The Community-Level Interventions for Pre-eclampsia (CLIP) cluster randomised trials in Mozambique, Pakistan, and India: an individual participant-level meta-analysis

Background: To overcome the three delays in triage, transport and treatment that underlie adverse pregnancy outcomes, we aimed to reduce all-cause adverse outcomes with community-level interventions targeting women with pregnancy hypertension in three low-income countries. Methods: In this individual participant-level meta-analysis, we de-identified and pooled data from the Community-Level Interventions for Pre-eclampsia (CLIP) cluster randomised controlled trials in Mozambique, Pakistan, and India, which were run in 2014–17. Consenting pregnant women, aged 12–49 years, were recruited in their homes. Clusters, defined by local administrative units, were randomly assigned (1:1) to intervention or control groups. The control groups continued local standard of care. The intervention comprised community engagement and existing community health worker-led mobile health-supported early detection, initial treatment, and hospital referral of women with hypertension. For this meta-analysis, as for the original studies, the primary outcome was a composite of maternal or perinatal outcome (either maternal, fetal, or neonatal death, or severe morbidity for the mother or baby), assessed by unmasked trial surveillance personnel. For this analysis, we included all consenting participants who were followed up with completed pregnancies at trial end. We analysed the outcome data with multilevel modelling and present data with the summary statistic of adjusted odds ratios (ORs) with 95% CIs (fixed effects for maternal age, parity, maternal education, and random effects for country and cluster). This meta-analysis is registered with PROSPERO, CRD42018102564. Findings: Overall, 44 clusters (69 330 pregnant women) were randomly assigned to intervention (22 clusters [36 008 pregnancies]) or control (22 clusters [33 322 pregnancies]) groups. 32 290 (89·7%) pregnancies in the intervention group and 29 698 (89·1%) in the control group were followed up successfully. Median maternal age of included women was 26 years (IQR 22–30). In the intervention clusters, 6990 group and 16 691 home-based community engagement sessions and 138 347 community health worker-led visits to 20 819 (57·8%) of 36 008 women (of whom 11 095 [53·3%] had a visit every 4 weeks) occurred. Blood pressure and dipstick proteinuria were assessed per protocol. Few women were eligible for methyldopa for severe hypertension (181 [1%] of 20 819) or intramuscular magnesium sulfate for pre-eclampsia (198 [1%]), of whom most accepted treatment (162 [89·5%] of 181 for severe hypertension and 133 [67·2%] of 198 for pre-eclampsia). 1255 (6%) were referred to a comprehensive emergency obstetric care facility, of whom 864 (82%) accepted the referral. The primary outcome was similar in the intervention (7871 [24%] of 32 290 pregnancies) and control clusters (6516 [22%] of 29 698; adjusted OR 1·17, 95% CI 0·90–1·51; p=0·24). No intervention-related serious adverse events occurred, and few adverse effects occurred after in-community treatment with methyldopa (one [2%] of 51; India only) and none occurred after in-community treatment with magnesium sulfate or during transport to facility. Interpretation: The CLIP intervention did not reduce adverse pregnancy outcomes. Future community-level interventions should expand the community health worker workforce, assess general (rather than condition-specific) messaging, and include health system strengthening. Funding: University of British Columbia, a grantee of the Bill & Melinda Gates Foundation. This individual participant-level meta-analysis included data from the CLIP cluster randomised controlled trials in India (run from November, 2014, to October, 2016), Pakistan (January, 2015, to December, 2016), and Mozambique (February, 2015, to February, 2017)7, 8, 9 as agreed by all CLIP principal investigators (ZAB, MBB, SSG, AAM, KM, RNQ, CS, ES) and detailed in the CLIP trial protocol. The included trials were cluster randomised controlled trials with 12 (India and Mozambique) or 20 clusters (Pakistan) comprising complex health system interventions. The Individual Patient Data (IPD) proposal was prospectively registered with PROSPERO, CRD42018102564, as were the individual CLIP trials ({“type”:”clinical-trial”,”attrs”:{“text”:”NCT01911494″,”term_id”:”NCT01911494″}}NCT01911494; the protocol is in the appendix [pp 12–120]). Ethical approval for the trials and individual participant-level data meta-analysis was granted by the University of British Columbia, Vancouver, BC, Canada (H12-03497) and each country’s relevant research ethics board (Aga Khan University, Pakistan, 2590-Obs-ERC-13; KLE University, India, MDC/IECHSR/2011-12/A-4, ICMR 5/7/859/12-RHN; and Centro de Investigação em Saúde da Manhiça [CIBS-CISM/038/14], Mozambique National Bioethic Committee [219/CNBS/14]). All eligible pregnant women provided written informed consent to participate. The units of randomisation (clusters) in India were primary health centres, in Pakistan were Union Councils (in Pakistan, provinces are divided into divisions, subdivisions [tehsil], and thereafter into Union Councils, comprising a large village and surrounding areas, often including nearby small villages) and all associated primary health centres, and in Mozambique were Administrative Posts (in Mozambique, provinces are divided into districts, which are further divided into Administrative Posts). Local teams chose potential clusters according to similar health-care infrastructure, accessibility for the surveillance team, and the absence of conflicting concurrent research activity. In India and Pakistan, internal pilot trials (each in an initial four clusters per country) preceded definitive trials. Pregnant women aged 15–49 years (12–49 years in Mozambique) were identified in their homes or local primary health centres by trained community health workers. Clusters were randomly assigned (1:1), via restricted, stratified randomisation according to population size, to either the intervention or control group. The trials were unmasked given the nature of the intervention. Participants in control clusters continued current local practice around antenatal care, referral to facilities, and initiation of therapy. The intervention aimed to address the so-called three delays in triage, transport, and treatment related to maternal mortality risk. The first step was community engagement involving women and their mothers, household male decision makers, mothers-in-law, and community leaders regarding pre-eclampsia awareness and education about birth preparedness and complication readiness, supported by culturally appropriate pictograms. Community group meetings were held in all countries, with additional one-on-one health-care worker-led meetings in women’s homes in clusters in Pakistan. In the second step, community health workers were trained to task-share pregnancy-hypertension-oriented care at CLIP visits in women’s homes, using the CLIP Pre-eclampsia Integrated Estimate of Risk (PIERS)-On-the-Move (POM) mHealth app for risk stratification.10 The team of community health workers already in place was trained to make the intervention scalable, if found to improve outcomes in a trial setting.5 Community health workers responded to emergency conditions (aided by country-specific pictograms), took women’s blood pressure using the Microlife BP 3AS1-2 device (Widnau, Switzerland) and assessed dipstick proteinuria at the first contact and any subsequent contact when hypertension was detected. Per-protocol, the minimum number of CLIP visits should have been at least every 4 weeks, with additional visits recommended at days 3, 7, and 14 after birth. Community health workers were directed by the POM app to either administer oral methyldopa 750 mg for blood pressure of 160/110 mm Hg or higher; administer intramuscular magnesium sulfate 10 g for suspected severe pre-eclampsia (miniPIERS11 risk for an adverse maternal outcome of at least 25%, severe systolic hypertension [at least 160 mm Hg], eclampsia, stroke, or vaginal bleeding); or refer the woman to a comprehensive emergency obstetric care facility for suspected pre-eclampsia or increased risk of stillbirth (4+ dipstick proteinuria value, or absent fetal movements for at least 12 h). In referral facilities shared by both intervention and control groups, evidence-based care was promoted through a small number of continuous professional development events (three in India, three in Pakistan, and six in Mozambique) spaced throughout the trial periods that focused on the WHO’s recommendations for prevention and treatment of pre-eclampsia and eclamspia.12 In intervention and control clusters, surveillance teams were trained to do regular cross-sectional surveys of households (every 3–6 months), except in India, where a prospective population-based surveillance system was established. After individual participant consent was obtained, data were collected on baseline individual-level and household-level information, antenatal care, and adverse maternal, fetal, and neonatal outcomes up to 6 weeks after birth (for the mother) or 28 days after birth (for the neonate). Women were defined as withdrawing from the trial if they declined further trial surveillance. Women were lost to follow-up if they were more than 6 weeks post partum (based on estimated delivery date) and more than one surveillance cycle from trial end. Women were defined as still on follow-up if they either had not delivered their baby or were post partum within 6 weeks of their estimated delivery date and within one surveillance cycle from the trial end. Overall coordination and data management was done by the Pre-eclampsia – Eclampsia Monitoring, Prevention and Treatment research group at The University of British Columbia.7, 8, 9 We assessed the risk of bias of each trial according to the five criteria for cluster randomised controlled trials in the Cochrane Handbook: recruitment bias, baseline imbalance, loss of clusters, incorrect analysis, and comparability with individualised randomised trials.13 Within Research Electronic Data Capture (REDCap; version 5, Vanderbilt University, Nashville, TN, USA) we extracted data, by trial group, for characteristics of: the trial, women enrolled, and the intervention, and outcomes. For this meta-analysis, as for the individual trials, the primary outcome was one or more of the maternal or perinatal mortality or morbidity outcomes, and the secondary outcomes were birth preparedness and complication readiness, the proportion of births that occurred in a facility, and delivery in a facility that is able to provide comprehensive emergency obstetric care (panel). Secondary outcomes included components of the primary outcome (including a composite of maternal mortality or morbidity, and a composite of stillbirth, neonatal death, or neonatal morbidity) and safety outcomes. Adverse events that we assessed were transport-related injury or death and infection-site haematoma or infection after intramuscular magnesium sulfate in the community. In India (given trial surveillance informed by facility data), we also included as part of this meta-analysis maternal systolic blood pressure of below 110 mm Hg on arrival at a facility after in-community methyldopa; respiratory depression, coma, or death during transport after in-community magnesium sulfate; and infection-site haematoma or infection after intramuscular magnesium sulfate in-community or at a facility. We were unable to follow up the details of a woman’s management and clinical courses after she was referred to facility. Serious adverse events were defined as being serious, unexpected (in nature, severity, or frequency), and thought to be related to the study intervention. Primary outcome The primary outcome is a combined maternal or perinatal outcome (either maternal, fetal, or neonatal death, or severe morbidity for the mother or baby). Maternal outcomes Maternal death was defined as the number of deaths during pregnancy or within 42 days of pregnancy (or last contact day if contact was not maintained to 42 days) per 1000 identified pregnancies, defined as the maternal death rate. Maternal morbidity was defined as the number of women with one or more life-threatening complication of pregnancy during pregnancy or within 42 days of pregnancy or last contact day if contact was not maintained to 42 day, per 1000 identified pregnancies. Complications of pre-eclampsia were defined as follows: Serious end-organ complications of pre-eclampsia Other major causes of maternal mortality Life-saving interventions Perinatal outcomes Perinatal and late neonatal death was defined as stillbirth (gestational age ≥20 weeks, ≥500 g, or both), early neonatal mortality (within 0–7 days of birth) and late neonatal mortality (within 8–28 days of birth) per 1000 identified pregnancies. Neonatal morbidity was defined as occurrence of a primary neonatal morbidity during 0–28 days of birth per 1000 identified pregnancies. Primary neonatal morbidities were: Secondary outcomes Birth preparedness and complications readiness: defined as having completed three or more of arranging for transport, obtaining prior permission for transport, saving money for obstetric care, identifying a skilled birth attendant, and identifying a facility for birth. In-facility birth, defined as any birth at a health-care facility. Birth at a comprehensive emergency obstetric care facility, defined as birth at any centre that provides basic functions and capability of doing a caesarean section, giving safe blood transfusions, and provision of care for sick and low-birthweight neonates, including resuscitation. For the primary individual participant-level data meta-analysis, we included all CLIP participants who were followed up with regards to the primary outcome of this meta-analysis. However, women who withdrew, were lost to follow-up, or still on follow-up at trial end were included in a sensitivity analysis in which we used the imputation from the primary trials in the individual participant-level data meta-analysis. Each trial was independent, no potential overlap existed between enrolled participants. With a planned sample size of approximately 60 600 women in 44 clusters, we would have 80% power to find a 20% reduction in the composite maternal and perinatal outcome, from a baseline of 10·2% and with an intraclass correlation coefficient (ICC) of 0·006. Furthermore, this sample size would give similar power to find a 20% reduction in maternal mortality or morbidity if the baseline rate were 1·7% and the ICC were less than 0·001 (appendix pp 122–54). A 20% reduction was chosen a priori as being clinically relevant by consensus within the CLIP group of experts and the research programme technical advisory group. We combined data from each trial dataset by study group. We summarised data as median (IQR) for continuous variables and n (%) for categorical variables. While in the primary trials we imputed outcomes for women with incomplete data, in this meta-analysis we only include data from women with complete follow-up.7, 8, 9 We assessed the treatment effect on the various outcomes using generalised mixed-effect models with random effects for both country and cluster, and fixed effects for the study group and baseline characteristics of maternal age, basic education, previous pregnancy (parity), and neonatal mortality rate from each country’s baseline survey as part of the feasibility studies. We used a one-stage approach, in which a single model is fitted directly using the results from each study, to make optimal use of data.14 The summary statistic was the adjusted odds ratio (OR; fixed effect for maternal age, parity [nulliparous vs parous], and maternal education, and random effects for country and cluster) with Wald-type 95% CI. We assessed between-trial heterogeneity using the τ2 (estimated as the variance term of the random effect for treatment in the mixed-effect model) and R2 statistics (as the ratio of the SEs of the treatment effect from a model with fixed slope and a model with a random slope).15 A τ2 close to 0 and R2 close to 1 were taken as indicating a lack of heterogeneity. Statistical significance (two-sided) was set at a p value of less than 0·05 for the composite maternal and perinatal outcome, and a p value of less than 0·001 for other analyses. We did four types of sensitivity analyses for the primary outcome and its components to assess the effect of potential sources of bias on our results. First, women who were defined as lost to follow-up or still on follow-up at the end of the trial were included to assess the potential effect of missingness. As for the primary analysis of the individual CLIP trials, mixed imputation was used to account for the risk associated with each woman, depending on her personal baseline characteristics, cluster characteristics, and time of enrolment relative to the beginning of the trial. Second, we restricted the adjusted analysis to women whose pregnancies continued to at least 20 weeks, because the average gestational age at recruitment was much earlier in India (approximately 11 weeks) than in Pakistan (approximately 21 weeks) and Mozambique (approximately 26 weeks).7, 8, 9 As for the individual CLIP trials and according to the statistical analysis plan (appendix pp 122–54), inclusion for this sensitivity analysis was: restricted to women who had 42 days of post-partum follow-up data; restricted to women with anticipated birth (according to estimated delivery date) or anticipated birth and 42 day of post-partum follow-up data within the trial timeline, to assess the effect of the intervention independent of gestational age at birth; and expanded to include women who were enrolled into trial surveillance only post partum, which might have reflected an effect of community engagement but was a protocol deviation. Third, we did an unadjusted analysis without accounting for baseline individual-level and cluster-level characteristics. Fourth, we did a so-called on-treatment analysis of women who received at least one community health worker-led POM-guided visit in intervention versus control clusters. In an additional planned secondary analysis, we explored in the intervention group whether an association existed between our primary outcome and the number of CLIP visits, measured as 0, 1–3, 4–7, or 8 or more visits, to reflect previous and current WHO recommendations for the frequency of antenatal care contacts.16 To account for factors related to the number of POM-guided visits and confounders, the analysis was restricted to women whose pregnancies continued beyond 20 weeks and was adjusted for maternal age, basic education, parity, time of enrolment in the trial, and distance from the household to a facility. We used R statistical software (version 3.5.2) for all analyses. The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all data in the study and final responsibility for the decision to submit for publication.