Menu
Menu
Menu
Menu
Human immunodeficiency virus (HIV) is common in pregnant women in many malaria-endemic regions and may increase risk of placental parasitemia. Placental malaria is more common in primigravidae than multigravidae, but the relationship between HIV and malaria across gravidities is not well characterized. We recruited pregnant Malawian women during the second trimester and followed them until delivery. Parasitemia was assessed at enrollment, follow-up visits, and delivery, when placental blood was sampled. There was no difference in risk of parasitemia between HIV-positive and HIV-negative primigravidae. Among multigravidae, HIV-infected women had greater than twice the risk of parasitemia as HIV-uninfected women throughout follow-up. Human immunodeficiency virus was also associated with more frequent peripheral parasitemia in multigravidae but not primigravidae. Both HIV and primigravid status were independently associated with higher peripheral and placental parasite densities. Although risk of parasitemia is lower in multigravidae than primigravidae, the HIV effect on risk of malaria is more pronounced in multigravidae. Copyright © 2012 by The American Society of Tropical Medicine and Hygiene.
The study population consisted of all healthy women in their second trimester of pregnancy attending the Mpemba and Madziabango Health Centers in Blantyre District in southern Malawi for antenatal care and delivery between March 2005 and February 2006. These two health centers are located in rural areas outside Blantyre, Malawi’s second largest city. Exclusion criteria included declining voluntary counseling and testing for HIV. Participants were administered a questionnaire by the study nurses at enrollment about demographic characteristics, socio-economic factors, and malaria prevention behaviors. Participating women were encouraged to attend visits at 26, 32, and 36–38 weeks gestation according to standard antenatal care guidelines. Women were administered sulfadoxine-pyrimethamine (SP) for intermittent preventive therapy in pregnancy (IPTp) or treated for clinical malaria according to national guidelines. The HIV-infected women were given nevirapine during labor and their infants were treated with nevirapine after delivery according to Malawi’s national guidelines at the time. All HIV-infected women were referred to an antiretroviral treatment program. Participants were screened for HIV infection at enrollment with two rapid HIV-1 antibody tests: Determine (Inverness Medical Innovations, Inc., Waltham, MA) and Unigold (Trinity Biotech, Bray, Ireland). The HIV infection was defined as a positive result by both rapid tests. There was greater than 95% agreement between the two tests. Patients with discordant results were excluded from analyses. At each visit, a blood sample was collected by finger prick for preparation of thick blood smears on slides. Peripheral malaria parasitemia was assessed through microscopic examination of stained thick blood smear slides on site by trained laboratory technicians. At delivery, placental, cord, and peripheral blood samples were collected, and thick blood smears were prepared and examined as described previously. For quality control, 10% of randomly selected slides were re-examined by the laboratory supervisor at Ntcheu District Hospital. Parasitemia was defined as the presence of parasites in thick blood smears. Malaria parasites were quantified against 200 white blood cells (WBCs). Parasite density was calculated assuming 6,000 WBCs/μL of blood. Frequency of peripheral parasitemia over follow-up was defined as the number of episodes of parasitemia during follow-up visits. Because we could not distinguish between recrudescence and reinfection, episodes of parasitemia were assumed to be independent across visits. Parasitemia was analyzed in four ways: 1) the presence or absence of parasitemia at enrollment, delivery, and anytime during the follow-up period; 2) average longitudinal risk of parasitemia during follow-up, which is the marginal probability of developing parasitemia over follow-up taking into account clustering, i.e., the possibility of more than one parasitemia episode per woman; 3) number of episodes of parasitemia during follow-up; and 4) peripheral and placental parasite density at delivery. Binomial regression was used to estimate prevalence and risk ratios for parasitemia. Parasitemia risk over visits was analyzed using weighted generalized estimating equations (wGEE) to account for the possibility that data were missing at random. In these models, each individual response is weighted by the inverse probability of a missing response given the other responses, i.e., the probability of a missing measurement given the other measurements for a given subject.18 We used the following model of covariates to estimate the weight: where θhi = Pr{missing response at visit h ∣ non-missing response at visit h-1}. In the above model, h indexes visits, i indexes subjects, and k indexes covariates. The set of covariates used in the dropout model included continuous maternal weight in kg, maternal age at enrollment, indicators for access to an unsafe water source, < 8 years of education, primigravidity, husbands' occupation, and low housing quality. The polytomous outcome, number of parasitemia episodes, was analyzed using generalized logistic regression. Parasite density at delivery was analyzed using zero-inflated negative binomial (ZINB) regression. The advantage of ZINB regression is that it takes into account the semi-continuous nature (excess zeros) of parasite density and allows for overdispersion in nonzero values of parasite density. This is preferable to comparing geometric mean parasite density using Student's t test and analysis of variance or to performing analyses on log-transformed parasite density for two reasons. First, because of the high degree of skew in parasite density, log-transformation may not result in a normal distribution of transformed values precluding use of tests with a normality assumption; additionally, the optimal Box-Cox transformation that would result in a normal distribution may vary from population to population, given factors such as seasonality and transmission intensity that tend to differ across populations. Second, back-transformation of transformed values may not result in sensible results. The ZINB regression appropriately takes into account the distribution of parasite density and allows for the estimation of a predicted mean change in parasite density, which has use in determining the impact of the risk/protective factors of interest on parasite burden. Covariate inclusion in regression modeling was decided using a causal diagram. Malaria preventive behaviors are endogenous to socioeconomic/demographic factors so these variables were coupled together. Age and gravidity were correlated to the point of exchangeability; therefore, gravidity was chosen for analysis because it was the main confounder of interest. The wGEE were conducted using the macro developed by Molenbergh and Verbeke.18 All analyses except one were conducted using SAS version 9.1 for Windows (SAS Inc., Cary, NC). Zero-inflation negative binomial regression was conducted using StataSE version 10 (StataCorp., College Station, TX). Informed consent was obtained from all participating women. The study was approved by the institutional review boards of the University of North Carolina at Chapel Hill and the University of Malawi College of Medicine.
N/A
