Effect of a novel vital sign device on maternal mortality and morbidity in low-resource settings: a pragmatic, stepped-wedge, cluster-randomised controlled trial

Background: In 2015, an estimated 303 000 women died in pregnancy and childbirth. Obstetric haemorrhage, sepsis, and hypertensive disorders of pregnancy account for more than 50% of maternal deaths worldwide. There are effective treatments for these pregnancy complications, but they require early detection by measurement of vital signs and timely administration to save lives. The primary aim of this trial was to determine whether implementation of the CRADLE Vital Sign Alert and an education package into community and facility maternity care in low-resource settings could reduce a composite of all-cause maternal mortality or major morbidity (eclampsia and hysterectomy). Methods: We did a pragmatic, stepped-wedge, cluster-randomised controlled trial in ten clusters across Africa, India, and Haiti, introducing the device into routine maternity care. Each cluster contained at least one secondary or tertiary hospital and their main referral facilities. Clusters crossed over from existing routine care to the CRADLE intervention in one of nine steps at 2-monthly intervals, with CRADLE devices replacing existing equipment at the randomly allocated timepoint. A computer-generated randomly allocated sequence determined the order in which the clusters received the intervention. Because of the nature of the intervention, this trial was not masked. Data were gathered monthly, with 20 time periods of 1 month. The primary composite outcome was at least one of eclampsia, emergency hysterectomy, and maternal death. This study is registered with the ISRCTN registry, number ISRCTN41244132. Findings: Between April 1, 2016, and Nov 30, 2017, among 536 223 deliveries, the primary outcome occurred in 4067 women, with 998 maternal deaths, 2692 eclampsia cases, and 681 hysterectomies. There was an 8% decrease in the primary outcome from 79·4 per 10 000 deliveries pre-intervention to 72·8 per 10 000 deliveries post-intervention (odds ratio [OR] 0·92, 95% CI 0·86–0·97; p=0·0056). After planned adjustments for variation in event rates between and within clusters over time, the unexpected degree of variability meant we were unable to judge the benefit or harms of the intervention (OR 1·22, 95% CI 0·73–2·06; p=0·45). Interpretation: There was an absolute 8% reduction in primary outcome during the trial, with no change in resources or staffing, but this reduction could not be directly attributed to the intervention due to variability. We encountered unanticipated methodological challenges with this trial design, which can provide valuable learning for future research and inform the trial design of future international stepped-wedge trials. Funding: Newton Fund Global Research Programme: UK Medical Research Council; Department of Biotechnology, Ministry of Science & Technology, Government of India; and UK Department of International Development. This pragmatic, stepped-wedge, cluster-randomised controlled trial evaluated the CRADLE VSA intervention in routine maternity care in low-resource settings. All methods were predefined and published,25 with no important methodological changes. Clusters were purposely selected to represent a range of low-resource settings. Ten clusters across eight countries were identified and agreed to participate: Addis Ababa in Ethiopia, Cap Haitien in Haiti, Freetown in Sierra Leone, Harare in Zimbabwe, Gokak in India, Kampala and Mbale in Uganda, Lusaka and Ndola in Zambia, and Zomba plus the Southern Region in Malawi. Each cluster comprised at least one urban or peri-urban secondary or tertiary health facility that provides comprehensive emergency obstetric care with multiple peripheral facilities that refer to the central hospital.25 Facilities were identified by the local primary investigators as the main facilities that refer to the central hospital within a feasible geographical area. Community health-care providers were included in implementation in clusters where they were supported at a district level and were active in routine maternity care provision (Ndola and Cap Haitien). Clusters crossed over from control to the CRADLE intervention in one of nine steps at 2-monthly intervals, with CRADLE devices replacing existing equipment at the randomly allocated timepoint. Ethics approval was granted by the King’s College London (UK) Research Ethics Subcommittee (LRS-14/15-1484) and in all countries before the start of the trial (appendix). Institutional-level consent on behalf of the cluster was obtained.28 See reference 25 for the protocol. All health-care providers working in the cluster facilities had access to the intervention. All women identified as pregnant or within 42 days of delivery, presenting for maternity care in a cluster facility or to community health-care providers, were exposed to the intervention. There were no exclusion criteria. The unit of randomisation was the cluster. A restricted method of randomisation was used such that there was zero rank correlation between events per month and order of randomisation to minimise imbalance between intervention and pre-intervention periods due to anticipated variation in the primary event rate between clusters. A computer-generated randomly allocated sequence run by the CRADLE statistician (PTS) determined the order in which the clusters received the intervention. All clusters were masked to the order until 2 months before receiving the intervention, when the next cluster to receive the intervention was informed. Because of the nature of the intervention, this trial was not masked. The two smallest clusters were randomised at the same time. Data were gathered monthly, with 20 time periods of 1 month. Before the intervention, various medical devices (where previously available) were used in routine maternity care, with management by local guidelines (pre-intervention period). The CRADLE VSA and training package was iteratively developed and piloted as previously described.25 At the randomly allocated timepoint, the training package was delivered in interactive group sessions to health-care providers from each of the clinical areas in the cluster facilities. Some health-care providers in each cluster became CRADLE Champions and provided ongoing training and support in their clinical areas. The local implementation team provided regular support to all facilities, with at least monthly contact. Existing equipment for measuring vital signs was usually removed from clinical use unless specific functions were needed (eg, repeated automated measures in a high-dependency unit). We did not include a transition period; outcomes occurring after implementation start were allocated to the intervention group (post-intervention period). After the end of the trial, clusters were able to continue using the intervention. Each cluster included primary (first point of access), secondary (first referral point), and tertiary facilities (specialty referral facility). Maternity unit staffing levels (doctors, nurses, midwives, clinical officers, and community health-care providers in Ndola and Cap Haitien, where active in routine care) and availability of key resources (magnesium sulphate, intensive care unit beds, and capacity for blood transfusion) were recorded throughout the trial period. Major changes to infrastructure, patient payment requirements, or environmental conditions were systematically evaluated each month in each site. Service impact was assessed by the proportion of women referred from peripheral facilities to higher-level care (collected for a 4-week period before and 3 months after implementation). Before the start of the trial, data collection methods were optimised based on existing resources available in each site. Outcomes were triangulated across multiple sources (including referral registers, ward registers, patient records, local mortality and morbidity records, and active case finding) to ensure data completeness. Source data consistency and quality were monitored by the local research team, with a proportion verified by the UK research team. There were no formal stopping rules. The trial ended after 20 months as planned. The primary outcome was a rate of a composite of maternal mortality or major morbidity: at least one of maternal death (all-cause mortality), eclampsia (occurrence of generalised convulsions with increased blood pressure in the absence of epilepsy or another condition predisposing to convulsions), or emergency hysterectomy (surgical removal of all or part of the uterus) per 10 000 deliveries per month, occurring during pregnancy, labour, or within 42 days of delivery. Predefined secondary maternal outcomes were the individual components of the primary outcome (eclampsia, emergency hysterectomy, and maternal death), intensive care unit admission (defined as admission or referral to a specific intensive care unit or equivalent defined highest-level care environment), and stroke. Secondary perinatal outcomes, collected per 1000 women with a primary outcome, were the rate of stillbirths and neonatal deaths. Neonatal death was defined as death of a live born infant within 28 days of delivery at or after 28 weeks of gestation. Stillbirth was defined as born without signs of life at or after 28 weeks of gestation. Because the intervention itself was not considered likely to lead to adverse events, and all major adverse pregnancy complications were included as outcomes, no additional adverse event reporting was undertaken. There were no changes to prespecified outcomes during recruitment, and only prespecified analyses were undertaken. Sample size estimation was informed by data from the feasibility phase and carried out using Hemming and Girling’s methods.29 Assuming 4000 deliveries per cluster per month, with nine clusters, each observed for 20 months, and a baseline event rate of 1% with a reduction to 0·75% post-intervention, 2450 outcome events were required to have power of 95% (selected to account for cluster variability in a stepped-wedge randomised controlled trial). A coefficient of variation of 0·4 (judged to be high, but plausible, based on our pilot data) and an intracluster correlation coefficient calculated as 0·0085 were selected for this study design. We did the planned comparison using risk ratios, but it did not converge for the majority of results. This finding is common in analyses of rates; therefore, results are reported as odds ratios (ORs) with 95% CIs. We used logistic regression with generalised estimating equations and a population-averaged model for the main analysis.30 For the primary analysis of the primary outcome and its individual components,31 we adjusted for three predictors: cluster (categorical), time from start of study (continuous; with an interaction between cluster and time so that each cluster had its own underlying time trend), and total time on the randomised intervention, with time before intervention given as zero (continuous). The analysis resulted in separate linear time trends in each cluster, as prespecified. The model for intervention analysis allowed for separate linear trends in each cluster before and after the intervention (or a change in trend at the time the intervention was introduced, known as linear splines or bent stick). This model was an amendment from the planned single linear trend in each cluster with a sudden change (or step) at the time of the intervention (no change in trend or slope, known as trend and step). We selected this alternative model because the unexpected variation led to greater instability in the analysis. The bent stick model achieved greater stability. Both analyses are presented and were adjusted for the same predictors. A significance level of 0·05 was used for all analyses. As was prespecified in the protocol, we also analysed the results for individual clusters using fixed linear trend and a step at the start of intervention because bent stick models were not reliable for clusters that implemented the intervention early or late and therefore had few data points with or without the intervention. Because individual patient data were only collected for women with a primary outcome, we had to treat all other women as having no event and therefore had no information from which to estimate the extent of missing data. The possible number of women who might have had a primary outcome without being recorded was not expected to have changed following the intervention. Sensitivity analysis removing data collected the week before and after implementation was originally intended to account for the learning phase following introduction of the intervention. However, this analysis proved impossible because only monthly delivery data could be collected. Analysis removing data collected at time periods during which there were major external influences was done as planned. Autoregressive correlation allows for decreasing correlations between observations over greater time periods; we did further analysis for alternative correlation structures as planned (appendix). We planned to adjust for any significant differences in the characteristics of clusters (number of facilities, obstetric resources, and personnel) before and after the intervention, but none were found. We calculated CIs using generalised estimating equations and robust standard errors adjusted for clustering. Statistical analyses were done in Stata, version 14.2 (by PTS). This study is registered with the ISRCTN registry, number ISRCTN41244132. The funders 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 the data in the study and had final responsibility for the decision to submit for publication.