Efficacy and safety of intermittent preventive treatment and intermittent screening and treatment versus single screening and treatment with dihydroartemisinin–piperaquine for the control of malaria in pregnancy in Indonesia: a cluster-randomised, open-label, superiority trial

Summary Background Plasmodium falciparum and Plasmodium vivax infections are important causes of adverse pregnancy outcomes in the Asia-Pacific region. We hypothesised that monthly intermittent preventive treatment (IPT) or intermittent screening and treatment (IST) with dihydroartemisinin–piperaquine is more effective in reducing malaria in pregnancy than the existing single screening and treatment (SST) strategy, which is used to screen women for malaria infections at the first antenatal visit followed by passive case detection, with management of febrile cases. Methods We did an open-label, three-arm, cluster-randomised, superiority trial in Sumba (low malaria transmission site) and Papua (moderate malaria transmission site), Indonesia. Eligible participants were 16–30 weeks pregnant. Clusters (antenatal clinics with at least ten new pregnancies per year matched by location, size, and malaria risk) were randomly assigned (1:1:1) via computer-generated lists to IPT, IST, or SST clusters. In IPT clusters, participants received the fixed-dose combination of dihydroartemisinin-piperaquine (4 and 18 mg/kg per day). In IST clusters, participants were screened with malaria rapid diagnostic tests once a month, whereas, in SST clusters, they were screened at enrolment only. In all groups, participants with fever were tested for malaria. Any participant who tested positive received dihydroartemisinin–piperaquine regardless of symptoms. The primary outcome was malaria infection in the mother at delivery. Laboratory staff were unaware of group allocation. Analyses included all randomly assigned participants contributing outcome data and were adjusted for clustering at the clinic level. This trial is complete and is registered with ISRCTN, number 34010937. Findings Between May 16, 2013, and April 21, 2016, 78 clusters (57 in Sumba and 21 in Papua) were randomly assigned to SST, IPT, or IST clusters (26 clusters each). Of 3553 women screened for eligibility, 2279 were enrolled (744 in SST clusters, 681 in IPT clusters, and 854 in IST clusters). At enrolment, malaria prevalence was lower in IST (5·7%) than in SST (12·6%) and IPT (10·6%) clusters. At delivery, malaria prevalence was 20·2% (128 of 633) in SST clusters, compared with 11·6% (61 of 528) in IPT clusters (relative risk [RR] 0·59, 95% CI 0·42–0·83, p=0·0022) and 11·8% (84 of 713) in IST clusters (0·56, 0·40–0·77, p=0·0005). Conditions related to the pregnancy, the puerperium, and the perinatal period were the most common serious adverse events for the mothers, and infections and infestations for the infants. There were no differences between groups in serious adverse events in the mothers or in their infants. Interpretation IST was associated with a lower prevalence of malaria than SST at delivery, but the prevalence of malaria in this group was also lower at enrolment, making interpretation of the effect of IST challenging. Further studies with highly sensitive malaria rapid diagnostic tests should be considered. Monthly IPT with dihydroartemisinin–piperaquine is a promising alternative to SST in areas in the Asia-Pacific region with moderate or high transmission of malaria. Funding Joint Global Health Trials Scheme of the Medical Research Council, Department for International-Development, and the Wellcome Trust.


eMethods
The protocol is available as supplemental information.

Details of study site and antenatal clinic
The study area consisted of several sub-districts. In each sub-district, there is one community health centre ("Puskesmas") that serves a population of around 30,000 people. Under each Puskesmas maternal health services are provided through sub-health Posts (Pustu) based in the communities and covering about 2-3 villages with a population size of 500-1500 people per village and through community integrated services ("Posyandu") held monthly in the village. Most women receive antenatal care through the Pustu and Posyandu, while most facility-based deliveries occur in the Puskesmas, or in the district hospital.
Each Puskesmas provides a monthly report to the district health office of the malaria smear positive cases including data from the Pustu and Posyandu. Based on this data the annual parasite incidence (API) is calculated as the annual number of positive malaria slides x 1000 / Total population for each village within the catchment area of each Puskesmas.

Malaria epidemiology
In Papua Indonesia, the study was conducted in Mimika district in southern Papua with its capital Timika. Modelling studies based on cross-sectional survey data suggested that in 2010 malaria transmission in most of this district is intermediate (PfPR2-10 predicted prevalence of 5-40%). 1,2 In 2013, the annual incidence of parasitaemia was 450/1000, with P. falciparum and P. vivax respectively causing 60% and 40% of cases, without significant seasonal fluctuation. [1][2][3] In the same year, a community-based cross-sectional survey involving 2,830 individuals of all ages found that 37.7% had detectable malaria parasitaemia in the peripheral blood by microscopy or polymerase chain reaction (PCR), and 13.9% by microscopy alone. Approximately 99% of these infections were due to P.falciparum and P.vivax mono-infections, and the remaining due to P.malariae. Although most infections were asymptomatic, those with any parasitaemia, including sub-microscopic infections, were at significant risk of anaemia. 3 In Sumba, the study was conducted in south-west Sumba district. Modelling studies based on crosssectional survey data suggested that in 2010 malaria transmission in most of this district was low (PfPR 2-10 predicted prevalence of <5%), although some areas have intermediate levels of transmission defined as PfPR2-10 5-40%. 1,2 In a large cross-sectional survey in 2007 involving 8,870 individuals of all ages, the prevalence of malaria by expert microscopy (any species) was 6.8% in the rainy season and 4.9% in the dry season. 4 The seasonal variation in malaria prevalence reflected changes in the prevalence of P.falciparum infection, which was higher in the rainy vs dry season (4.9% vs 2.9%). There were no seasonal differences in the prevalence of P.vivax (~2.2%) and P. malariae (~0.1%). 4 In the same area, a screening study in 2012 found that approximately 3.2% of pregnant women were positive by RDT and 6.6% by PCR during routine scheduled antenatal clinic visits. 5 Antimalarial drug resistance In Papua, both P.falciparum and P.vivax parasites in Mimika district are highly resistant to sulfadoxine-pyrimethamine and chloroquine. 6 Prior to the switch to DP as first-line treatment in March 2006, at least 95% of patients with P.falciparum malaria treated with chloroquine mono-therapy had recrudescent infections by day 28 7 and this was approximately 50% for the combination of chloroquine plus sulfadoxine-pyrimethamine. [7][8][9] High-grade chloroquine resistant P.vivax was first reported from Papua, Indonesia and Papua New Guinea in the late 1980, early 1990s. 8,10,11 A recent review of its impact 12 showed that in Papua, 60-90% of patients had recurrent malaria within 28 days 6,7,12 and 2% of patients infected with P.vivax treated with chloroquine monotherapy subsequently require admission to hospital. 6,13 In 2006 to 2008, 61.8% of P.vivax parasites carried a quadruple mutant genotype in the genes encoding for sulfadoxine-pyrimethamine resistance. 14 The clinical efficacy of DP against both P.falciparum and P.vivax in Papua remains excellent 9 years after extensive use since its introduction in March 2006. 15 In Sumba, chloroquine resistant P.falciparum is widespread. In a survey in 2007, the prevalence of the 76T allele of the pfcrt gene was 89% and half of these parasites also carried pfmdr1 mutant alleles. 16 Although recent data is lacking, it is believed that the levels of resistance to sulfadoxinepyrimethamine in Sumba is low, in contrast to Papua and other parts of Indonesia. In 2007, only 1% of P.falciparum isolates carried the double dhfr/dhps mutant genotype, and none the quadruple or quintuple mutant genotype. 16 A single therapeutic study in 2010 in patient with P.vivax malaria, showed that 98% of patients with P.vivax infections were successfully treated with sulfadoxinepyrimethamine and only 3.3% of P.vivax parasites carried the quadruple mutant genotype reflecting sulfadoxine-pyrimethamine resistance, compared to 61.8% in Papua. 14 Over 99% P. falciparum isolates carried wild type K13 markers. 17

Randomisation
The ANC clinics constituted the units of randomisation and were identified in advance by the lead investigators based in Indonesia (RA, JRR, DS) and their clinic identification number provided to the trial statistician (BF) based in the UK. A 1:1:1 allocation ratio was used. To minimize imbalances across treatment groups with respect to baseline malaria prevalence and risk factors for malaria, multivariate matching was used, based on the Government's annual parasite incidence (API) data for the two years preceding the study, geographical area (site [Sumba and Papua] and then sub-district within each site), and clinic size (prior annual number of new ANC attendees). In this way, the 78 eligible clinics was blocked into 26 sets of 3 matched clusters. A total of 20 randomisations were then generated using computer-generated random numbers, in which the three clinics in each triple were allocated to the three (arbitrarily labelled) groups A, B and C. For each randomisation, an imbalance score was calculated based on prevalence of positive RDTs in previous 12-months, geographical area and clinic size. The three randomisations with the "best" (smallest) imbalance score were then selected and each allocated a new computer-generated random number -the randomisation allocated the largest of these numbers was used in the study. A list with the dummy allocation for each cluster (as A, B, C) was then sent by the trial statistician based in the UK to the principal investigators in Indonesia.

Public ceremony
A public randomisation ceremony, organised by the principal investigators, was held in each of the two sites, attended by District health officials and village representatives. A district health official not involved in the study, first drew one of three identical looking opaque sealed envelopes which contained the dummy allocation from one box. The content of each envelope containing the dummy allocation for each cluster was then displayed to the audience. A second health official, then drew, from the second box, one of three other identical looking opaque sealed envelopes containing the actual allocation. Prior to opening this second set of envelopes, the health official labelled each with dummy code from the first set of envelops (A B or C) after which they were opened to reveal the allocated assignment.

Definitions of morbidity endpoints
Birthweight data The aim of the study was to measure birthweight within 24 hours after birth. Birth weights taken 24-48h hours (n=8, 0.4%), and 48-168 hours after delivery (n=2, 0.1%) were corrected for the physiological fall in birth weight in breastfed infants occurring in the first days following delivery 18,19 by a factor +2% and +4%, respectively to obtain the estimated weight at birth. 20,21 All analyses used corrected birthweight unless indicated otherwise. Low birth weight was defined as the corrected birthweight <2,500 gram.

Gestational age and preterm
Gestational age was assessed at enrolment using the date of the last menstrual period, fundal height, and at delivery using the modified Ballard score. If more than one gestational age measurement was available we used estimates in the following order of preference: Neonatal clinical exam within 96 hours of delivery (modified Ballard score), last menstrual period (if known), and fundal height at enrolment. Preterm was defined as a gestational age of less than 37 completed weeks.
Small for gestational age (SGA) SGA was defined as birthweight below the tenth percentile of an external reference population for a given gestational age and sex. Small for gestational age (SGA) was defined as birthweight below the tenth percentile for a given gestational age and sex, using the new INTERGROWTH reference population, 22 which was also used to calculate the birthweight-for-gestational age Z-scores.

Congenital malaria
Any asexual malaria parasitaemia detected by microscopy, RDT or Loop-mediated isothermal amplification (LAMP) or PCR in cord blood or in the peripheral blood within 7 days of birth.

Assessment of compliance and tolerance to DP intake
A reminder phone text message or phone call was made on the day of the 2 nd and 3 rd dose of each course to participants who were given DP to take at home. If women were not contactable by phone, a field staff visited participant's homes to ensure the study drug was taken. On day 3, a home visit was made to all women who received DP to check compliance and whether the participant experienced any side effects.

Laboratory methods mRDTs, malaria microscopy, haemoglobin assessment and histopathology
The mRDTs used targeted histidine-rich protein-2 (HRP2) and parasite lactate dehydrogenase (pLDH) (sensitivity to detect 200 parasites/µl: P.falciaprum=85%, P.vivax=74%) (First Response Malaria Ag pLDH/HRP2 Combo [I16FRC30], Premier Medical Corporation Ltd, India). 23 All mRDT-positive women (positive HRP2-or pLDH-bands) in any arms were treated with dihydroartemisinin-piperaquine. Women with a history of dihydroartemisinin-piperaquine intake in the previous 4 weeks received quinine-clindamycin. 7 Malaria smears were not used for point of care. All smears were read first by an expert microscopist on site who was blind to the mRDT results. All positives and a random selection of 10% of negative smear results (randomly selected) were read by the senior expert microscopist at the Eijkman Institute, Jakarta who was blinded to the results of the first reading. If one of the two smears were declared positive, LAMP/PCR findings were used. Malaria infection was defined as the presence of asexual Plasmodium parasites (any species) in a thick blood smear. Parasite densities were counted against 300 white blood cells and expressed per 8,000 parasites per microlitre. Smears were declared negative if no parasites were detected after examining 200 high power fields. Thin smear was used to identify malaria species (PCR confirmed species were subsequently used in the analysis). Placental incision smears (thick and thin) were read and parasite density calculated similar to maternal peripheral smears. Haemoglobin levels were determined using portable HemoCue Hb 201+ (HemoCue AB, Ängelholm Sweden) machines following manufacture instructions. Malaria rapid diagnostic test (RDT) was performed as per the manufacturer's instruction. Tissue samples for placental histopathology were collected from the maternal side of the placenta and fixed with 10% neutral buffered formalin and then processed, stained with hemotoxin-eosin, and examined under standard light and polarized microscope following standard procedures. 24 Histopathological slides were read in duplicate by two independent readers who were blinded to the placental and maternal smear results, and any discrepant results were resolved by one of the investigators (RA).

DNA extraction
Dried blood spots (DBSs) for LAMP and PCR assays were collected on Whatman's #3 filter paper, air dried and placed in individual plastic bags with desiccant and stored at room temperature. From these spots, genomic DNA (gDNA) was extracted using chelex-100 ion exchanger (Biorad Laboratories, Hercules, CA). Briefly, 6 mm filter paper disc punches were incubated in 0.5% saponin in PBS overnight, centrifuged for 10 minutes at 12000 rpm, supernatant discarded, washed in PBS, centrifuged for 5 minutes at 12000 rpm and the supernatant discarded (this procedure was repeated 3 times). The sample was then heated at 100 o C in 150 µl of 20% Chelex 100-Ion Exchanger for 10 minutes and centrifuged for 10 minutes at 12000 rpm. The resultant 100 µl supernatant was stored at -20 o C.

Loop mediated isothermal amplification (LAMP)
LAMP assays were conducted at the Eijkman institute in Jakarta using to the Eiken Loopamp™ MALARIA Pan Detection kit procedures (Eiken Chemical Company, Japan). 15 μL of DNA plus 15 μL of water was added to the malaria Pan reaction tube with one negative and one positive control included in each 16 reactions; primers, buffers and enzymes were reconstituted by inverting the samples in the lid of the reaction tubes, tubes were briefly spun and incubated at 65 °C for 40 minutes before polymerase inactivation at 80 °C for 5 minutes. 25,26 The limit of detection of LAMP assays is ~1 parasite/ µl.

Real-time PCR (quantitative PCR)
A multiplex real-time PCR was used to simultaneously to detect the four main species of Plasmodium: falciparum, vivax, ovale and malariae. Primers and probes were used at identical concentrations as previously published 27,28 in total reaction volume of 10 μL. Each reaction contained 2 μL of gDNA, 1x Quantifast Pathogen PCR Master Mix and primer/probe concentrations as outlined in eTable 1. Amplification and real-time measurements were carried out using the Rotor-Gene Q 5plex HRM Platform (Qiagen, Hilden Germany) in a 72-optical tube format. The thermal cycling profile was as follows: 95°C for 10 minutes, followed by 95°C for 15 seconds and then 60°C for 60 seconds for 38 cycles. Cycle threshold (Ct) values were calculated and analysed with the Rotorgene Q series software version 1.7 (Qiagen Inc, Valencia, CA, USA). Positive DNA controls for each species (provided by Malaria Reference Laboratory, Public Health England) and non-template controls (NTCs) were also included. The limit of detection of each of these primer/probe sets are between 0.1-1 parasite/ µl. Nested PCR (nPCR) Nested PCR was performed using previously described primers and reaction conditions. 29 The limit of detection of this assay is estimated to be 6-10 parasites/ µl.

Definition of LAMP/PCR positivity
All samples were first tested for malaria using Loop mediated isothermal amplification (LAMP) in the laboratories of the Eijkman Institute in Jakarta Indonesia. The multiplex real-time PCR was then used to determine the species. In addition, a random sample of 5% of the LAMP negatives was also tested using qPCR. Any LAMP-qPCR discordant samples were tested using nested PCR. The following algorithm was used to define LAMP/PCR positivity (see Figure 1 Definition of PCR/LAMP positivity).

Sample size and power calculations
Original sample size calculations The trial was originally designed to detect a 50% or greater reduction in malaria infection at delivery, from 10.0% in the SST group to 5.0% in any of the interventions group with 90% power, 2-sided alpha of 0.025, and an assumed ICC of 0.002, and accounting for a 13% efficiency loss due to varying cluster sizes, and 20% loss to follow-up. This required 3,198  ). An interim sample re-estimation was then conducted in a blinded manner using the observed pooled event rate in each site across the 3 arms and the observed ICC values.
The pooled event rate of the primary endpoint was then used to estimate the frequency in the control arm (SST), by assuming a 50% reduction in the IPT arm relative to the control arm (SST) (RR=0.50, as per protocol), and assuming no reduction in the IST arm (RR=1.0) (based on new data from recently completed IST trials in Kenya and Malawi). 30, 31 For example, if the observed pooled prevalence was 10%, then this was assumed to be a combination of SST=12%, IST=12% and IPT=6% (0.5 x 12%), and an equal distribution of women by arm (1:1:1 allocation). Similarly, for a prevalence of 20% this was assumed to be the summary of 24% in the SST and IST arms and 12% in the IPT arm.
Power and sample size calculations where then conducted using NCSS/PASS to estimate the required extension that would provide at least 80% power to detect a 50% reduction in the primary endpoint across the two sites pooled using the 'metapow' command in Stata, and 80% to 90% power to detect a similar reduction in Papua alone, while allowing for 13% loss in efficiency due to cluster size variation and 20% loss to follow-up.
This suggested that in Papua a total of 1,290 women overall and 903 completers (301/arm, 43 in 7 clusters) were required on top of the 989 women recruited in Sumba (i.e. 2,279) to achieve at least 80% power overall across the two study sites pooled (alpha 0.0167, ICC=0.005). This sample size was also estimated to provide 87% power to detect a 50% reduction from 24% to 12% in Papua alone and also had 80% power to detect a 50% difference if the prevalence of malaria was only 21% or if the ICC was 0.01 instead of 0.005 (NCSS/PASS). Analysis

Cardiac monitoring
Electrocardiography was performed in a subgroup of women in the IPT-DP arm to determine whether previously documented transient QTc prolongation associated with DP increases in magnitude with subsequent courses. The study was conducted in the Papua site Indonesia. Written informed consent was obtained from all women.
Pregnant women enrolled in the IPT arm of the main trial willing to complete the study schedule were eligible. Following enrolment, women had an ECG measured at baseline and then again 4-6 hours after taking the 3 rd dose of each course of DP. This timing was chosen as it represents the time of the expected maximum concentration of DP (anticipated Tmax), which has been shown to correlate with the expected maximal prolongation of the QT interval. 32 With each subsequent monthly treatment course, the ECG was repeated 4-6 hours after the 3 rd dose. All ECGs were done in triplicate 30 to 60 seconds apart. ECGs were read on site, and again by a cardiologist at Cardiabase, the Banook Group, France and results reported back to the study team in Indonesia. SAEs (QTcF > 480 ms or delta QTcF from baseline >60 msec) were reported in an expedited manner. Any woman with a QTcF >480 ms or delta QTcF from baseline >60 msec were withdrawn from receiving additional doses of DP, but followed per study protocol.
It was estimated that 33 women were required to allow detection of a 20ms difference in QTc from baseline following exposure to DP, assuming an estimated standard deviation of 30 ms, with 90% power, at a significance level 0.05 using a two-sided one-sample t-test and allowing for 20% loss to follow-up.
The primary endpoint was the change in QTc from baseline (hour 0; i.e. prior to the first course of DP) to 4-6 hours following receipt of the third dose with each course of DP.
The mean of the triplicate ECGs measurements taken 30 to 60 seconds apart were used for analysis.
The primary analysis was based on Fridericia's method to obtain heart rate corrected QTc intervals (observed QT interval divided by cube root of RR interval, in seconds [QT / (RR) 0.33 ]). A sensitivity analysis was conducted using the same analytical approach but now using the Bazett's method to obtain QTc intervals (observed QT interval divided by the square root of RR interval, in seconds [QT / (RR) 0.5 ]). QTcB value using Bazett's correction were not considered for clinical care, but were also calculated for data analysis.

Other sub studies
As indicated in the original trial protocol, the study included a second main objective "To determine the acceptability, feasibility and cost effectiveness of SST, IST and IPT alongside the randomised control trial." The results of the acceptability and feasibility studies have been published previously. [33][34][35] The cost-effectiveness analysis is ongoing and will also be published elsewhere.
12 eResults eTables eTable 2: Follow-up visits schedule (intention to treat population) 48 ITT=intention to treat population, PPP=per protocol population, Crude=unadjusted for co-variates, IST=intermittent screening and treatment during pregnancy with dihydroartemisinin-piperaquine. IPT=intermittent preventive treatment during pregnancy with dihydroartemisinin-piperaquine. SST=Single screening and treatment during pregnancy with dihydroartemisinin-piperaquine; py=person years. P-values for interaction terms were obtained post-hoc using the Altman-Bland method. 37 * Outcomes represents binary outcome except for the data for the incidence per 100 person-years. † The p-value for interaction could not be computed because the relative risk in at least one of the study sites could not be computed because of zero events in at least one of the study arms. 19 eTable 7: Site-treatment interaction P-values for secondary outcomes related to malaria at the time of delivery

20
ITT=intention to treat population, PPP=per protocol population, Crude=unadjusted for co-variates, IST=intermittent screening and treatment during pregnancy with dihydroartemisinin-piperaquine. IPT=intermittent preventive treatment during pregnancy with dihydroartemisinin-piperaquine. SST=Single screening and treatment during pregnancy with dihydroartemisinin-piperaquine; py=person years. P-values for interaction terms were obtained post-hoc using the Altman-Bland method. 37 * Outcomes represents binary outcome except for the data for the incidence per 100 person-years. † The p-value for interaction could not be computed because the relative risk in at least one of the study sites could not be computed because of zero events in at least one of the study arms.

22
ITT=intention to treat population, PPP=per protocol population, Crude=unadjusted for co-variates, IST=intermittent screening and treatment during pregnancy with dihydroartemisinin-piperaquine. IPT=intermittent preventive treatment during pregnancy with dihydroartemisinin-piperaquine. SST=Single screening and treatment during pregnancy with dihydroartemisinin-piperaquine; py=person years. P-values for interaction terms were obtained post-hoc using the Altman-Bland method. 37 * Outcomes represents binary outcome except for the data for the incidence per 100 person-years. † The p-value for interaction could not be computed because the relative risk in at least one of the study sites could not be computed because of zero events in at least one of the study arms. 23 eTable 9: Site-treatment interaction P-values for malaria outcomes and non-malaria sick visits during pregnancy 48 ITT=intention to treat population, PPP=per protocol population, Crude=unadjusted for co-variates, IST=intermittent screening and treatment during pregnancy with dihydroartemisinin-piperaquine. IPT=intermittent preventive treatment during pregnancy with dihydroartemisinin-piperaquine. SST=Single screening and treatment during pregnancy with dihydroartemisinin-piperaquine; py=person years. P-values for interaction terms were obtained post-hoc using the Altman-Bland method. 37 * Outcomes represents binary outcome except for the data for the incidence per 100 person-years. † The p-value for interaction could not be computed because the relative risk in at least one of the study sites could not be computed because of zero events in at least one of the study arms. 25 eTable 10: Site-treatment interaction P-values for key secondary outcomes newborn 49 ITT=intention to treat population, PPP=per protocol population, Crude=unadjusted for co-variates, IST=intermittent screening and treatment during pregnancy with dihydroartemisinin-piperaquine. IPT=intermittent preventive treatment during pregnancy with dihydroartemisinin-piperaquine. SST=Single screening and treatment during pregnancy with dihydroartemisinin-piperaquine; py=person years * Outcomes represents binary outcome except for the data for the incidence per 100 person-years. † The p-value for interaction could not be computed because the relative risk in at least one of the study site could not be computed because of zero events in at least one of the study arms. 27 eTable 11: Maternal and fetal mean haemoglobin, birthweight, gestational age and mean birthweight-for gestational age Z-score by treatment group and site  For IST and SST this includes both treatments based on RDT positivity detected during a scheduled screening event, and data from treatment given to RDT positive women presenting for unscheduled sick visits (there were none in the IPT arm). b incidence rate per 100 person-years c includes MeDRA's preferred terms for 'abdominal pain', 'abdominal pain lower' and 'abdominal pain upper' d includes MeDRA's preferred terms for 'rash pruritic' and 'rash macular' e includes MeDRA's preferred terms for 'diarrhoea' and 'diarrhoea haemorrhagic' f Late vomiting (>30 minutes following drug administration). All of these events occurred within the first 3 days after the start of drug intake, i.e. during or within the 24h after drug intake but excluding the first 30 minutes. 30 eTable 13: Serious Adverse Events: case descriptions of maternal deaths Arm IST Post-partum haemorrhage A 33-year-old gravidae 3, enrolled in IST arm presented at 36 weeks gestation to the Timika General Hospital with contractions and fever. She gave birth to a live baby after four hours by spontaneous vaginal delivery. After delivery, she had retention of the placenta and died due to post-partum haemorrhage on the same day. She had been enrolled into the study two months earlier at 28 weeks of gestation by fundal height examination. She had one previous live birth and one abortion. Her physical condition and vital signs at enrolment were normal and fetal viability was confirmed by Doppler. The mRDT was negative and her Hb was 9.5 g/dL. She received ferrous sulfate tablets 200mg/day and calcium supplement 500mg/day. The physical examination at her first scheduled antenatal follow up visit at one month was normal. Her mRDT at that visit was negative. Because all her RDTs were negative throughout her antenatal follow-up, she did not receive DP during her pregnancy. IST Spontaneous abortion in field followed by maternal death at home A 31-year-old primigravidae enrolled in IST arm had a post-partum death. The study staff became aware of her death three weeks later when she failed to attend the scheduled follow up visit. The history of event was as follows: In the afternoon of the previous day she left her home to go to the field about 10 kilometres from her village. When she left home, she was in good condition according to her sister. While in the field, she went into labour and gave birth spontaneously to a live born baby. She wrapped the baby, with the umbilical cord intact, using her clothes and went to a house in the nearest village to find help. The baby had died by the time she arrived. The residents of the house got the village traditional birth attendant to deliver her placenta. There was no history of post-partum bleeding and the following day she went back home. While at home she fainted. The family decided to take her to the local hospital, but she died on the way. She was enrolled into the study on 2 weeks before her death. At enrolment, her gestational age by fundal height examination was 26 weeks and fetal heart was detected by Doppler. Her physical examination was normal. Her haemoglobin was 10.3g/dL and mRDT was negative. She did not receive DP during her pregnancy and she did not take any other medications.

SST
Septic shock secondary to pneumonia A 20-year-old primigravida enrolled in SST arm presented to hospital emergency room with haemoptysis and dyspnoea. She had vomited about 50 cc of blood about three hours before she came to hospital. There was no previous history of dyspnoea or haemoptysis. She was admitted comatose with a diagnosis of septic shock secondary to pneumonia. Soon after admission her condition deteriorated, and she was declared dead 46 minutes later. The blood sugar on admission was 170 mg/dl and malaria microscopy were negative. She received the following medication intravenously: Asering 1500 cc/24 hours, ceftriaxone 2x2g, antrain 1g/8hrs, ranitidine 50mg/8hrs, tranexamic acid 1000mg/8hrs. She was enrolled into the study five months earlier at 17 weeks gestation estimated by fundal height. Her physical examination was normal. The mRDT was negative. She subsequently attended three scheduled ANC visits. Her medical and obstetric history and examinations were all normal during these visits. She received supplementation with ferrous sulfate 200mg daily and calcium (kalk) 500mg daily. She did not receive DP during her pregnancy. RR=Relative Risk, IRR=Incidence rate ratio, Adjusted RR or IRR obtained using the same covariates as in Figure 2 of the main text. * Data represents n/N (%) except for the data for the incidence per 100 person-years which represent the number of women with an event, the number of events, the follow-up person time and in brackets, the incidence rate per 100 person-years. † RR/IRR and p-value could not be computed because of zero events in at least one of the arms. eFigure 6: Malaria at delivery (primary outcome) and key secondary outcomes in the per protocol population (IST vs SST) RR=Relative Risk, IRR=Incidence rate ratio, Adjusted RR or IRR obtained using the same covariates as in Figure 2 of the main text. * Data represents n/N (%) except for the data for the incidence per 100 person-years which represent the number of women with an event, the number of events, the follow-up person time and in brackets, the incidence rate per 100 person-years. † RR/IRR and p-value could not be computed because of zero events in at least one of the arms. eFigure 7: Malaria at delivery (primary outcome) and key secondary outcomes in the per protocol population (IPT vs IST) RR=Relative Risk, IRR=Incidence rate ratio, Adjusted RR or IRR obtained using the same covariates as in Figure 2 of the main text. * Data represents n/N (%) except for the data for the incidence per 100 person-years which represent the number of women with an event, the number of events, the follow-up person time and in brackets, the incidence rate per 100 person-years. † RR/IRR and p-value could not be computed because of zero events in at least one of the arms. RR=Relative Risk, IRR=Incidence rate ratio, Adjusted RR or IRR obtained using the same covariates as in Figure 2 of the main text, RDT=Rapid diagnostic test for malaria, fever-RDT=RDT taken among women with document fever or a history of fever in the last 48 hours in all 3 arms. * Data represents n/N (%) except for the data for the incidence per 100 person-years which represent the number of women with an event, the number of events, the follow-up person time and in brackets, the incidence rate per 100 person-years. † RR/IRR and p-value could not be computed because of zero events in at least one of the arms. 45 eFigure 15: Malaria and non-malaria sick visits during pregnancy by treatment group in the intention-to-treat population (IST vs SST) RR=Relative Risk, IRR=Incidence rate ratio, Adjusted RR or IRR obtained using the same covariates as in Figure 2 of the main text, RDT=Rapid diagnostic test for malaria, fever-RDT=RDT taken among women with document fever or a history of fever in the last 48 hours in all 3 arms. * Data represents n/N (%) except for the data for the incidence per 100 person-years which represent the number of women with an event, the number of events, the follow-up person time and in brackets, the incidence rate per 100 person-years. † RR/IRR and p-value could not be computed because of zero events in at least one of the arms. 46 eFigure 16: Malaria and non-malaria sick visits during pregnancy by treatment group in the intention-to-treat population (IPT vs IST) RR=Relative Risk, IRR=Incidence rate ratio, Adjusted RR or IRR obtained using the same covariates as in Figure 2 of the main text, RDT=Rapid diagnostic test for malaria, fever-RDT=RDT taken among women with document fever or a history of fever in the last 48 hours in all 3 arms. * Data represents n/N (%) except for the data for the incidence per 100 person-years which represent the number of women with an event, the number of events, the follow-up person time and in brackets, the incidence rate per 100 person-years. † RR/IRR and p-value could not be computed because of zero events in at least one of the arms. Except for one woman, all women were afebrile at the time of drug administration; the one woman who was febrile, had an axillary temperature (37.7 o C) at enrolment (first course) and had P. falciparum parasites on the malaria smear. Out of the 33 women, 5, 5, 13, 9 and 1 received 1, 2, 3, 4, or 5 courses respectively. A total of 126 ECGs were taken in triplicate. There were no clinical cardiac adverse events Overall, the best correction of the QT interval for heart rate was obtained with Bazett's formula. The heart rate differed slightly during the course of pregnancy, and was highest during the first baseline visit resulting in a lower QTc value with Fridericia's method (eTable 14 There was no evidence that the mean increase in QTcF or QTcB increased with subsequent courses (eTable 14) and eFigure 20).
With Fridericia's methods, two women had QTcF values exceeding 480ms and none had values above 500 ms. Among the two women with values exceeding 480ms, this occurred after the first course of DP in one woman, when her QTcF was 484ms compared to 426ms at baseline, a 58ms (13.6%) increase. In the other woman the QTcF increased from 418 at baseline to 443 after the first course, and 475 and 489 after the 2 nd and 3 rd course respectively, which was a 24, 57 and 71ms increase compared to baseline. Both women had a normal sinus rhythm and no other abnormalities on the ECG. Neither of these women received a subsequent course. The latter woman was the only woman who showed a consistent increase in the QTcF values with each subsequent course. All other women showed no change or a relative decline in the magnitude of QTcF prolongation with each subsequent course.
With Bazett's method there were 7 women with QTc values above 480 ms; None at baseline, 4 after the 1 st course (12.1%), 2 after the 2 nd (7.1%) and 1 after the 3 rd course (4.4%). Among 2 of these the value exceeded 500 ms. These were the same two women with values exceeding 480 ms with the Fridericia's method.

Overview design
This is an open-label three-arm parallel-group matched cluster-randomised controlled superiority trial conducted in two rural sites in Eastern Indonesia with low levels of P.falciparum and P.vivax malaria comparing the efficacy, safety and cost-effectiveness of intermittent screening and treatment (ISTp) and intermittent preventive therapy (IPTp) with the current single screening and treatment (SSTp) strategy. Dihydroartemisinin-piperaquine (DHP) is used in all three arms. The trial is designed to detect 50% reduction in any malaria infection at delivery (peripheral or placental, any species, detected by microscopy, RDT, histology [acute/chronic] or PCR/LAMP) from 10% to 5% in women at delivery. The unit of randomization is antenatal-clinics. The initial study design (protocol v2.0) required 26 clusters of 41 women per cluster/arm for an overall sample size of 3198 women. The revised sample size requires a total of 2279 women from 26 clusters per arm; 989 women in Sumba (57 clusters) and 1290 in Timika (21 clusters). It is open label because it will not be possible to blind the participants to their allocation, although laboratory staff undertaking trial related diagnostic tests will be blinded. Health service related studies are conducted to assess the acceptability of the 3 interventions and the feasibility of screening policies. This study will collect data which will allow an analysis of the costeffectiveness of the three different strategies proposed.

Trial objectives:
1. To compare the efficacy if IPTp -DHP or ISTp-DHP with RDTs in the 2 nd and 3 rd trimester with the current strategy of SSTp-DHP to reduce the risk of any malaria infection at delivery among women protected by long lasting insecticide treated nets (LLINs) in areas with relatively low P.falciparum and P.vivax transmission in eastern Indonesia. 2. To estimate the acceptability, feasibility and cost effectiveness of each of SSTp-DHP, ISTp-DHP and IPTp-DHP within the randomised control trial.

Initial sample size
The unit of randomisation is the clinic providing antenatal care (Puskesmas and Posyandu). The average number of women per clinic for the study period was estimated to be 58 (i.e. 29 pregnancies per year). We estimate that 41 of the 58 (70%) women would fulfil the eligibility criteria and provide informed consent, and that 80% of them (33) would contribute to the primary endpoint. The remaining 20% will be lost or have incomplete delivery data. The study was designed to detect at least a 50% reduction in malaria infection at delivery, from 10

New sample size (protocol v 3.0)
The trial in Sumba site was ended after recruitment of 989 women. New sample size calculations were conducted to determine the sample required for completion of the trial using recruitment in Timika site only. Power calculations were performed using Stata Metapower to estimate the sample size required for the trial to achieve at least 80% power using 10,000 simulations with the data from Sumba treated as 1 trial with 989 women, and data from Timika collected till January 2015 as another trial.
The Timika extension was simulated as a new trial. Blinded data pooled across the 3 arms for each site was used to obtain observed estimates of the pooled frequency of the primary endpoint and ICC value. The prevalence of the primary endpoints used in the new sample size calculation were 4.1% in Sumba and 24% in Timika. The observed ICC value was 0.0005. All analysis was conducted blinded. No interim analysis of the effect size was conducted.
A total of 2279 women; 989 women in Sumba (57 clusters) and 1290 in Timika (24 clusters) would achieve approximately 81% power to detect a 50% reduction in malaria infection from approximately 15% in the control arm to 7.5% in any of the intervention arms, using an alpha of 0.0167 to allow for 3 comparisons (compared to 0.025 in the original study which allowed for only 2 comparisons), and allowing for 13% efficiency loss due to varying cluster size and 20% loss to follow-up up. The new sample size would also have 87% power to detect a 50% reduction in Timika alone if the average prevalence of infection in the control arm is at least 24%.

Randomisation and allocation
The ANC clinics constituted the units of randomisation. A 1:1:1 allocation ratio was used. To minimize imbalances across treatment groups with respect to baseline malaria prevalence and risk factors for malaria, multivariate matching was used, based on malaria indicators available, such as the prevalence of positive RDTs or microscopy at antenatal visit in the 12-month period prior to the trial (ANC registry data), geographical area and clinic size (prior annual number of new ANC attendees); in this way, the 78 eligible clinics was blocked into 26 sets of 3 matched clusters.
The trial statistician at LSTM computer-generated lists of sets of triple-matched clusters and forwarded these to the trial site in Indonesia. The allocation of clusters to each of the three study arms were done as a public event. District Health Officials and village elders were asked to draw opaque sealed envelope from a box. Each sealed envelope contained the allocation, and after drawing the envelopes, they were opened and allocation recorded and study arm assigned. Signed envelopes containing the final list of clinic names and their allocation were sent to the trial statistician and a copy kept in the trial site in the TMF.
Minimization of selection and confounding bias is achieved through central block randomisation taking baseline data on malaria risk into account. The matched design with clinics as the unit of randomisation will minimize contamination between individual women and avoid allocation errors. The endpoints (malaria infection, birth weight, etc.) are measured at delivery, most of which take place in the health facility. The primary outcome, malaria infection is an objective verifiable measure, performed by laboratory staff unaware of the randomisation allocation.

Purpose of the analysis plan
The purpose of this document is to outline the statistical analysis plan for the STOPMiP trial. The primary objective of this trial was to compare the effectiveness in reducing malaria infection at delivery of IPTp-DHP or ISTp-DHP with RDTs in the second and third trimesters against the strategy of SSTp-DHP. The target study population was women protected by long lasting insecticide treated nets (LLIN) in areas with relatively low to moderate transmission of P.falciparum and P.vivax in Indonesia.
The SAP is based on the version of the amended protocol (V3.0 18June15) and approved by the Research Ethics Committees of the Liverpool School of Tropical Medicine (Sponsor) and the Eijkman Institute (collaborating institute and primary ethics committee in Indonesia). One interim analysis was planned half-way (when delivery numbers reached 50% of initial sample size) for assessing whether to stop the trial early (or one of the study arms) due to safety, efficacy or futility. With ending of recruitment in Sumba site before 50% deliveries of the total sample size was reached and recalculation of sample size for Timika site with single site option, the interim analysis stated in the protocol version 2.0 was dropped. A secondary analysis to compare the two interventions ISTp and IPTp was added. As the study design uses matched cluster randomisation methodology, all analyses will use GEE multilevel random effect log-binomial for binary and Poisson or negative binomial models for count data (primary outcome, incidence rates, event occurrences), with adjustment both for intra-cluster correlation at the level of randomisation (clinic) and for important covariates. Although the primary purpose of the matching in the randomisation is to optimise the balance between the study groups, covariate adjustment will be made for matching variables where appropriate (provided perfect matching was not achieved).

Malaria infection endpoint definitions
For all the definitions of malaria infection at delivery reference is made to any species of Plasmodium (falciparum, vivax, malariae, ovale, knowelsi, etc.).

Booking visit
1. Booking visit maternal malaria infection: infection detected in peripheral blood at the first antenatal visit, (yes/no) a. Standard microscopy b. PCR/LAMP c. Microscopy and PCR/LAMP d. RDT with fever/ history of fever • Excludes RDTs with no fever as this was a routine part of the SSTp and ISTp intervention but not for IPTp.
Antenatal (during pregnancy, after enrolment, before delivery) 2. Antenatal maternal malaria, (yes/no): any plasmodium detected in the peripheral blood of the mother prior to time of delivery by either a. Standard microscopy at all scheduled antenatal visits or unscheduled visits b. PCR/LAMP at all scheduled antenatal visits or unscheduled visits c. RDT at only unscheduled visits or at a scheduled visit with fever/ history of fever • Excludes all RDTs at scheduled visits and scheduled visits with no fever as this was a routine part of the SSTp and ISTp intervention but not for IPTp.
3. Antenatal 3 rd trimester peripheral malaria infection mother at last scheduled visit in the 3 rd trimester , (yes/no) a. Antenatal peripheral malaria infection mother detected during the last scheduled antenatal visit in the 3 rd trimester, before delivery. b. Otherwise the same diagnostic criteria are used as for antenatal peripheral malaria infection mother, described for antenatal malaria above a. Documented fever (>=37.5°C), or recent history of fever in the past 48 hours, or other symptoms of acute illness that resulted in a women seeking care or alerting the study team to request a home visit, and b. Maternal malaria patent infection detectable by Microscopy or RDT • Excludes immediate follow-up visits related to the primary episode; if not defined, use 14 days exclusion period for that endpoint 16. Asexual parasite density by microscopy a. Parasite density expressed per mm3, quantified against 300 leucocytes on assumed white blood cell count of 8000/mm3. b. The parasite density is defined by natural log transformation of the above count

Molecular definition of malaria infection detected by LAMP and PCR data
The study involves quantitative and nested PCR confirmation of all LAMP positive samples and a random sample of 5% of the LAMP negatives. The following algorithm will be used to define LAMP/PCR positivity (see Figure 1 Definition of PCR/LAMP positivity):

4.2.
Morbidity endpoint definitions 1. Birthweight a. Uncorrected birthweight (grams) (continuous) weight taken within 24 hours of birth using digital scales (precision +/-10grams) in live singleton babies. Birthweights taken more than 24 hours after delivery will not be considered because of the physiological fall in birth weight in breastfed infants occurring in the first days following delivery. [1,2] b. Corrected birthweight (grams) (continuous); weight taken within 7 days (168 hours) after birth in live singleton babies. Birthweights taken more than 24 hours after delivery will be corrected for the physiological fall in birth weight in breastfed infants occurring in the first days following delivery. [1,2] i. Birth weights taken 24-48h hours, and 48-168 hours after delivery will be corrected by a factor +2% and +4%, respectively to obtain the estimated weight at birth. [3,4] ii. Birth weights within 24 hours will not need to be corrected. 2. Newborn Gestational age (days) (continuous): derived gestational age at booking in days based on gestational age assessment methods at booking assessed in order of priority as follows: a. By gestational age from Ballard score estimated within 96 hours of delivery b. By Last Menstrual Period if known and if Ballard examination is not available c. By fundal height measurement if no other measure of gestational age is available.
• Excludes immediate follow-up visits related to the primary episode; if not defined, use 14 days' exclusion period for that endpoint • Note: these events are mutually exclusive of clinical malaria (i.e. all-cause sick-clinic visits minus sick clinic visits due to clinical malaria = non-malaria sick-clinic visits) • Delivery visits will be ignored in the clinical visits analysis hence ignoring the placental information 21. Non-malaria sick-clinic visits, infant (count) (same as for maternal) resulting in seeking care for an infant) 22. All-cause sick-clinic visits, maternal (count) • The sum of Clinical malaria and non-malaria sick-clinic visits with fever or history of fever in last 48 hours • Excludes immediate follow-up visits related to the primary episode; if not defined, use 14 days' exclusion period for that endpoint 23. All-cause sick-clinic visits, infant (count) • The sum of Clinical malaria and non-malaria sick-clinic visits • Excludes immediate follow-up visits related to the primary episode; if not defined, use 14 days' exclusion period for that endpoint 24. Maternal anaemia, (yes/no): Hb<11.0 g/dL (measured by HemoCue (Angelhom, Sweden), either venous or capillary blood). 25. Maternal Moderate to severe anaemia, (yes/no): Hb<9.0 g/dL (measured on Hemocue, either venous or capillary blood) (used to provide adequate power as Hb< 8 g/dL is rare, and to be in the midpoint between any anaemia (above) and severe anaemia (below) 26. Maternal severe anaemia, (yes/no): Hb<7.0 g/dL 27. Fetal anaemia, (yes/no): Hb<12,5 g/dL in umbilical cord blood at birth, which is 2 standard deviations below the mean cord Hb in developed countries [6] 28. Congenital malformations, (yes/no): Physical abnormality of live born baby detected at delivery or newly noted abnormality during the infant visits (7 days or 6-8 weeks post-natal). 29. Neonatal jaundice, (yes/no): Reported presence of jaundice in neonate within first seven days of life. : Treatment compliance will be defined as a percentage (total number of tablets taken/total number of tablets expected)*100, and divided into 3 equal groups (tertiles). 5. Treatment compliance (continuous): Treatment compliance will be defined as a percentage (total number of tablets taken/total number of tablets expected)*100, and treated as a continuous variable. 6. Regimen compliance will be defined as a percentage of the number of scheduled visits attended (total number of scheduled visits attended/total number of scheduled visits expected by gestational age at enrolment and delivery)*100, and then ranked into 5 equal groups (quintiles). a. Exclude visits that could not have occurred because the woman delivered before that scheduled visit date.

4.4.
Definitions for other variables 1. Season (tertiles): Each pregnancy will be defined to have occurred in the predominantly rainy vs dry season using rainfall data collected in the study area. This will be done by categorising the women into three equal groups based on the mean daily, weekly or monthly rainfall during the 6-month period prior to the date of delivery (i.e. during the 2nd and 3rd trimester of pregnancy). This can include rainfall data prior to her enrolment in the study. 2. Gravidity will be computed and triangulated from the various variables in the enrolment questionnaire and categorised into nominal (not ordinal) categorical variables. The nominal variable will be used because the relationship between gravidity and the primary outcomes is not linear. The following categories will be used: a. Gravidity by number, (G1, G2, G3, G4+): i. First pregnancy (G1) ii. Second pregnancy (G2) iii. Third pregnancy (G3) iv. Fourth pregnancy G4+ • Computed based on the combination of variables in the booking form ('Primi yes/no, gravid, previous livebirths, stillbirths and miscarriages). b. Pauci-Gravidae-2 (G1+G2) i. first and second pregnancies (G1-G2) c. Multigravidae (G3+) • Third or more pregnancies • if data is incomplete, this will be changed into (yes/no) primary school completed, junior high completed, senior high/academy/university completed. 4. Socio-Economic Status (SES), (quintiles): Categories will be based on the combination of ownership of household items, materials used for the floor, roof, walls of house and use of fuel, type of toilets and drinking water source in the socioeconomic CRFs and ranked according to World Bank wealth index score. 5. Study site: (Sumba, Timika) 6. Study clusters: will be based on the ANC of enrolment in each study site and not the place of delivery. 7. Place of residence (Urban, rural, not-known): will be based on information provided in the enrolment forms. 8. Place of delivery: is categorised to hospital, Puskesmas, home, private clinic and others (Pustu or Polindes or on the road/vehicle) 9. ITN use at enrolment: binary (yes/no): a single variable which takes into account the responses to this question in booking visit CRF 10. ITN use during pregnancy: binary (yes /no); a single variable which takes into account the responses to the question at scheduled visits, such as if a woman answers less than 50% of the time during pregnancy that she slept under a bednet the previous night than she is considered as a "non-user" vs. a "user" who slept under a bednet more or equal 50% of the time during pregnancy. 11. Beetlenut use: will be categorised into low, moderate or high by tertiles 12. Cigarette smoking: will be categorised to low, moderate or high using tertiles

Primary outcome
The primary endpoint will be the presence or absence of malarial infection (any species) at delivery (yes/no) and a composite of either 1. Placental malaria (placental blood or tissue) by microscopy or RDT or histology (acute/chronic) or PCR/LAMP, or 2. Maternal malaria (maternal blood) by microscopy or RDT or PCR/LAMP

Secondary efficacy outcomes
Antenatal (from 1 day after enrolment to 1 day prior to delivery) 1. Maternal malaria by PCR/LAMP (count)

Intention to treat analysis (ITT)
The unit of analysis will be individual women (participants). The primary analyses will be based on the ITT principle, so will include all randomised women not considered screening failures and for whom there is an outcome.

Per protocol Population (PP);
Per protocol population will be defined as: 1. All women not considered screening failure and received either: a. The study intervention and took all of the study doses on each occasion when measured; or b. An approved alternative treatment for symptomatic malaria according to protocol that replaced the need for the scheduled intervention; or c. Received the potential number of scheduled visits prior to delivery • Note: 'potential' visits implies that visits that were scheduled to occur after the observed delivery date are not considered as missing visits (i.e. a woman enrolled at 20 weeks and who came again at 24 and 28 weeks, but delivered at 30 weeks will fulfil the criteria of per protocol even though she will have missed the 32 and 36 weeks visits).

AND 2. Women who contributed information to the specific endpoint investigated
Women will be excluded from the per protocol population if they used prohibited medication.

Safety Population
All women who received at least one dose of study drug Eurartesim in IPTp arm or in ISTp or SSTp (if malaria-positive), and have completed sufficient follow-up to provide information on potential adverse events, defined as attendance of the next scheduled study visit from the last dose of investigational product received. We would separately account for women who received DHP under the national programme during unscheduled visits.

Reporting guidelines
We will follow the Consolidated Standards of Reporting Trials (CONSORT) 2010 statement; extension to cluster randomised trials guidelines for reporting of clinical trials (http://www.consortstatement.org/).

Data Pooling and standalone estimates
Effect estimates will be computed and presented as a single summary pooled estimate for both sites, with appropriate adjustment for site differences, and in addition for each site separately. All effect estimates will take the cluster design into account.
It is anticipated that the prevalence of the primary outcome may be low or even zero in some clusters, which could affect the ability of some analysis methods to converge (see also section 7.5.2 below). Should this happen, a cluster-level analysis will be performed using linear regression with weighting to account for varying cluster size. If this also fails, consideration will be given, as a last resort, to combining proximate clusters with similar geographical and demographic properties within the same study arm; any such combinations will be done incrementally to ensure that this process is minimised as far as possible.

Pre-scheduled stopping of study participants and use of data
In case the intervention was stopped before the pre-scheduled end, either by a decision of the study woman herself, or by the study team, and data was collected after stopping the intervention, the information will be included in the full analyses set.

Missing data
Missing data will be dealt with differently for the primary endpoint and the independent variables as follows.

Endpoints
Missing data on the primary and secondary endpoints will not be imputed.

Covariates
Missing values for covariates will be imputed for the covariate adjusted analysis of primary endpoint.
If the missing data for all pre-selected covariates is less than 5% of observations, missing values for these covariates will be imputed by means of multiple imputations (10 multiple datasets will be created) using the SAS procedure MI or similar procedures. Missing data will be assumed missing at random (MAR), (probability that an observation is missing can depend on the observed values of the individual, but not on the missing variable values of the individual). Imputations will be done on continuous as well as categorical variables. If categorical variables are created from continuous variables the imputations will be conducted on the continuous variable. We will first investigate the Missing Completely at Random (MCAR) assumption by modelling the probability of missing data on treatment assignment and other independent variables. If any of the independent variables are significant then missing data depends on covariates, a violation of MCAR. Then missing data will be assumed MAR. Results derived from multiple under MAR imputation and complete-cases analysis without multiple imputation will be compared in a sensitivity analyses. Models under Missing Not At Random assumption (selection and pattern mixture) will not be done. Focus will be on MAR assumption and how its violation can be investigated in a sensitivity analysis.

Linear regression analysis for adjusted analysis of dichotomous outcomes
Risk ratios or odds ratios?
Because the study outcome will be common in some strata the odds ratio is not likely to approximate the risk ratio and be further from 1 than the risk ratio (i.e. more extreme). Because risk ratios are easy to interpret and because odds ratios are sometimes misinterpreted as risk ratios, the study will use risk ratios as the measure of relative association for dichotomous outcomes to assist the public health interpretation of the findings.

Log binomial regression and alternative strategies in case of non-convergence
The primary linear regression analysis method to obtain risk ratios and corresponding 95% Confidence Intervals (CI) for dichotomous variables will be log binomial regression (PROC GENMOD in SAS, GLM in Stata). A well-known limitation of log-binomial regression is problems with convergence. If a model does not converge with the default syntax for log-binomial regression, we will use generalized linear models (GLM) with a log link or COPY-method. The advantage of COPY-method over 'robust Poisson method' is that it produces correct approximation of maximum likelihood estimates (MLE) and can be run using the existing PROC GENMOD procedure that had failed on the original data. There is a COPYmethod SAS MACRO for executing the method and will be used if there is a need to use COPY-Method. The MACRO first runs the PROC GENMOD procedure on the original dataset. If no convergence occurs, it automatically switches to the COPY-method and MLE set to 1000 copies. If problems of convergence are encountered with this method, Cheung's modified OLS method will also be attempted [7]; in addition, consideration will be given to using zero-inflated Poisson regression methods As stated above (section 7.2), if the convergence problems are identified as being caused by low or zero incidence of the outcome measure, as a last resort consideration will be given to combining and/or excluding proximate clusters with similar geographical and demographic properties within the same study arm.

Reporting conventions
Descriptive statistics Variables will be checked for the presence of outliers, using tabulation and box plots. Continuous variables with an approximately normal distribution will be summarised by their mean, standard deviation and skewed continuous variables by their median and the interquartile range (25 th percentile to 75 th percentile). Parasite densities will be log-transformed and expressed as the geometric mean (95% CI). Categorical variables will be summarised by their frequency and percentage.
Means, standard deviations and any other statistics other than quartiles will be reported to one decimal place greater than the original unit of measure. Quartiles, such as median, or minimum and maximum will use the same number of decimal places as the original data. Estimated parameters, not on the same scale as raw observations (e.g. regression coefficients) will be reported to three significant figures.

Measures of associations and P-value reporting
Analyses will be conducted at either the 5% or 2.5% significance level, as appropriate, allowing for multiple testing of two intervention arms compared to the control. Estimates and their 95% confidence intervals (CI) will be produced using SAS 9.3 or v 9.4 (version may change) or SPSS v 22 or Stata v13 or v14. We will also report p-values. P-values ≥0.01 will be reported to four decimal places in the analysis; p-values less than 0.0001 will be reported as '<0.0001', as per The Lancet's convention.

Participant disposition and Flow chart
A flow chart will be drawn up showing the number clusters allocated to each study arm per site and the number of women screened, enrolled, and followed-up in each study arm, and the number contributing to the primary analysis and per-protocol. The number screened and not enrolled and the reasons for non-enrolment will be reported, as well as the number and reasons of women who were lost for follow up, or who were withdrawn from study for safety reason or because of death.

Demographic, clinical and laboratory measures
All baseline characteristics will be summarised by intervention group and overall. No inference testing will be conducted on the baseline variables, but marked differences (e.g. >10% relative difference) will be noted and taken into account in the post-hoc multiple regression analyses.

Measures of Social Economic Status (SES) Asset index.
The educational, income and socio-economic status parameters will be summarised in table form. To develop a single measure of SES index Principal Component Analysis (PCA) will be used to generate scores for ranking. PCA is a multivariate data analysis technique and it will reduce the dimension of this pool of variables to a smaller set of principal components capturing as much information (variability) from the data as possible. The summary SES index will be added to the baseline table.

Binary outcomes
For binary endpoints, the following will be calculated: Unadjusted 3. P-value for crude mean difference Adjusted 4. Adjusted mean difference (95% CI) 5. P-value for adjusted mean difference

Forest plots of efficacy parameters
Results will be presented using forest plots for dichotomous and continuous variables. The graphics component will represent primary measure of association, i.e. the crude and adjusted Relative Risk (Reduction) and the adjusted and crude mean difference. In addition, columns with number of events and women per group, and Risk Difference (RD) (dichotomous variables) and the number of women, mean (SD) per group, and crude mean difference (95%) (Continuous variables) will be added. Forests plots will show results by site (Timika and Sumba), and overall (summary estimate stratified by site).

Crude and adjusted effect estimates
The primary analyses will be ITT using the full analytical population. The primary measures of association are the risk ratio (RR) (95% CI) between the two groups obtained using the generalized linear regression models with binomial distribution and log link function.
The primary efficacy endpoint will be any malarial infection at delivery as defined in outcomes.
Both the crude (unadjusted) risk ratio (RR) and the adjusted RR will be computed using the generalised linear regression model. In the first model the response variable is the primary endpoint variable (yes/no) and the independent variable is treatment group. In the second model, additional independent variables will be included to adjust for potential confounding (overall and stratified by gravidity). The independent variables for adjustment are given in Section 10.3.3. The cluster variables will be included in all models.

Adjustment for baseline independent variables in the multiple regression models
The aim of the modelling is to obtain a valid estimate of an exposure-disease relationship i.e. a valid measure of the treatment effect of ISTp-DHP or IPTp-DHP relative to the control arm, adjusted for confounding. We will use the same independent co-variates in each multivariate model to allow for consistency across the models. The variables will be categorised into groups as indicated below (section 10.3.3).

Variable specification
Variables that will be included will include variables that are likely to be prognostic for the primary outcome but are not in the causal pathway, as predefined on the basis of the literature, and variables that are possibly prognostic for the primary outcome. All analyses will include the variable for ANC as the cluster variable.
The following variables will be considered a priori:

Sensitivity analysis
Multiple Imputation for handling missing data in potential confounders The results of the estimate (95% CI) obtained from multiple imputation for the missing covariate data statistical models will be compared with the complete-case (i.e. participants with missing covariate data are excluded) estimate (95% CI). The primary analysis reported will be the complete-case estimate, irrespective of whether the MI and complete-case estimates differ. Nonetheless the differences will be explicitly explained.

Corrected birth weights
The results of the statistical models using uncorrected birthweight will be compared with the initial results using corrected birthweights in a sensitivity analyses. Three different sensitivity analyses will be conducted: 1 using birthweight collected within 24 hours of birth (these are all uncorrected by definition), and one using all birthweight collected within 1 week, but without correction, and one using uncorrected birthweights collected within 1 week, but using timing of measurement as covariate. In an event that there are differences between these results (e.g. >10% relative difference in effect estimate [e.g. RR 1.4 vs RR 1.6]) the results without correction will be taken as the final results. If the difference is <10% the corrected birthweight will be used (as this results in a bigger sample and minimizes the potential for overestimation of the frequency of small for gestational age). Any differences will be explicitly explained.

Effect stratified by gravidity
The primary efficacy outcomes will also be analysed stratified by gravidity (G1/2 vs G3+).

Outcomes at delivery
The secondary efficacy outcomes outlined in Section outcomes will be analysed using similar crude and adjusted analysis. For the modelling approaches, the same independent variables as identified in the models for the two primary endpoints will be used for adjustment. Results will be expressed identical to the methods described above for the primary outcomes.

Count data outcomes
For the secondary efficacy outcomes that are count of episodes during follow up, these outcomes will be analysed using Poisson regression with the time of follow up as an offset. The incidence rate ratio for the treatment group effect will be estimated and its 95% CI presented.
In an event of over dispersion, then the Negative binomial regression model will be fitted to the data instead of the Poisson regression model. In an event where the number of episodes is very small and there are lots of zero episodes then a log binomial model will be fitted to the data where the dependent variable will be defined as (0=no episodes, 1=one or more episodes). A zero-inflated Poisson regression model will also be fitted to the data in an event of a lot of zero adverse events. All these models will be compared using the Akaike Information Criteria (AIC). A model with a smaller AIC will be considered as the final model under these conditions. Both crude and adjusted analyses will be conducted similar to the primary endpoints. Variables considered for the full models will be the same as those for the primary endpoints.

Continuous outcomes
Similar considerations will be used for the assessment of continuous variables results expressed as mean differences (95% CIs) calculated by multiple linear regression models with independent variables as treatment group.

Safety analysis
For each safety outcome, the number of events and incidence of SAEs will be tabulated by system organ class, preferred term and by severity and causal relationship with the study drugs and compared between the three groups (IPTp-DHP versus ISTp or SSTp) using the appropriate count data regression model to estimate the treatment effect and its 95% confidence interval (CI) computed similar to the secondary efficacy outcomes.
Safety endpoint in the ISTP and SSTp arms will be reported by DHP exposure and non-exposure to DHP.

Further Adverse Events analysis.
All adverse events will be categorised as serious or non-serious. The total number of adverse events will be a count outcome and will be analysed using similar methods for count data as above. The independent variables in these analyses will be serious (yes/no) and treatment category, either: (1 = IPTp, 2= ISTp, 3= SSTp-arm and tested positive and received the drug, 4= ISTp; 5 =SSTp arm and tested negative and never received the drug).

Number of intervention visits
Because the study was designed to allow variation between the number of scheduled visits as a function of the gestational age at enrolment (4+ scheduled follow-up visits for women who were enrolled early in pregnancy and 3 for women enrolled later in pregnancy), we will explore the difference in treatment effect on the primary endpoint between 3 and 4+ scheduled visits among the per protocol population. This will be done by including interactions between the number of scheduled visits (3 vs 4+) and treatment group so that an estimate of the comparison between the two treatment groups is estimated for each of these two strata.

Compliance with study drug
For definitions of treatment and regimen compliance measures see Section 3.7, Definitions for other variables, page 14. Further exploratory analysis will be conducted of the distribution and impact of regimen compliance on the primary endpoints. This will be done by including interactions between regimen compliance and treatment group so that an estimate of the comparison between the two treatment groups is estimated at each level of compliance. This will be done using quintiles of the regimen compliance variable to explore the shape of the relationship, as well as a continuous variable (0 to 100%). This analysis will exclude women who delivered prior to their last scheduled pre-natal visit. We will look at determinants of compliance, including whether dose intolerance is a predictor of subsequent compliance.

Treatment response
The percent (%) of women who were parasitaemic at each visit and are still parasitaemic at the next visits (defined as within 63 days inclusive [i.e. 9 weeks, the time typically used in extended in-vivo tests) will be compared between women in IPTp-DHP and ISTp arm using survival analysis and the hazard ratio (95% CI CI) reported for 28 (+/-3 days) (about 1 visit later), 42 days (+/-3 days) and 63 days (=/-3 days; i.e. about 2 months). Only fully treatment adherent women of IPTp arm will be included in this analysis (for definition, see Section 3.7, Definitions for other variables, section 3.8 page 14). Bednet use at enrolment and at delivery, and IRS and place from where bednet received Table 2.3 Obstetric History Table 2.4 Information copied from ANC card Table 2.6 Medical history/ illness symptoms Table 2.7 Prior and concomitant medication Table 2.8 Physical examination Table 2.9 RDT result and drug intake Table 2.10 Social economic status Section 3 Medication Table 3.1 Prior medication Table 3.2 Concomitant medication Section 4 Follow-up information Newborn: Outcome of delivery, baby measurements, Physical abnormalities Table 5.3 Ballard score: neuromuscular score, physical maturity score, total score, gestational age Table 5.3 Delivery malaria smear test (maternal) Maternal, Placental (incision smear) and Cord blood smear results Table 7.4 Maternal, placental, cord blood PCR/LAMP Table 7.5

List of tables
Placental histopathology results

Section 8
Adverse events (note: all adverse events tables will be done for overall, by study arm and by exposure and non-exposure to study drug) Summary of serious adverse event by system organ class and preferred term: infants Table 8.2.4 Summary of adverse event by severity, system organ class and preferred term: infants Table 8.2.5 Summary of adverse event by causality, system organ class and preferred term: infants Section 9 Efficacy Table 9.1 Primary outcome analysis Table 9.2 Secondary efficacy outcomes Table 9.3 Secondary safety outcomes Table9. 4 Malaria infection endpoint definitions Table 9. 5 Morbidity endpoint definitions Section 10 Covariate adjusted analysis

Discussion
Limitations 20 Trial limitations, addressing sources of potential bias,