/60 (72%) of ARV-naive and 41/49 (84%) of ARV-experienced women were treated with a boosted PI regimen (versus unboosted nelfinavir regimen) at some point during their pregnancy. All NNRTI-based regimen use was conducted at the first two hospitals (UCSF and SFGH). Of the 18 women in whom an NNRTI regimen was used at some point in pregnancy, 2/18 (11.1%) were before 2000; 14/18 (77.8%) were between January 2000 and March 2006; and 2/18 (11.1%) were between April 2006 and April 2009. No woman was on an NNRTI-based regimen in the third hospital during their specified study period of 2006–09. During the course of the index pregnancy, 20 women (14%) required a single regimen change, and two women required two regimen changes. Regimen change was defined as addition, discontinuation or substitution of one or more drugs in the woman's regimen. A total of 18 (13%) women underwent directly observed therapy of ARVs at some time during their pregnancy to optimise adherence. Overall, 102 (74%) participants had HIV resistance profile testing, 67% of ARV-naive and 82% of ARV-experienced women. Twenty-one (15%) of the total 138 women demonstrated clinically significant mutations, consisting of 4 (5%) ARV-naive versus 17 (27%) ARV-experienced (P < 0.001). Of these participants, only one ARV-naive woman required regimen change due to resistance profile, the remaining 20 were on a drug regimen during the index pregnancy in which the woman's HIV-1 appeared to be susceptible and did not require change in their ARV regimen because of resistance profile. At time of delivery, eight women (four ARV-naive and four ARV-experienced) required caesarean delivery for HIV perinatal transmission prevention indication based on known viral load. There were no cases of MTCT among the 135 women for whom we had complete delivery data.
During the index pregnancy, 93% of ARV-naive women reached viral load below the level of detection compared with 92% of ARV-experienced women (Table 2). The median number of days (based on use of midpoint of the interval in which an event occurred representing the ‘time’ of the event) to achieve a viral load < 400 copies/ml in the entire cohort was 26.0 (range 3.5–162.5; interquartile range [IQR] 17.5–41). When assessing based on ARV-exposure history, the median number of days to achieve a viral load < 400 copies/ml in the ARV-naive group was 25.0 (range 3.5–133; IQR 16–34) days compared with 27.0 (range 8.0–162.5; IQR 18.5–54.3) days in the ARV-experienced group (P = 0.02). Mean gestational age at which the women achieved viral load < 400 copies/ml was 27.1 weeks for the ARV-naive group and 25.7 weeks for the ARV-experienced group (P = 0.23). There was no statistically significant difference between ARV-naive and ARV-experienced groups with respect to number of women with undetectable viral loads at each visit and number of days between interval viral load assessments.
|Viral load <400 copies/ml||ARV-naive (n = 70)||ARV-experienced (n = 61)||P-valuea|
|HIV viral load < 400 copies/ml achieved in pregnancy, n (%)||65 (92.9)||56 (91.8)||0.82‡|
|Days to viral load < 400 copies/ml, median (range; IQR)b||25 (3.5–133; 16–34)||27 (8–162.5; 18.5–54.3)||0.02§|
|Gestational age at time of viral load < 400 copies/ml (weeks), mean (SD)||27.1 (6.4)||25.7 (5.6)||0.23†|
|HIV viral load <400 copies/ml at time of delivery, n (%)||63 (90.0)||52 (85.3)||0.41‡|
|Viral load <1000 copies/ml||ARV-naive (n = 67)||ARV-experienced (n = 59)||P-valuea|
|HIV viral load <1000 copies/ml achieved in pregnancy, n (%)||64 (95.5)||56 (94.9)||0.60‡|
|Days to viral load <1000 copies/ml, median (range; IQR)c||13.5 (3.5–39.0; 7.0–19.3)||15.8 (3.0–121.5; 10.3–24)||0.03§|
|Gestational age at viral load <1000 copies/ml (weeks), mean (SD)||26.6 (6.17)||25.1 (5.77)||0.15†|
|HIV viral load <1000 copies/ml at time of delivery, n (%)d||63 (95.5)||51 (87.9)||0.19‡|
Twelve (9%) of the 138 women had a baseline HIV viral load < 1000 copies/ml before initiation of HAART during pregnancy and were, therefore, excluded from the analysis of time to reach viral load < 1000 copies/ml. Of the 126 women with baseline viral load > 1000 copies/ml before HAART initiation, 96% of ARV-naive women reached a viral load <1000 copies/ml compared with 95% of ARV-experienced women (Table 2). The median number of days to achieve viral load of <1000 copies/ml in the entire cohort was 14 (range 3.0–121.5; IQR 8.5–21). When assessing based on ARV-exposure history, the median number of days to achieve viral load < 1000 copies/ml in the ARV-naive group was 13.5 (range 3.5–39.0; IQR 7–19.3) days compared with 15.8 (range 3.0–121.5; IQR 10.3–24) days in the ARV-experienced group (P = 0.03). Mean gestational age at which the women achieved viral load < 1000 copies/ml was 26.6 weeks for ARV-naive group and 25.1 weeks for ARV-experienced group (P = 0.15). There was no statistically significant difference between ARV-naive and ARV-experienced groups with respect to number of women with viral loads < 1000 copies/ml at each visit and number of days between interval viral load assessments.
Figure 1A displays the non-parametric maximum likelihood estimator curves (Kaplan–Meier–Turnbull) and fitted accelerated failure time survival curves comparing days to achieve viral load <400 copies/ml for ARV-naive and ARV-experienced pregnant women. Figure 2A displays the corresponding survival curves comparing days to achieve viral load <1000 copies/ml. The non-parametric tests demonstrated a statistically significant reduction in time to suppress viral load <1000 copies/ml in ARV-naive relative to ARV-experienced pregnant women (P = 0.038), but not at <400 copies/ml (P = 0.172). These results were similar to the corresponding accelerated failure time model results (P = 0.045 and P = 0.176, respectively) with only ARV exposure history as a predictor, and the accelerated failure time model fitted curves provide a reasonably smooth approximation to the raw Kaplan–Meier–Turnbull curves. Figure 2A,B show the survival curves for time to reach viral load < 400 copies/ml and <1000 copies/ml, respectively, adjusted for all covariates in the final multiple predictor model. For these plots, the other final model predictors were fixed at reference levels/values of UCSF/SFGH for site, PI-based regimen, nulliparous for birth, and the mean values across all women for the continuous predictors of adherence, HIV log10 viral load, and CD4 cell count. Based on the corresponding final multiple predictor models neither the adjusted <400 copies/ml (P = 0.29) nor <1000 copies/ml (P = 0.21) survival curves had statistically significant differences with respect to ARV exposure history. For group predictors, the group which was most common was used as the reference.
In the single predictor survival models, baseline HIV log10 viral load, baseline CD4 cell count, adherence, and ARV regimen class used during index pregnancy were statistically significantly associated with time to achieve viral load < 400 copies/ml (Table 3). In the final multiple predictor model, baseline HIV log10 viral load at time of HAART initiation or restart during pregnancy was associated with a longer time to achieve viral load <400 copies/ml, whereas better adherence and receiving an NNRTI-based regimen were statistically significantly associated with shorter time to achieve viral load below level of detection. When adjusting for the specified covariates in the final model, each HIV log10 copies/ml increase in baseline viral load resulted in a 40% reduction (95% confidence interval [95% CI] 8–61%, P = 0.02) in chance of achieving a viral load < 400 copies/ml at any particular time (referred to as the relative decrease in hazard). Subjects with higher adherence had a 1.29-fold (95% CI 1.08–1.54, P = 0.01) relative increase in hazard for each 10% increase in adherence. Additionally, those women receiving NNRTI-based regimen compared with those receiving a PI-based regimen had 2.48-fold increase in hazard (95% CI 1.19–5.60, P = 0.01). The ARV exposure history was not statistically significant in the multiple predictor model with an estimated 1.31-fold increase in hazard (95% CI 0.79–2.15, P = 0.29).
|Variable||Single predictor||Multiple predictor|
|Unadjusted relative hazard (95% CI)||P-value||Adjusted relative hazard (95% CI)||P-value|
|Viral load <400 copies/mla (n = 131)|
|ARV exposure history|
|Naive||1.32 (0.88–1.98)||0.18||1.31 (0.79–2.15)||0.29|
|HIV viral loadb (log10 copies/ml)||0.55 (0.36–0.84)||0.01||0.60 (0.39–0.92)||0.02|
|CD4 cell countb (cells/μl) per 50 unit increase||1.07 (1.01–1.14)||0.03||1.04 (0.99–1.09)||0.15|
|Gestational ageb (weeks)||1.01 (0.99–1.04)||0.37||–|
|Adherence per 10% increase||1.27 (1.09–1.48)||0.01||1.29 (1.08–1.54)||0.01|
|UCSF/SFGH||1.30 (0.88–1.93)||0.19||1.15 (0.73–1.80)||0.54|
|NNRTI-based||2.21 (1.43–3.41)||<0.001||2.48 (1.33–4.63)||0.01|
|Otherd||2.71 (1.15–6.41)||0.02||2.60 (1.16–5.83)||0.02|
|Viral load <1000 copies/mla (n = 126)|
|ARV exposure history|
|Naive||1.52 (1.01–2.30)||0.05||1.38 (0.83–2.29)||0.21|
|HIV viral loadb (log10 copies/ml)||0.53 (0.33–0.88)||0.01||0.54 (0.34–0.86)||0.01|
|CD4 cell countb (cells/μl) per 50 unit increase||1.12 (1.06–1.19)||<0.001||1.12 (1.03–1.22)||0.01|
|Gestational ageb (weeks)||1.02 (0.99–1.05)||0.20||–|
|Adherence per 10% increase||1.54 (1.23–1.94)||<0.001||1.79 (1.34–2.39)||<0.001|
|UCSF/SFGH||1.52 (0.97–2.38)||0.07||2.04 (1.19–3.53)||0.01|
|NNRTI-based||2.27 (1.32–3.91)||0.01||2.95 (1.23–7.06)||0.02|
|Otherd||1.72 (0.72–4.11)||0.23||1.41 (0.53–3.77)||0.49|
In the single predictor survival models, ARV exposure history, baseline HIV log10 viral load, baseline CD4 cell count, adherence and ARV regimen class used during index pregnancy were statistically significantly associated with time to achieve HIV viral load <1000 copies/ml (Table 3). In the final multiple predictor model, baseline HIV log10 viral load at time of HAART initiation or restart during pregnancy was associated with a longer time to achieve a viral load <1000 copies/ml, whereas higher baseline CD4 cell count, better adherence, women at the UCSF/SFGH study site, and receiving NNRTI-based regimen were associated with shorter time to achieve a viral load <1000 copies/ml. When adjusting for the specified covariates in the final model, each HIV log10 copies/ml increase in baseline viral load resulted in a 46% decrease in hazard (95% CI 14–66%, P = 0.01). Subjects had an estimated 1.12-fold increase in hazard for each 50-count increase in CD4 cell count (95% CI 3–22%, P = 0.01) and for each 10% increase in adherence had an associated 1.79-fold increase in hazard (95% CI 1.34–2.39, P <0.001). Subjects at the UCSF/SFGH site had 2.04 times the hazard of women at the Baylor site (95% CI 1.19–3.53, P = 0.01). Additionally, those women receiving an NNRTI-based regimen compared with those receiving a PI-based regimen had 2.95 times higher hazard (95% CI 1.23–7.06, P = 0.02). The ARV exposure history was not statistically significant in the multiple predictor model with an estimated 1.38-fold increase in hazard (95% CI 0.83–2.29, P = 0.21).
We found that the majority of both ARV-naive and ARV-experienced HIV-infected women achieved a viral load < 400 copies/ml by the time of delivery. Furthermore, for the entire cohort, it took a median of 3.7 weeks (26 days) to reach HIV viral load of <400 copies/ml and 2 weeks (14 days) to reach a viral load of <1000 copies/ml. Nonetheless, 25% of women required longer than 5.9 weeks (41 days) to reach a viral load <400 copies/ml and 3 weeks (21 days) to reach a viral load <1000 copies/ml after initiation or restart of HAART during pregnancy. In addition, our data, indicated a pattern that ARV-naive pregnant women may be more likely to achieve viral loads <400 copies/ml and <1000 copies/ml than those who are ARV-experienced, although this was only statistically significant for a group comparison without covariates for the <1000 copies/ml level. The associations we found between viral suppression and baseline HIV viral load, baseline CD4 cell count, as well as medication adherence were expected and have been noted in a few previous studies that included pregnant women.[16, 17, 19, 20] Our findings also demonstrated that receiving an NNRTI-based regimen was associated with a shorter time to achieve viral load <400 copies/ml and <1000 copies/ml. This is similar to the European Collaborative Study and Bell et al.,[16, 18] who have previously reported that women receiving NNRTI-based versus PI-based regimens had faster time to undetectable viral load.
A similar proportion of the women in our cohort reached viral load below the level of detection of 400 copies/ml during pregnancy compared with other studies.[15-17, 19] Conversely, our findings indicated that the median time to reach a viral load below level of detection was shorter than what has been previously reported in some, but not all, studies.[15-17, 19] However, our findings require further confirmation, because the previous studies may have differed significantly in design, methodology and population characteristics, and so a direct comparison is limited. Our study also evaluated median time to achieve viral load <1000 copies/ml based on the current standard of care to perform caesarean delivery for women with viral load >1000 copies/ml at term gestation, as examined previously by Wright et al. The median number of days to reach a viral load <1000 copies/ml among women in our study was estimated at less than half of that previously presented by others.
Our investigation is one of the few published reports that compared ARV-naive and ARV-experienced pregnant women.[17, 20, 21] The results indicate a possible difference in time to achieve specified viral load threshold based on simple group comparisons of ARV-naive versus ARV-experienced women for <400 copies/ml (not statistically significant) and <1000 copies/ml (statistically significant). This is different from findings of other investigators who reported that ARV exposure history was not associated with HIV virological response in pregnant women.[20, 21] Additionally, while our study estimated only small differences in median time to viral suppression between ARV-naive and ARV-experienced women, Schalkwyk et al. who also performed this comparison found that it took, on average, 20 days longer for ARV-experienced women to achieve viral suppression. One may hypothesise that, given the frequent use of HIV genotype resistance assays in our cohort for the majority of ARV-experienced women (82%), they were prescribed fully suppressive regimens with at least three active antiretroviral medications and, therefore, would be expected to experience excellent virological outcomes. However, there may still be either inherent physiological or social factors that contribute to this observed difference between ARV-naive versus ARV-experienced pregnant women.
Our study had several limitations inherent in its retrospective design that warrant mention. Our statistical power to fully examine the potential differences between ARV-naive and ARV-experienced women may have been limited by incorrect estimation of relative hazards, as the observed relative hazards were not as large as those estimates we used for the power analysis. As baseline viral load measurements may have been performed up to 4 months before initiation of HAART and subsequent viral load measurements were performed thereafter at different time intervals (generally ranging from a median of 2 to nearly 6 weeks apart), our results may have underestimated the rapidity of viral decay. However, although intervals in which viral load was measured were not consistent for all women, there were no statistically significant differences in the interval days between viral load measurements between the two groups at each visit. Additionally, laboratory assays, and therefore level of detection, varied based on institution and year of blood draw. These various factors could potentially influence the reliability of our calculation of time to viral load level of <400 copies/ml or <1000 copies/ml, especially given the different assay thresholds used in various study time periods.
For the purposes of summary estimates, we used an interval mid-point to calculate time to viral suppression (though this was not used in our statistical modelling, where the interval censoring was appropriately modelled). We know from previous studies that viral decay is biphasic and, therefore, not perfectly modelled using this mid-point estimation.[12-14] Therefore, the summary estimates in our study may have underestimated or overestimated the time to reach our specified viral load outcome levels. Nonetheless, the estimates may still be able to guide management decisions pertaining to average time needed to achieve critical viral load levels for obstetric management purposes, as our findings contribute to an overall greater understanding of this issue.
Furthermore, there were several potential sources of bias as a result of patient population and practice management differences at different study sites and over the duration of the study period. For example, the number of pregnancies differed over the study time periods at the different study sites, which may have influenced our findings. Additionally, there was a change in PI versus NNRTI-based HAART regimen use patterns over time, where the majority of NNRTI-based regimen use was conducted at the UCSF-SFGH hospital site and occurred predominantly from 2000 to 2006, which may have also contributed to bias. Nonetheless, we did find that NNRTI-based regimens demonstrated a significantly faster time to achieving viral loads <400 copies/ml and <1000 copies/ml.
Different ARV regimen types and dosing used among the different sites and during different periods, as well as potentially different pharmacokinetic properties in pregnancy, may have also influenced the time required to achieve viral load below a detectable level. Hence, multiple predictor survival models were used to control for ARV regimen types. Yet despite controlling for ARV regimen type, there may have been additional bias because of differences in boosted versus non-boosted (nelfinavir) regimens, which differed in ARV-experienced versus ARV-naive women. However, the majority of women in both groups received boosted PI regimens in our study, and therefore this potential confounder was probably limited when comparing these two groups.
Finally, given the prolonged half-life of NNRTIs and possible virological implications, use of at least 6 months between last use of NNRTIs and reinitiation of HAART for ARV-experienced women may have been the preferred exclusion criterion as demonstrated by Lockman et al. However, when designing our study, such specific data were not available to guide in the criteria for establishment of the optimum time needed for NNRTI washout, and therefore a minimum of 2 months was chosen as acceptable. Furthermore, the majority of our women (66%) had >6 months between last use of NNRTI regimen and reinitiation of HAART in the index pregnancy. Additionally, of the 27 ARV-experienced women with a history of NNRTI use, 21 (78%) were reinitiated on a PI-based regimen during the index pregnancy but six (22%) received an NNRTI-based regimen. Hence, the potential influence due to insufficient NNRTI washout on our findings was probably small overall in our cohort.
Despite these limitations, the inclusion of both ARV-naive and ARV-experienced women and the ethnic heterogeneity of our cohort may broaden the clinical generalisability of our data compared with other studies. Although previous data have reported shorter time to achieving viral suppression in pregnant women of Western African origin, our results may be considered with respect to a more ethnically diverse pregnant patient cohort than we often encounter in our routine clinical management. Our data support previous findings that adherence is the key, potentially modifiable, variable associated with time to viral load below level of detection.[17, 20] Our group and others have used creative approaches to optimising adherence, including the use of directly observed therapy on labour and delivery units.[27-31] We also found that use of an NNRTI-based regimen may be associated with increased likelihood of achieving a viral load < 400 copies/ml and <1000 copies/ml. Given updated Centers for Disease Control and Prevention guidelines recommending third-trimester HIV antibody screening in many settings across the USA, there may be an increasing proportion of HIV-infected pregnant women identified later in pregnancy. Our data also seem to suggest that for women starting ARVs later in pregnancy an NNRTI-based HAART regimen may be considered to potentially result in a faster time to achieve viral loads <400 copies/ml and <1000 copies/ml. By more quickly lowering the viral load to <400 copies/ml and <1000 copies/ml, the risk of antepartum HIV transmission would be lower and, simultaneously, these women could more likely to avoid the need for a caesarean delivery to prevent intrapartum HIV transmission and associated risks.[33-35] Given the risk of hepatotoxicity with the use of nevirapine among women with CD4 cell counts >250 cells/μl, HIV-infected women who are newly diagnosed late in pregnancy or who present for care late in pregnancy would likely benefit most from the use of efavirenz, now considered safe after the first trimester.[9, 10] Although randomised controlled trials would be able to assess different ARV regimen viral decay properties during pregnancy to further confirm our findings, such study designs may be difficult to conduct. These trials face several challenges, including ethical, study design and cost-efficacy considerations. Furthermore, additional prospective, well-designed observational cohort studies conducted with comprehensive assessment of potential confounders and using uniform viral load assays, level of detection, and viral load measurement frequency would be useful in determining viral load decay in pregnant women for corroboration of our results and assessment of specific characteristics, which may influence time to achieve viral load suppression during pregnancy. Moreover, these data also reinforce the importance of adherence to ARVs during pregnancy. Lastly, the near universal achievement of a viral load <1000 copies/ml among the women in our cohort suggests that, with aggressive assessment of HIV viral load and alteration of ARV therapy when necessary, few HIV-infected women would need a caesarean delivery solely as an HIV prevention intervention. Even when HIV-infected women present for prenatal care later in the pregnancy, our findings demonstrate that immediate initiation of HAART may still allow in most women for viral load to decrease within a period of weeks to levels that would allow for a vaginal trial of labour.
Our data may help to inform the timing of HAART initiation in pregnancy and choice of antiretroviral regimen for women initiated on HAART later in pregnancy. Given the increased risks of caesarean delivery among HIV-infected women,33–35 reserving caesarean delivery for strictly obstetric indications would be prudent. Indeed, the most effective means of preventing MTCT of HIV, the use of HAART to achieve viral suppression, is also the most important clinical tool we have to promote the long-term health of HIV-infected women.
No financial or personal conflict of interest disclosures exist for any of the authors.
NA was principal investigator, principal author and contributed to the statistical analysis; AS and JK were co-authors and contributed to the statistical analysis, AS was also co-investigator. NL, MM, JL, CF and MS were co-authors and co-investigators and DC was a principal investigator and co-author.
This study was approved by the investigational review boards (IRB) at all three participating institutions: University of California, San Francisco (UCSF) and San Francisco General Hospital Bay Area Perinatal AIDS Center (SFGH BAPAC) from August 1997 to April 2009 (IRB# H11529-28926); and the Baylor College of Medicine from April 2006 to April 2009 (IRB# H18412).
This research was supported in part by the National Institutes of Health Loan Repayment Program Award for Clinical Research, 2005–2007 and 2007–2009. It was also supported by the National Center for Advancing Translational Sciences, National Institutes of Health, through UCSF-CTSI Grant Number UL1 TR000004. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the National Institutes of Health.
We would like to thank Chengshi Jin in the Division of Biostatistics at the University of California at San Francisco for her valuable contributions to the statistical analysis.