DECIDE: A cluster-randomized controlled trial to reduce unnecessary caesarean deliveries in Burkina Faso

Background: In Burkina Faso, facility-based caesarean delivery rates have markedly increased since the national subsidy policy for deliveries and emergency obstetric care was implemented in 2006. Effective and safe strategies are needed to prevent unnecessary caesarean deliveries. Methods: We conducted a cluster-randomized controlled trial of a multifaceted intervention at 22 referral hospitals in Burkina Faso. The evidence-based intervention was designed to promote the use of clinical algorithms for caesarean decision-making using in-site training, audits and feedback of caesarean indications and SMS reminders. The primary outcome was the change in the percentage of unnecessary caesarean deliveries. Unnecessary caesareans were defined on the basis of the literature review and expert consensus. Data were collected daily using a standardized questionnaire, in the same way at both the intervention and control hospitals. Caesareans were classified as necessary or unnecessary in the same way, in both arms of the trial using a standardized computer algorithm. Results: A total of 2138 and 2036 women who delivered by caesarean section were analysed in the pre and post-intervention periods, respectively. A significant reduction in the percentage of unnecessary caesarean deliveries was evident from the pre- to post-intervention period in the intervention group compared with the control group (18.96 to 6.56% and 18.27 to 23.30% in the intervention and control groups, respectively; odds ratio [OR] for incremental change over time, adjusted for hospital and patient characteristics, 0.22; 95% confidence interval [CI], 0.14 to 0.34; P < 0.001; adjusted risk difference, - 17.02%; 95% CI, - 19.20 to - 13.20%). The intervention did not significantly affect the rate of maternal death (0.75 to 0.19% and 0.92 to 0.40% in the intervention and control groups, respectively; adjusted OR 0.32; 95% CI 0.04 to 2.23; P = 0.253) or intrapartum-related neonatal death (4.95 to 6.32% and 5.80 to 4.29% in the intervention and control groups, respectively, adjusted OR 1.73; 95% CI 0.82 to 3.66; P = 0.149). The overall perinatal mortality data were not available. Conclusion: Promotion and training on clinical algorithms for decision-making, audit and feedback and SMS reminders reduced unnecessary caesarean deliveries, compared with usual care in a low-resource setting. Trial registration: The DECIDE trial is registered on the Current Controlled Trials website: ISRCTN48510263. We conducted the DECIDE trial (DECIsion for caesarean DElivery) at 22 public hospitals in Burkina Faso from May 2, 2014, to November 2, 2016. We included public hospitals with a functioning operating room, at least 200 caesarean sections performed in the year before the initiation of the study, no previously implemented audits of caesarean indications, and signed consent forms from the director of the hospital and the head of the maternity ward to participate. National academic hospitals were excluded because of the high number of junior clinicians in training (student midwives or doctors, interns or residents). All healthcare professionals involved in caesarean decision-making in the participating hospitals were included in the study. These included doctors (general practitioners and obstetrician-gynaecologists) and midwives. The first 100 women and their newborns who delivered by caesarean section in each participating hospital during the pre- and the post-intervention periods were included in the analysis regardless of the reason or timing of the caesarean. Women whose caesareans were performed in another hospital and who were subsequently transferred to a participating hospital were not included in the study. The DECIDE trial was a stratified, facility-based parallel cluster-randomized trial. To avoid contamination bias between clinicians in the same service, the unit of randomization and intervention was the hospital, while the unit of analysis was the women who delivered by caesarean section. Randomization was stratified according to three different types of hospitals: regional hospitals, district hospitals in the two largest cities (Ouagadougou and Bobo Dioulasso) and district hospitals outside those two cities. The study included a 6-month pre-intervention (baseline) period from May 2 to November 2, 2014, a 1-year intervention period from May 2, 2015, to April 30, 2016, and a 6-month post-intervention period from May 2 to November 2, 2016. After the baseline period, hospitals were randomly assigned to the intervention group or control group. All participating hospitals were randomly allocated simultaneously to minimize the risk of allocation bias. To avoid imbalance in the size of the two groups, we used computer-generated, blocked randomization within each stratum, with blocks consisting of four centres or, for strata with fewer than eight hospitals, two centres. Investigators were informed of the allocation just before the rollout of the intervention. Information on the women who underwent caesarean sections during the study was abstracted by trained midwives from hospital registers and medical records, whose quality and archiving were regularly monitored by the study coordinator. Data were collected daily using a standardized questionnaire, in the same way at both the intervention and control hospitals. Data completeness and quality were assessed during daily maternities’ staff meetings, through quarterly on-site visits and queries sent to on-site data collectors to resolve discrepancies identified by the data manager. Data collectors were aware of the randomization assignments but were not involved in outcome assessments. Access to the database was restricted to the data manager until the trial was completed. Evidence-based clinical algorithms were developed during the baseline period of the trial to help healthcare professionals in the caesarean decision-making process for the four main indications for caesarean reported by clinicians in Burkina Faso [18], namely, labour dystocia (obstructed/prolonged labour), foetal distress, previous caesarean section and pre-eclampsia/eclampsia, which represent 80.3% of all caesareans in study hospitals in the baseline period. We conducted a literature review to discern the diagnostic reasoning underlying evidence-based indications for caesareans (Additional file 1) and to generate a provisional list of good practice criteria, with preference given to evidence obtained through randomized controlled trials. The provisional list of criteria was sent to 16 international and national experts (gynaecologist-obstetricians, midwives and a public health physician) (Additional file 2), who gave their opinions on the relevance of each criterion and proposed others. The criteria retained were those validated by at least two thirds of the experts. We then developed clinical algorithms (for details, see Additional files 3, 4, 5, 6, and 7) for managing the four main indications for caesareans on the basis of these agreed criteria, as well as the corresponding definitions of unnecessary caesarean sections. The first 3 months of the 1-year intervention period focused on the training of healthcare professionals. The chiefs of the maternity units of the hospitals in the intervention group were trained to use these clinical algorithms (2-day training) and to conduct clinical audits of caesarean indications (1-day training). Subsequently, these trained chiefs set up audit committees in their own hospitals (which consisted of physicians and midwives) and trained all healthcare professionals to use the algorithms for caesarean decision-making. Algorithms were printed on posters and posted in the delivery room of each hospital in the intervention group. With a view to sustainability, no financial incentive was provided to the chiefs or healthcare professionals. The initial provider training, conducted in Ouagadougou, the capital city of Burkina Faso, was led by two experts of the Society of Gynaecologists and Obstetricians of Burkina (SOGOB). The training was based on the WHO guidelines for managing complications of pregnancy and childbirth [19] and for clinical auditing [20]. During the 9 months after the training period, iterative weekly SMS reminders for appropriate caesarean decision-making were sent to all healthcare professionals involved in caesarean decision-making in intervention hospitals (Additional file 8), and audits of caesarean indications were launched by audit committees with the support of one researcher (CK) during his quarterly educational outreach visits. Monthly audits were recommended, and each audit cycle included five standardized steps according to the approach proposed by the WHO [20] [1]: identification of women who had caesarean deliveries for the main indications addressed by the clinical algorithms during the previous month [2]; data collection regarding the management of labour and delivery on standardized forms [3]; assessment by the local audit committee, with the use of clinical algorithms, of the relevance of the indications for caesarean delivery [4]; formulation of recommendations for best practices and the evaluation of previous recommendations, both performed by the committee; and [5] provision of informal and formal feedback to healthcare professionals. During the 6-month post-intervention period, healthcare professionals in the intervention group were encouraged to continue performing clinical audits without supervision, to assess sustainability. No intervention was planned for the control group as part of this project. To assess contamination bias, we searched for any quality improvement programmes ongoing during the study period in the control hospitals that could impact caesarean rates. We also monitored staff turn-over and transfers between hospitals. We did not control or monitor if SMS reminders were shared or forwarded from staff in the intervention arm to staff in the control arm because this was not technically possible. The primary outcome was the percentage of unnecessary caesarean sections among all caesareans. Fifteen clinical categories of unnecessary caesareans were prespecified on the basis of the literature review and expert consensus (Table 1), grouped under the main four indications reported in the hospitals of Burkina Faso [18]. To avoid classification bias, caesarean sections were classified as necessary or unnecessary based on a standardized computer algorithm. This algorithm, developed as part of this study, was based on the established criteria (Table 1) and was applied to the database. Definition of unnecessary caesarean delivery 1Temperature of 37.5 °C 2Three contractions in 10 min, each contraction lasting more than 40 s 3Fetal heart rate (between 120 and 160 beats/min) without slowing 4Weight below the 10th percentile for gestational age 5Signs of severe pre-eclampsia: blood pressure ≥ 160/110 mmHg; albuminuria, with urinary albumin ≥ 3+ or ≥ 3 g/24 h; oliguria, with urinary flow < 30 mL/h; headache; epigastric pain; vision disorders; neurological disorders; seizures; haemolysis; low platelet count and high levels of liver enzymes Secondary outcomes included the percentage of unnecessary caesareans for each of the four indications; the relative contribution of each indication and each group of the Robson classification [3] to all caesarean sections performed; the percentage of caesarean sections performed before and after the onset of labour; the rates of intra-hospital maternal death among women delivering by caesarean section; intrapartum-related neonatal death (fresh stillbirths and immediate neonatal deaths before 24 h) among births by caesarean section; and quality caesarean decision-making score among healthcare professionals. Clinical decision-making competency and skills were evaluated using hypothetical patient vignettes framed around selected decisional algorithms (33 vignettes and 51 related questions) [12]. The results of this evaluation were used to generate a decision-making score for the main indications of caesarean section in Burkina Faso. During quarterly visits, CK conducted participant observations of audit committee meetings in intervention hospitals. The healthcare professionals’ views on the relevance of the caesarean indications were collected, as well as on the reasons for unnecessary caesarean sections and their recommendations for action. The sample size was calculated to maximize statistical power while minimizing the number of clusters [21]. To account for clustering by hospital, we assumed an intraclass correlation coefficient of 0.02, estimated based on the percentage of unnecessary caesareans in 10 hospitals in Burkina Faso [7]. We calculated that we would have to enrol 22 hospitals, with a total of 2200 women delivering by caesarean section each in the baseline and post-intervention period, for the study to have 80% power to detect a 50% relative reduction with the intervention in the percentage of unnecessary caesareans, assuming a baseline percentage of 25%, at a two-sided alpha significance level of 0.05. In the primary intention-to-treat analyses, the intervention effect on the primary outcome was estimated as the difference between the allocation groups in the change in individual women’s risk of unnecessary caesarean birth from the baseline to the post-intervention period. The binary individual-level outcome relied on the generalized estimating equation (GEE) extension of the logistic regression model, with an exchangeable covariance structure, to account for the clustering of women within hospitals [22]. Using the difference-in-differences approach [23], the additional reduction in the percentage of unnecessary caesareans in the intervention group, relative to the reduction in the control group, was estimated by the odds ratio (OR) with the 95% confidence interval (CI) for the interaction between indicators of the trial group (intervention vs control) and time (post-intervention vs baseline) from the GEE model. The GEE model-based two-sided Wald test of this interaction, at α = 0.05, was used to test the significance of the intervention effect. The same approach was used to assess the effect of the intervention on hospital-based maternal and intrapartum-related neonatal mortality. In the sensitivity analysis, for the primary outcome, we considered all caesarean deliveries for previous caesarean sections as non-avoidable caesareans because women may request an elective caesarean section to prevent maternal or perinatal poor outcomes. The GEE model for the primary outcome was adjusted for the stratification variable, namely, hospital type, as well as for variables selected a priori as potential risk factors for unnecessary caesareans [18], including (a) the baseline characteristics of the hospitals (systematic use of a partograph, the 24-h availability of laboratory tests and the 24-h availability of an anaesthetist), (b) the qualifications of the healthcare professional who decided on the caesarean section (qualifications) and (c) the characteristics of the individual women (spouse’s and woman’s occupations, spouse’s and woman’s education levels, lack of a prenatal visit, maternal age, referral from another healthcare facility and time of caesarean section). To assess whether the intervention effect varied according to hospital type, we tested the corresponding three-way interactions: hospital type × intervention × time at two-tailed α = 0.05. All secondary binary outcomes related to caesarean practice and case fatality were analysed using the same methods as those for the primary outcome. The GEE model was then adjusted for variables selected a priori as potential risk factors for intra-hospital maternal mortality, including (a) hospital baseline characteristics (24-h availability of laboratory tests and an anaesthetist), (b) qualification of the healthcare professional who performed the caesarean, and (c) women’s characteristics (residence, age, parity, previous caesarean delivery, any pathology during pregnancy, prenatal visit attendance, multiple pregnancy, referral from another health facility, caesarean performed before labour vs during labour). To assess the effect of the intervention on perinatal mortality, the model was adjusted for the same variables used for maternal mortality plus birth weight. To assess the effect of the intervention on clinical decision-making for caesarean section, quantified by the quality decision-making score, we adapted the difference-in-differences approach, described above for the primary outcome, to the analysis of a quantitative healthcare professional-level outcome. Specifically, for each score, we estimated the multivariable mixed linear model, with 218 healthcare professionals (123 at the pre-intervention period and 95 at the post-intervention period) as the units of the analysis. An exchangeable covariance structure was assumed to account for the correlation between the two complexity scores, namely, the baseline period and post-intervention period, within the same hospital. The multivariable mixed linear model was adjusted for the effects of the year, the randomization group and their interaction for the stratification variable. Statistical analyses were conducted by one of the co-authors (NC) who was unaware of the hospital assignments. All analyses were conducted using Stata version 12.0. as well as SAS, version 9.3, to check the accuracy. This study is registered with Current Controlled Trials, as number ISRCTN48510263. The WHO Statement proposes the use of the Robson classification as the global standard for assessing, monitoring and comparing caesarean rates within healthcare facilities over time, and between facilities [2]. In this paper, we report the contribution of each group of the Robson classification as a secondary outcome which was not stated in the protocol. Indeed, the 2015 WHO recommendation to use this classification was made after the writing of the protocol.