Community-level interventions for pre-eclampsia (CLIP) in Mozambique: A cluster randomised controlled trial

Objectives: Pregnancy hypertension is the third leading cause of maternal mortality in Mozambique and contributes significantly to fetal and neonatal mortality. The objective of this trial was to assess whether task-sharing care might reduce adverse pregnancy outcomes related to delays in triage, transport, and treatment. Study design: The Mozambique Community-Level Interventions for Pre-eclampsia (CLIP) cluster randomised controlled trial (NCT01911494) recruited pregnant women in 12 administrative posts (clusters) in Maputo and Gaza Provinces. The CLIP intervention (6 clusters) consisted of community engagement, community health worker-provided mobile health-guided clinical assessment, initial treatment, and referral to facility either urgently (<4hrs) or non-urgently (<24hrs), dependent on algorithm-defined risk. Treatment effect was estimated by multi-level logistic regression modelling, adjusted for prognostically-significant baseline variables. Predefined secondary analyses included safety and evaluation of the intensity of CLIP contacts. Main outcome measures: 20% reduction in composite of maternal, fetal, and newborn mortality and major morbidity. Results: 15,013 women (15,123 pregnancies) were recruited in intervention (N = 7930; 2·0% loss to follow-up (LTFU)) and control (N = 7190; 2·8% LTFU) clusters. The primary outcome did not differ between intervention and control clusters (adjusted odds ratio (aOR) 1·31, 95% confidence interval (CI) [0·70, 2·48]; p = 0·40). Compared with intervention arm women without CLIP contacts, those with ≥8 contacts experienced fewer primary outcomes (aOR 0·79 (95% CI 0·63, 0·99); p = 0·041), primarily due to improved maternal outcomes (aOR 0·72 (95% CI 0·53, 0·97); p = 0·033). Interpretation: As generally implemented, the CLIP intervention did not improve pregnancy outcomes; community implementation of the WHO eight contact model may be beneficial. Funding: The University of British Columbia (PRE-EMPT), a grantee of the Bill & Melinda Gates Foundation (OPP1017337). The full protocol has been published (https://clinicaltrials.gov/ct2/show/{"type":"clinical-trial","attrs":{"text":"NCT01911494","term_id":"NCT01911494"}}NCT01911494; Appendix 1) and ethically approved (Centro de Investigação em Saúde da Manhiça (CISM, CIBS-CISM/038/14), Mozambique National Bioethic Committee (219/CNBS/14) and University of British Columbia (UBC, H12-03497)). The CLIP Mozambique trial was one of three independently powered cRCTs (others in Pakistan and India; all {"type":"clinical-trial","attrs":{"text":"NCT01911494","term_id":"NCT01911494"}}NCT01911494), and was conducted in Maputo and Gaza Provinces, southern Mozambique (Fig. 1) with average population per cluster of 24,526 inhabitants (range: 7,499–36,663) [14]. The socioeconomic status varies widely between and within clusters. Generally, households in Gaza possess more assets than those in Maputo province. In Ilha Josina and Calanga, nearly-half of the households are in the poorest wealth quintile [14]. Agriculture is the main occupation for women in this area [14]. Map of study area and enrollment by cluster. A formative 12-month feasibility study delineated baseline characteristics of the study area and barriers and facilitators related to implementation of the intervention [8], [9], [10], [14], [15]. The CLIP Mozambique Trial did not include a pilot phase. The unit of randomisation (cluster) was the administrative post and all villages and primary health centres (PHCs) within that post (minimum population of 25,000) were included in that cluster; in areas where population size was much lower than 25,000, two administrative posts were joined together to form one cluster. Twelve potential clusters were chosen by the local team, according to similar health care infrastructure, accessibility to the surveillance team, and the absence of conflicting concurrent research activity; four were from Maputo (3 de Fevereiro, Maluana/Maciana, Ilha Josina/Calanga, Magude) and eight from Gaza (Xilembene, Chicumbane, Nhancutse, Chibuto, Macia, Chissano, Mazivila, Messano) Provinces. Restricted, stratified randomisation was undertaken according to population size. There were 21 allocations that met these minimisation criteria; of these, one was selected using a random number sequence generator to assign the six intervention and six control clusters. Women of reproductive age were identified based on six-monthly household surveillance that followed written informed consent by the head-of-household. Pregnant women were eligible if they were aged 12–49 years. All participants provided written informed consent; if under 18 years, the participant provided written assent and her guardian, written consent. In intervention and control clusters, consent covered six-monthly household data collection. In intervention clusters, consent also included CHW-provided home visits (see ‘Interventions’). The surveillance team informed the cluster CHWs about eligible and consented pregnant women with registration data. All recruited participants were followed up to satisfy ethical requirements that women complete their postpartum visit. In the intervention arm (six clusters), the CLIP intervention package consisted of both community engagement and CHW-led visits with eligible women in their homes or PHC, with a focus on the detection and management of pregnancy hypertension (Figure S1 and Table S2). Community engagement was delivered from the time of cluster randomisation until the end of follow-up. Sessions were held with community leaders, women in the communities and their mothers, mothers–in-law, and husbands. Pregnant women and women of reproductive age were engaged during CHW home visits, nurse-led talks at PHCs and community meetings with community activists hired by the trial. Family members also participated in these meetings, usually husbands or partners, mothers-in-law or elders. Cluster-level mobilisers and the CLIP community liaison officer led events for community stakeholder groups (e.g., community leaders, small business entrepreneurs, and neighbourhood secretaries). Fifteen nurses, 48 CHWs, and three community activists were trained to deliver key messages addressing the ‘three delays’ in triage, transport, and treatment, including: (i) awareness of symptoms, signs, and potential consequences of pre-eclampsia and eclampsia; and (ii) birth preparedness and complication-readiness, including prior permissions for care-seeking, savings for obstetric emergencies, and transport planning, if needed. Activities were supported by culturally-appropriate pictograms developed with local input and describing maternal and perinatal risks associated with pregnancy hypertension. To address the specific transport problem, 77 meetings were held with 665 participants. A community transport scheme was designed to ensure adequate emergency obstetric care-preparedness among pregnant women and communities. A start-up fund was provided by CLIP, based on the anticipated local number of obstetric emergencies. The communities agreed to contribute funds to sustain the scheme post-trial. A committee assigned by the community and composed of three to four influential community members managed the funds. The CHW-provided contacts were recommended to occur every four-weeks (<28 weeks), fortnightly (28–35 weeks), weekly (36 weeks until birth), within 24 h of birth and on postpartum days 3, 7, and 14. CHW tasks were guided by the tablet-based PIERS On the Move (POM) mobile health (mHealth) application (app) [16], [17] that included miniPIERS (Pre-eclampsia Integrated Estimate of RiSk) time-of-disease risk stratification (Table S2) [18], [19] with pictograms as visual prompts. CHW-provided contacts were from enrolment until six weeks postpartum. There were 50 CHWs trained to provide the CLIP intervention. These included part of 48 new CHWs hired to address the low coverage of CHWs in the study areas (intervention and control); these hires were in collaboration with UNICEF and the Gaza and Maputo Provincial Directorates that agreed to sustain this workforce post-trial. The CHWs underwent 15 days’ didactic and participatory training in five modules, including: understanding pre-eclampsia and eclampsia and the CLIP protocol, conducting blood pressure (BP) measurement and using the CLIP POM tool, and undertaking skilled basic communication and obtaining informed consent. After study initiation, each CHW received on-the-job training and follow-up until all study procedures were understood and well-implemented, with monthly refresher training. In brief, the POM directed CHWs to first observe women to rule out emergency conditions (illustrated by pictograms) that would warrant immediate referral to facility. In the absence of emergency conditions, CHWs were directed to measure BP twice, using standardised methods and a semi-automated digital device validated for use in pregnancy (Microlife BP 3AS1-2) [20]; a third measurement was required if either the systolic or diastolic BP differed by more than 10 mmHg between the first two readings. Hypertension and severe hypertension were defined by systolic BP ≥140 mmHg and ≥160 mmHg, respectively; isolated diastolic hypertension did not elicit a response. The CLIP POM tool stratified women into one of three care pathways: usual ante-/postnatal care, non-urgent referral to a comprehensive emergency obstetric care (CEmOC) or higher facility within 24 h, or urgent referral to a CEmOC or higher facility within 4 h. In addition, women could be recommended to receive community-initiated oral antihypertensive therapy (methyldopa 750 mg for severe hypertension) or 10 g intramuscular magnesium sulphate (for evidence of complicated pre-eclampsia) (Figure S1 and Table S2). Women in control clusters received routine ANC provided at PHCs by nurses and doctors. The workforce was supplemented by an additional 87 CHWs in control clusters who received basic CHW, but neither BP measurement nor CLIP-specific, training. At both intervention and control cluster hospitals, facility enhancement (6 sessions) occurred to promote evidence-based care of hypertensive pregnant women. Adverse events were monitored in intervention and control clusters with special interest to relative maternal hypotension on arrival at facility (defined as sBP < 110 mmHg) after methyldopa administration in the community; injection site haematoma or infection, either respiratory depression, coma or death after magnesium sulphate administration in the community; transport-related injury (life or limb) or death during transport; ≥20% of women referred to facility being sent back to their communities without follow-up (monitoring community engagement and the CLIP POM). The adverse events are presented by treatment group: number of adverse events (overall and by type), number of women with one or more adverse event(s) (overall and by type). In both intervention and control clusters, the CISM Demography Department undertook six-monthly cross-sectional demographic surveys to identify eligible pregnant women and collect baseline and outcome data. In total, 12 field supervisors and 60 field workers received two-weeks’ training. The demography team undertook weekly field monitoring and supervision throughout the trial, and a web-based application (manhica-dbsync) was modified for quality control and data management between surveillance forms. Data collection tools were created iteratively with local investigators, derived from existing validated questionnaires where possible (e.g., WHO 2012 Verbal and Social Autopsy) [5], and translated to Portuguese. Initially women’s demographic characteristics, and past medical and obstetric history were collected, with subsequent six-monthly data collection on care-seeking during pregnancy and pregnancy outcomes. The mobile application synchronised with the CISM servers biweekly, facilitating transfer of information about deceased mothers and babies into the verbal autopsy instruments used by field officers for all maternal and perinatal deaths. Data were collected on electronic tablets (customised OpenHDS application) to give surveillance staff access to all contemporaneous trial data. Data management protocols ensured data security by encryption, data tracking through user identification numbers and audit trails, and effective data synchronisation between within-cluster devices, the CISM server, and the UBC REDCap (Vanderbilt University, USA) server. The primary outcome was a composite of all-cause maternal, fetal, and newborn mortality and major morbidity (defined in Panel). Mortality was assessed until 42 and 28 days after birth for mothers and offspring, respectively, and described per 1000 identified pregnancies. Maternal morbidity was not limited to those related to hypertensive pregnancy, as it was possible that the CLIP intervention might favourably influence outcomes in general. Neonatal morbidity reflected problems related to early delivery or delivery of a baby in poor health. All deaths, as well as survived morbidities, were reviewed by an independent panel (two obstetricians, two paediatricians, and one epidemiologist), masked to the cluster of origin, and excluding individuals who cared for the woman or baby under review. The major secondary outcomes were birth preparedness and complication readiness, delivery in facility able to provide emergency obstetric care, and proportion of facility births. Other outcomes included the impact of the intensity of contacts on the incidence of the primary outcome and its components, gestational age at birth, and mode of delivery. The requirement for 12 clusters over two years was estimated assuming an (i) annual birth rate of 40/1000/year in each cluster; (ii) a baseline incidence of our primary outcome (of one or more of maternal, fetal, or neonatal mortality or major morbidity) of 11·1% in intervention clusters; (iii) an intra-cluster co-efficient (ICC) of 0·002; (iv) a 20% reduction in all cause maternal, fetal, and newborn morbidity and mortality in intervention vs. control clusters; (v) an alpha of 0·05; (vi) 80% power and (vii) 10% loss to follow-up. The sample size was calculated using ‘Iceberg Sim’ software (version 2.0), based on simulations of 5000 Monte Carlo samples based on the input parameters. Power was calculated as the total number of the times within the 5000 samples there was a 20% difference in the primary composite outcome between intervention and control clusters, at p < 0.05. The data upon which the estimates were made were derived from the published rates or were provided by site investigators. All pregnancies, except those of women who withdrew, were included in our primary, intention-to-treat analyses. The unit of analysis was pregnancy, classified as ‘followed-up’ (complete postpartum trial surveillance), ‘lost-to-follow-up’ (estimated date of delivery [EDD] at ≥3 weeks before trial end but without follow-up data), and ‘still-on follow-up’ (EDD < 3 weeks before trial end). To mitigate potential bias due to differential loss-to and incomplete follow-up, the primary outcome of women who were lost-to, or still-on, follow-up was imputed ten times via multiple imputation by chain equations and Rubin’s rules [21]. Imputation models were based on all primary analysis adjustment factors (see below) and interactions between trial arm and enrolment date (accounting for possible lag in intervention effects). In each imputed data set, the adjusted odds ratio (OR) for the intervention effect was estimated using a multi-level logistic regression including a random intercept term for each cluster. To improve precision, models were adjusted for variables at individual (i.e., age, parity, maternal primary education, previous delivery locations, and husband’s primary education) and cluster-level (i.e., baseline neonatal mortality rate, CHW density, and population density). Sensitivity analysis including only complete cases was conducted with the same adjustment variables. Further, analogous multi-level logistic regression models were fit to assess the sensitivity of the primary result to various other factors, including adjustment, missing data for a component of the primary outcome (when others were documented), gestational age at birth, and postpartum follow-up to <42 days, as well as cluster-level aggregate analysis. Where sensitivity analyses included imputation, results were pooled (Rubin’s rules) [21]. In an additional planned secondary analysis, we explored within the intervention arm, whether there was a relationship between our primary outcome and the intensity of CLIP contacts, categorised as 0, 1–3, 4–7, or ≥8, to reflect prior and current WHO recommendations for the frequency of antenatal contacts [22]; to account for factors related to the number of POM-guided visit and confounders, the analysis was adjusted for maternal age, basic education, parity, enrolment timing in the trial, and distance from the household to facility. All analyses were repeated for each component of the primary outcome, albeit without imputation. Secondary and other outcomes were compared by baseline factor-adjusted multi-level models, as above. Statistical significance (two-sided) was set at p < 0·05 for the primary and p < 0·01 for other analyses, without adjustment for the interim analysis. R statistical software was used throughout. An interim analysis was undertaken once complete data were received for women making up half of the planned sample size and reviewed by the data safety and monitoring board (DSMB). The stopping rule for both benefit and harm required an observed difference between groups associated with an alpha < 0·001 (power 80%). The DSMB reviewed all reported adverse events for participant safety.