Impact of malaria during pregnancy on pregnancy outcomes in a Ugandan prospectivecohort with intensive malaria screening and prompt treatment

The study was conducted in Mbarara district, southwestern Uganda. Thispredominantly rural area lies at an altitude of about 1,500 m above sea leveland has moderate levels of malaria transmission [32]. Between October 2006 and May 2009, 1,218 pregnant women with anestimated gestational age ≥13 weeks were enrolled in a prospectiveobservational cohort. The first 1197 women in this cohort screened for malariawith a positive rapid diagnostic test (RDT) confirmed by a positive blood smearwere invited to participate in an additional study comparing the efficacy andtolerance of artemether–lumefantrine with oral quinine for the treatmentof uncomplicated falciparum malaria published elsewhere [33].

At baseline, a comprehensive assessment of the pregnant women’ssocio-demographic characteristics and health status was performed, including amedical and obstetrical history, clinical and obstetric examination, ultrasoundevaluation, blood smear and hemoglobin measurement. Estimated gestational age bywas determined by ultrasound in all women enrolled in the study between week16–20 of pregnancy (72% of the cohort). For the remaining mothers, i.e.,those recruited after the 20^th week of gestation, we turned to apublished model that predicts gestational age from symphysis-fundal height (SFH)measurements and calibrated it using the data from the 16–20 week group [34], and then used these results to predict gestational age at deliveryin the subset of mothers without ultrasound (Additional file 1).

At delivery, blood smears were obtained from the mother, placenta, cord andnewborn to test for the presence of Plasmodium and malaria pigment.Placental histology was available only for a subset of the cohort (n=260).Placental malaria cases were classified according to the presence of parasitizederythrocytes, intervillous inflammation and haemozoin deposition [18, 35]. Newborns were given an initial standardized physical examination bya medical officer, weighed to the nearest 10g using a SECA mechanical typescale, and measured for length to the nearest centimeter using a portablestadiometer (Shorr productions, US). Infants delivered outside of a healthfacility were examined within 24 hours of birth by a study medical officer.

Paracheck® RDTs were performed using a finger-prick blood sample andinterpreted according to the manufacturer’s instructions. Thick and thinblood smears were prepared and stained with Giemsa. Parasitaemia was calculatedby counting parasites against 200 white blood cells (or 500, if nine parasitesor fewer were counted against 200 white blood cells). Placental smears weretaken by incising a fresh placenta on the maternal surface halfway between thecord and the periphery, and were then examined for the presence of parasites andpigment [35].

HIV testing and treatment was proposed to all participants and performedaccording to national guidelines [36], which include cotrimoxazole prophylaxis for people infected withHIV. Haemoglobin was measured from a fingerprick sample by the HaemocueB-Haemoglobin analyzer (Ängelholm, Sweden).

Low birth weight was defined as <2,500 g measured within 24 hours of birth;preterm as newborn gestational age <37 weeks at delivery; stillbirth as thedelivery of a non-living foetus ≥28 weeks gestation; and miscarriage asthe delivery of a non-viable foetus either at <28 weeks gestation or weighing<500 g.

Various parameters of malaria exposure during pregnancy were described andanalysed for their temporal change and for their association with maternalcharacteristics or study interventions that may have affected MiPcharacteristics. Peripheral malaria was defined as the occurrence of a positiveperipheral blood smear. After a treated malaria episode, a subsequent episodewas considered a recurrence only after a minimum of 14 days, with at least onenegative blood smear during this period [10]. Placental malaria was defined as the detection by microscopy of anyparasite in a placental or cord blood smear.

The risk of peripheral malaria infection was analysed with a mixed-effectsPoisson model [37, 38]. Since the occurrence of malaria before enrolment in the study couldnot be observed (left censoring), the at-risk time period was defined as theinterval from study enrolment to delivery. Lead time bias was (partially)accounted for by including the gestational age at enrolment as a covariate. Eachindividual follow-up (from enrollment to delivery) was split into intervalselapsing from one visit to another, and the log duration of these intervals wasincluded as an offset. Baseline risk was modeled using a spline function. Thelevel of parasitaemia (log transformed) was analysed using a linear model. Whenmore than one malaria episode was observed in a pregnancy, the maximalparasitaemia level recorded per episode was used as a dependent variable. Thepresence of fever and the occurrence of placental malaria infection wereanalysed with logistic models. In each model, maternal age, gravidity, HIVstatus, education level, residency area (rural versus urban), and gestationalage at inclusion were considered as potential risk factors.

The number of IPT doses was introduced as a time-dependent covariate in the modelfor peripheral malaria risk. However, IPT was interrupted after the treatment ofa malaria infection, making the number of IPT doses an endogenous variable [39]. Since data were censored at the first malaria episode, only therelationship between the number of IPT doses received up to the beginning of atime interval and the risk of the first malaria episode during this timeinterval was assessed (using a log-linear model).

The adverse outcomes evaluated in this study were stillbirth, preterm delivery,low birth weight and IUGR. IUGR was defined as a birth weight below the 10thpercentile of the birth weight-for-gestational age. Type I (symmetric) IUGR andtype II (asymmetric) IUGR were distinguished according to whether the Rohrerindex was above the 10th percentile of Rohrer index for gestational age or not.United States population-based references were used as standard [40, 41]. The association of each outcome with the various parameters ofmalaria exposure and with maternal characteristics were analysed separately forthe full cohort, the subset of mother--newborn pairs with no or only oneperipheral malaria infection, and the subset of mother--newborn pairs withultrasound assessment of gestational age at baseline. Maternal age, educationlevel (no education, primary level or ≥secondary level), residence area(rural versus urban), HIV status, number of clinic follow-up visits before birthoutcome (<4 versus ≥4), and the newborn’s gender and gestationalage at birth were included in all models. Stillbirth was analysed as a binaryvariable using a logistic model. Preterm delivery was analysed with gestationalage at birth included as a continuous or binary variable (gestational age <37weeks) using respectively a linear and logistic model. Weight and length atbirth were both considered as continuous variables and analysed with a linearmodel adjusted for gestational age at birth. Parasitaemia was categorized asnone, low (log parasitaemia ≤6 log parasites/μL) or high (>6 logparasites/μL). Late malaria infection was defined as a peripheral malariainfection occurring in the last two weeks before delivery. To better understandthe effect of malaria infection timing independently of the enrolment timing,the association between birth weight and gestational age at infection (<15,15-<20, 20-<24, ≥24 weeks) was analysed with a linear modelrestricted to the subset of mothers with no or only one malaria infection andwith a gestational age <15 weeks at enrolment.

All analyses were performed using the open source statistical software R [42].

Written informed consent for study participation was obtained from allparticipants to the study. The study was approved by the institutional reviewboards of Mbarara University of Science and Technology, Uganda National Councilfor Science and Technology, and France’s “Comité de Protectiondes Personnes - Ile-de-France XI”. This study was registered withClinicalTrials.gov, number NCT00495508.

Of the 1,218 women enrolled in the cohort, 149 (12%) were excluded from thisanalysis because they were lost to follow-up before their pregnancy reached anoutcome (Figure 1). These excluded women were youngerand had shorter follow-up (p=0.0001). One maternal death unrelated to malariawas observed (from sepsis five days after a caesarean section for obstructedlabour in a term pregnancy).

Characteristics of the 1,069 women included in this cohort are summarized inTable 1. Women with malaria were enrolledsomewhat later in their pregnancy than those with no infection detected(Table 1). Most of the mothers delivered in ahealth facility (84% at the regional hospital and 3% in a private clinic), withthe remainder delivering at home. Median gestational age at enrolment was 19weeks (Inter Quartile Range, IQR: 16–22) and median follow-up time was 21weeks (IQR: 16–24). Almost 50% of the mothers reported the use of bed netsbefore the first visit. The mean number of visits, equivalent to the mean numberof screening tests, was high (18, IQR: 10–23) (Table 1).

Mothers without ultrasound assessment of gestational age were more likely to livein remote rural areas (OR: 1.68, 95%CI: 1.21 - 2.31) and to be at a moreadvanced stage of the pregnancy at enrolment (mean gestational at enrolment was23.5 weeks in mothers without ultrasound, versus 18.5 weeks in those withultrasound, p<0.001). Of the 1,018 live births, 40 were excluded from theanalysis on birth weight and length (Figure 1).

In total, there were 57 (7%) infants with low birth weight, 39 (6%) with type IIIUGR (asymmetric) and 89 (13%) type I IUGR (symmetric). In analysis adjusted forgender and for estimated gestational age at delivery, birth weight was reducedin primiparae, in mothers with low education level and in those who attended≤4 follow-up visits (Table 4). Greater birthweight was observed in mothers with higher average hemoglobin level duringpregnancy (+30g, 95%CI: 10–50), however this association disappeared inmultivariate analyses adjusted for MiP. Of the various parameters of malariaexposure during pregnancy, all but placental malaria and symptomatic malariawere associated with lower weight at birth (Figure 3). However, the presence of malaria pigment in the placental smear wasassociated with reduced birth weight in univariate analysis (−0.26, 95%CI:-0.49 to −0.032) though not in adjusted analysis (p=0.2). Type I IUGR wasnot associated with any aspect of malaria in pregnancy (Additional file 2). By contrast, type II IUGR was increased in motherswith more than one malaria episode and with symptomatic malaria (Additional file3). Median gestational age at last malariainfection was 21 weeks (19–22) in women with type I IUGR and 23 weeks(22–24) in women with type II IUGR (p=0.08).

Analyses restricted to mothers with ultrasound-verified gestational age, or tothose with no or only one malaria infection, found qualitatively similarassociations, although sometimes with only borderline significance(Figure 3). Birth weight was not associated withthe timing of the first malaria infection in mothers with gestational age <15weeks at enrolment (p=0.8). This conclusion did not change when the cut-offsused for gestational age at enrolment was varied. No association was foundbetween malaria exposure during pregnancy and newborn length.