Global rates of mother-to-child HIV-1 transmission are heterogenous; estimates range from 14 to 48% . In non-breastfeeding populations the risk of perinatal HIV transmission is most commonly 15-25% [1-3]. The variation in observed rates may be due to differences in risk factors or to viral characteristics in specific populations, and also to differences in timing of transmission. Although many factors have been associated with an overall increased risk of perinatal HIV transmission including increased maternal viral load, low CD4+ T-lymphocyte counts, decreased maternal cell-mediated immunity, vaginal delivery, prolonged rupture of membranes, prematurity, low birth-weight, and prolonged labor [1,4], risk factors according to time of transmission have not been well characterized.
In the absence of breastfeeding, perinatal HIV transmission can occur either in utero or intrapartum. It has been proposed that among non-breastfed infants, in utero and intrapartum transmission can be distinguished based on early infant HIV-1 DNA PCR or viral culture results . According to this widely used hypothesis, a positive virological result from a blood sample obtained within 48 h of birth indicates presumptive in utero transmission, whereas an initial negative result (within 1 week of birth) followed by a positive result indicates intrapartum transmission. With some modifications of the time-frame for positive results, this definition has been adopted by several investigators [6-12] who have estimated the relative amount of intrapartum transmission to be in the range 62-75% [6-9,11,13]. Determination of the relative contributions of presumed in utero and intrapartum transmission, and an understanding of risk factors specific for time of transmission, is important for the design and interpretation of interventions to reduce perinatal transmission, especially strategies targeting late pregnancy.
Although the mechanisms of in utero and intrapartum transmission are not well understood, it has been suggested that risk factors for in utero transmission probably include maternal factors such as high viral load, low CD4+ T-lymphocyte count and impaired cell- mediated immunity . Maternal viral load, by direct contact with blood or secretions, and immunological factors might also be associated with intrapartum transmission. Theoretically, obstetric factors should be associated with intrapartum transmission and one study found duration of ruptured membranes to be of significance . However, maternal viral load, which may be the most important determinant of transmission, has not been investigated in relation to timing of transmission.
We have separately reported the overall findings on transmission rate and risk factors in a non-breastfeeding perinatal cohort with HIV-1 subtype E infection [14,15]. Maternal viral load was the strongest risk factor for transmission, and gestational age, mode of delivery and maternal natural killer (NK) cell percentage were independent predictors of transmission also. In the current analysis, we estimate the relative and absolute rates of in utero and intrapartum transmission, assess whether maternal and delivery risk factors for perinatal transmission are related to time of transmission, and estimate the population-attributable fraction for maternal viral load, by time of transmission.
The Bangkok Collaborative Perinatal HIV Transmission Study recruited HIV-infected pregnant women (first and second 2nd trimesters) from two large Bangkok hospitals during 1992-1994 to participate in a prospective cohort study . Specific counseling was given to avoid breastfeeding, in accord with Thai national guidelines for HIV-positive women. A detailed description of the study methodology and the main findings of the overall transmission rate and risk factors have been reported previously [14,15]. The study was approved by the Ethical Review of Research Committee, Ministry of Public Health, Bangkok, Thailand and by the Institutional Review Board, CDC, Atlanta, USA.
Maternal enrollment sera and sequential child plasma samples were tested to confirm HIV antibody status with the Genetic Systems HIV-1/HIV-2 EIA (Genetic Systems, Redwood, WA, USA) and the Novapath HIV-1 Immunoblot (BioRad Laboratories, Hercules, CA, USA).
Qualitative HIV-1 DNA PCR was performed on all available birth, 2-month and 6-month infant venous blood samples. An aliquot of these samples, collected in EDTA anticoagulant, was prepared in Bangkok as a leukocyte DNA lysate and frozen in liquid nitrogen. Diagnostic PCR was performed in duplicate on batched, coded samples and negative and positive controls using two sets of gag primer pairs, SK38/39 and SK145/150. This CDC PCR assay is estimated to detect a minimum of 10 proviral copies/25μl of DNA lysate, representing approximately 150000 leukocytes.
Quantitative HIV-1 RNA PCR was performed on coded maternal delivery plasma samples using the Amplicor HIV-1 Monitor Test Kit, version 1.0 (Roche Diagnostic Systems, Inc., Branchburg, NJ, USA) according to the manufacturer‚s instructions with some modifications. A new primer mix (SK145 and SK151), provided by Roche, was added (10μl) to efficiently quantify RNA of HIV-1 subtype E. This modification gives similar results to those obtained with the new primer set included in version 1.5. The starting specimen volume was 200μl. With this volume the assay has a threshold sensitivity of 200 HIV RNA copies/ml and a linear range for quantification of 400-750000 copies/ml.
Lymphocyte phenotyping was performed on fresh mother and infant venous blood samples collected in EDTA collection tubes using the FACScan flow cytometer (Becton Dickinson Immunocytometry Systems, San Jose, CA, USA) and a standard six-tube, two-color monoclonal antibody panel (Becton Dickinson). NK cells were defined as CD3-, CD16+ and CD56+.
Definition of infant HIV-1 infection
Infants were considered to be HIV-1-infected if they had two HIV-1-positive DNA PCR tests, or one positive PCR test and a CDC AIDS-defining condition. Infants were defined as uninfected if they tested PCR HIV-1-negative on two samples, including one obtained at 6 months of age or older, or if they seroreverted to HIV-negative status on EIA testing. Infants not meeting these criteria were considered to have unknown infection status.
Study sample and definition of time of transmission
Mother-infant pairs with known transmission outcomes were included in this analysis if the infant had a PCR test result on a sample obtained within 72 h of birth (birth sample). The time of birth and time of specimen collection were both recorded, and the time (h) after birth was calculated. Among infected infants, in utero infection was presumed if the birth PCR test result was positive. Otherwise, infection was presumed to have occurred intrapartum.
To investigate factors associated with timing of perinatal HIV transmission two comparisons were made: (i) in utero-infected infants (positive birth PCR) versus all infants with a negative birth PCR test result (uninfected infants and intrapartum-infected infants); and (ii) intrapartum-infected infants versus uninfected infants. Exact binomial 95% confidence intervals (CI) were calculated for the estimated HIV transmission rates. Continuous variables, including maternal RNA level, CD4+ T-lymphocyte and NK cell absolute values and percentages, duration of ruptured membranes and duration of labor were dichotomized at the overall sample median or at values reported by others. One woman had a viral level below detection; she was assigned a value of 100 copies/ml for statistical analysis. Frequency data were analyzed by χ2 (or Fisher‚s exact) or χ2 for trend tests. Odds ratios (OR) and mid-P 95% CI were computed as the measure of effect . Multivariate analyses were performed using multiple logistic regression, adjusted OR were estimated and 95% CI were computed.
To test whether a risk factor was related to time of transmission, a nested response logistic regression model was used . The model as applied here is a discrete time survival model ; the grouped time points of infection were in utero and intrapartum. Constancy of effect, or homogeneity of OR, was assessed by testing the significance of an interaction term between the risk factor and time of transmission in the logistic model. The population-attributable fraction was estimated for maternal viral load at different reference cut-off points by logistic regression, adjusted for other significant risk factors . CI were calculated on the logit scale .
Of 295 HIV-infected pregnant women who delivered during the study, 281 (95.3%) had infants with known HIV-infection status. There was a total of 68 infected infants and the overall transmission rate was 24.2%. Sixty-one of the infected infants (90%) had two or more positive PCR tests; the other seven infected infants met the alternate definition for infection.
Of the 281 mother-infant pairs with known transmission outcome, 218 (78%), including 49 (72%) of the 68 infected infants, had a PCR test result for a specimen collected within 72 h of birth, and were included in this analysis. There were no significant differences (data not shown) in mother‚s age or any maternal, delivery, or other perinatal factors between the 218 infants with a birth PCR result and the 63 additional infants with known infection outcomes but without a birth PCR result. There were also no significant differences in factors among infected children included in or excluded from the birth PCR analysis. The median time of collection of the birth sample was 22 h; 82% of samples were collected within 48 h of birth.
For the 218 infants with a birth PCR test result, the overall HIV transmission rate was 22.5% (Table 1). Twelve infected infants tested HIV-1 positive by PCR at birth, giving an in utero transmission rate of 5.5%. Thirty-seven infected infants tested HIV-1 negative at birth and subsequently tested HIV-1 positive, giving an intrapartum transmission rate of 18.0%. PCR testing at birth detected 12 out of 49 infected infants, suggesting that 24.5% (95% CI, 14.0-37.9) of transmissions occurred in utero and 75.5% occurred intrapartum. Similar results were obtained when the analysis was limited to samples collected within 48 h of birth.
High maternal viral load at delivery and low birth-weight (<2500g) were independently associated with in utero transmission (Table 2). Maternal viral load above the sample median (18500 copies/ml), which included 10 (83%) out of 12 in utero infections, was associated with a 5.8-fold (95% CI, 1.4-38.8) increased adjusted odds of transmission. When analyzed as a continuous variable, the adjusted odds of transmission was 2.9 (95% CI, 1.2-7.8) per unit log increase in maternal viral load. Low birth-weight also was associated with in utero transmission [adjusted OR (AOR), 5.6; 95% CI, 1.3-21.1]: one-third of infants infected in utero were low birth-weight. Pre-term infants (<37 weeks gestational age, by modified Ballard) had a 25% risk for in utero transmission compared to 5% for full-term infants (OR, 6.7; 95% CI, 0.8-35.6), although this was not statistically significant. Maternal immunological markers, such as CD4+ T-lymphocyte and NK cell percentage and absolute count (not shown), and obstetric factors such as mode of delivery and duration of membrane rupture and labor, were not associated with in utero transmission. However, because the number of cases was small, this analysis had limited power to identify factors associated with in utero transmission.
Maternal delivery viral load was also strongly associated with intrapartum transmission (Table 3), whether dichotomized at the sample median (AOR, 4.4) or analyzed as a continuous variable (AOR, 2.31/log increase). In addition, low CD4+ T-lymphocyte and low NK cell percentages were associated with increased intrapartum transmission (the absolute counts for these variables were not associated with transmission). Delivery by cesarean section appeared to be protective for intrapartum transmission (11.1% transmission versus 19.0% for vaginal delivery), but the number of cesarean births with a PCR result was small (n = 27) and this association was not significant. There was no suggestion that other obstetric factors, or prematurity and low birth-weight, were associated with intrapartum transmission. Lack of association for duration of ruptured membranes and length of labor was still observed when cesarean deliveries were excluded from the analysis.
To evaluate whether factors found to be associated with in utero or intrapartum transmission were dependent on time of transmission, a nested logistic model was fitted to the data and was tested for interaction between the factor and time of transmission (not shown). Maternal viral load was not associated with time of transmission (P=0.99 for no interaction), as elevated viral load was associated significantly with increased risk for both in utero and intrapartum transmission. In addition, viral load among mothers who transmitted in utero (mean, 4.7 log10) was comparable to viral load among mothers who transmitted intrapartum (mean, 4.6 log10; P=0.6). The effects of NK cell percentage were not related to time of transmission (P=0.9); they were in the same direction and magnitude for both in utero and intrapartum transmission, but associated significantly with intrapartum transmission only (Table 3). The non-significant association with in utero transmission was probably due to low power. In contrast, the effects of low birth-weight (P=0.03) and maternal CD4+ T-lymphocyte percentage (P=0.02) were associated with time of transmission. Low birth-weight was significantly associated with in utero transmission (Table 2), whereas its effect for intrapartum transmission showed a trend in the opposite direction (Table 3). Conversely, CD4+ T-lymphocyte percentage was associated significantly with intrapartum transmission; its effect for in utero transmission also showed a trend in the opposite direction. For the two risk factors related to time of transmission, infants infected in utero were compared directly with infants infected intrapartum. Infants infected in utero were more likely to be of low birth-weight (OR, 8.8; 95% CI, 1.3-74.8) and infants infected intrapartum were more likely to have mothers with a CD4+ T-lymphocyte percentage <21% (OR, 4.9; 95% CI, 1.1-25.5).
Because maternal viral load was associated with both in utero and intrapartum transmission and is amenable to direct intervention, the population-attributable fractions for different viral load cut-off points were estimated (Table 4). Adjusted for low birth-weight, the population-attributable fraction for in utero transmission was 67% for maternal viral levels greater than the sample median of 18500 copies/ml (4.27 log10). Because of the small number of in utero infections, more extensive analysis of different cut-off points was not possible. For intrapartum transmission, adjusted for NK and CD4+ T-lymphocyte percentages, the population-attributable fraction was 53% for maternal viral levels greater than the median, and 69% for levels greater than 10000 copies/ml (4log10). Table 4 also shows the level-specific estimates for half-log increments in viral load from 4log10 to greater than 5log10.
Using the current working definition of timing of perinatal transmission in a cohort of non-breastfed infants born to HIV-1 subtype E-infected mothers in Bangkok, this study confirms that approximately three-quarters of HIV-infected infants are infected intrapartum and has found that maternal plasma viral load at delivery is a strong, independent risk factor for both in utero and intrapartum transmission. This is important because maternal viral load should be amenable to direct intervention.
To our knowledge, this is the first analysis that has evaluated the relationship between maternal viral load and timing of perinatal HIV transmission. Clear evidence was found that elevated maternal viral load is a strong risk factor for both in utero and intrapartum transmission. Whereas in utero transmission is probably associated with cell-free or cell-associated virus crossing the placenta, increased intrapartum risk may be due to maternal-infant transfusions, disruptions of the placental barrier or direct contact with maternal blood or mucosal secretions in the vaginal canal. The finding that viral load, as a determinant of transmission, is independent of timing was anticipated in a recent review , but this study provides the first evidence to support this hypothesis. This finding suggests that intervention strategies aimed at reducing maternal viral load might be successful at reducing both in utero and intrapartum transmission.
The population-attributable fraction is an estimate of the reduction in transmission that might be expected if a risk factor could be eliminated. For in utero transmission, the population-attributable fraction associated with viral load greater than the sample median was 67%. Because of the small number of cases, it was not possible to explore further the relationship between viral load and in utero transmission. Further investigations and combined analysis may help to define this relationship better. For intrapartum transmission, the population-attributable fraction estimate was 53% for viral load greater than the median and 69% for levels greater than 10000 copies/ml. Below 10000 copies/ml, intrapartum transmission risk appeared to be low (<5%). In this population sample, viral load reductions to below 10000 copies/ml at delivery would be achieved for most women with an intervention that produces a 0.5-1 log10 reduction. Our data did not suggest that the magnitude of effect obtained by reducing viral load may be more important for in utero transmission, but this should be examined in other studies. Short-course antiretroviral intervention studies targeted at preventing late in utero and intrapartum perinatal HIV transmission currently in progress in developing countries, including a study recently concluded by our group which showed a 50% reduction in transmission [22,23], may help to confirm the role of maternal viral load and timing of transmission.
Our estimate that 25% of infected infants are infected in utero is the same as estimates reported in CDC studies in Zaire and New York [6,7] and is consistent with estimates of approximately 35% reported in other studies [8,9,11]. Differences may be due to different PCR methods, the completeness of the sample, and the time of the ‚birth sample‚ collection. Using birth DNA PCR results (within 2-3 days of birth), it is difficult to distinguish late in utero transmission from intrapartum transmission because the window before detecting HIV DNA in the blood may vary and may be as long as 1 week [6,9,11]. Additional analyses with more sensitive, standardized assays such as RNA PCR, may help to distinguish better between in utero and intrapartum transmission. Nevertheless, the consistent conclusion from studies worldwide, conducted in populations with different HIV-1 subtypes, is that in the absence of breastfeeding most perinatal HIV infections (65-75%) occur during labor and delivery. This suggests that most perinatal transmission, in the absence of breastfeeding, may be amenable to short-term interventions targeted at the end of pregnancy.
In this analysis, the overall mother-to-child HIV transmission rate was 22.5% (95% CI, 17.3-28.4), a result consistent with other transmission estimates in women who did not breastfeed and who did not receive antiretroviral therapy [2,3]. An absolute in utero transmission rate of 5.5% (95% CI, 3.0-9.2) was estimated; this is again consistent with previous estimates from New York  and Africa [6,10]. In addition, the absolute intrapartum transmission rate was estimated to be 18% (95% CI, 13.2-23.7), almost identical to rates reported in Africa, with breastfeeding [6,10], but somewhat higher than the 13% rate reported in New York . Intrapartum transmission may be more variable than in utero transmission.
We recently reported that low maternal NK cell percentage was a risk factor for HIV-1 transmission . In the current analysis, it was found that the NK cell association was independent of time of transmission. NK cells are a marker for cell-mediated immunity, and NK cell number declines in HIV-infected individuals . Decreased maternal cell-mediated immunity has been associated with perinatal transmission and has been hypothesized to be a risk factor for in utero transmission . The present NK cell results suggest that maternal cell-mediated immunity can influence transmission directly, independently of viral load , particularly at delivery. Functional studies might help to clarify the relationship between NK cell number and activity. If these NK cell findings are confirmed in other studies, then NK cell percentage might be used as a marker to target women at high risk for transmission. Our finding that lower CD4+ T-lymphocyte percentage at delivery was associated with increased risk of intrapartum transmission further supports the hypothesis that maternal immunosuppression at delivery is a risk factor for intrapartum transmission, but the mechanism and the opportunities for intervention are not clear.
Low birth-weight and prematurity have been associated with perinatal HIV transmission in numerous studies, including ours. Although these factors are hypothesized to be associated with in utero infection, there have been no data to support this, and debate continues over whether low birth-weight and prematurity result from or predispose to infection [1,25]. In one African study, there was no relationship between birth-weight and timing . In New York, low birth-weight and prematurity were associated with intrapartum transmission . In the current analysis, low birth-weight was associated with in utero transmission, as was prematurity, although the association with prematurity was borderline. Contrary to the New York study, low birth-weight and prematurity were not associated with intrapartum transmission. Some of the difference may be due to differences in the causes of low birth-weight and prematurity in the two populations and to differences in the occurrence of these conditions. For example, in the Bangkok study, there was virtually no smoking or drug use, antenatal registration was early and regular, and women received multivitamin supplements. Interestingly, in these two cohorts of HIV-infected women, the prevalence of low birth-weight infants in Bangkok was 10%, whereas in New York it was 28% . Exclusion of low birth-weight from the present analyses, because of concern about whether it is a cause or consequence of HIV infection, does not change the other conclusions.
It was expected that obstetric factors would be found to be associated with intrapartum transmission [1,4]. However, neither mode of delivery, duration of ruptured membranes nor length of labor were independent risk factors for intrapartum transmission. Longer duration of ruptured membranes among vaginal deliveries was associated with intrapartum transmission in the New York study . One explanation for the difference in findings might be that women in the present study had shorter duration of ruptured membranes and few had prolonged ruptured membranes. In the complete cohort, cesarean section was protective for overall mother-to-child transmission . However in the reduced sample, the protective effect of cesarean section for intrapartum transmission was not significant.
In summary, this study demonstrates that maternal viral load, as measured at delivery, is strongly associated with both in utero and intrapartum HIV-1 perinatal transmission, and suggests that interventions aimed at reducing maternal viral load may be effective in reducing transmission at both of these times. The results reported here provide further evidence that as much as 75% of perinatal HIV-1 infections occur during labor and delivery, and suggest that a successful intervention offered late in pregnancy could reduce substantially both intrapartum and overall transmission risk. These findings may help to explain the success, in the absence of breastfeeding, of short-course zidovudine given in late pregnancy and delivery to reduce perinatal transmission .
The authors thank R.J. Simonds and J. Karon for critical review of the manuscript, V. Batter and L. Carr for assistance with data management and analysis, J. Rapier for DNA PCR testing, T. Granade and S. Philips for RNA PCR testing, and W. Pokapanichwong, J. Laosakkitiboran and P. Yuentrakul, for coordinating the field study. We gratefully acknowledge the dedicated field work of the study nurses and social workers: K. Neeyapun, P. Tothong and S. Pinyovanichkul (team leaders); N. Chookaew, S. Henchaichon, S. Jalanchavanapate, B. Jetsawang, K. Klumthanom, S. Phurksakusamesuk, C. Prasert, S. Samsukkree, S. Sorapipatana, S. Suwanmaitre and C. Yuvasevee.