Effect of continuum-of-care intervention package on improving contacts and quality of maternal and newborn healthcare in Ghana: A cluster randomised controlled trial

Objective To evaluate the effect of a continuum-of-care intervention package on adequate contacts of women and newborn with healthcare providers and their reception of high-quality care. Design Cluster randomised controlled trial. Setting 32 subdistricts in 3 rural sites in Ghana. Participants The baseline survey involved 1480 women who delivered before the trial, and the follow-up survey involved 1490 women who received maternal and newborn care during the trial. Interventions The intervention package included training healthcare providers, using an educational and recording tool named â € continuum-of-care card’, providing the first postnatal care (PNC) by retaining women and newborns at healthcare facility or home visit by healthcare providers. Outcome measures Adequate contacts were defined as at least four contacts during pregnancy, delivery with assistance of skilled healthcare providers at a healthcare facility and three timely contacts within 6 weeks postpartum. High-quality care was defined as receiving 6 care items for antenatal care (ANC), 3 for peripartum care (PPC) and 14 for PNC. Results The difference-in-difference method was used to assess the effects of the intervention on the study outcome. The percentage of adequate contacts with high-quality care in the intervention group in the follow-up survey and the adjusted difference-in-difference estimators were 12.6% and 2.2 (p=0.61) at ANC, 31.5% and 1.9 (p=0.73) at PPC and 33.7% and 12.3 (p=0.13) at PNC in the intention-To-Treat design, whereas 13.0% and 2.8 (p=0.54) at ANC, 34.2% and 2.7 (p=0.66) at PPC and 38.1% and 18.1 (p=0.02) at PNC in the per-protocol design that assigned the study sample by possession of the continuum-of-care card. Conclusions The interventions improved contacts with healthcare providers and quality of care during PNC. However, having adequate contact did not guarantee high-quality care. Maternal and newborn care in Ghana needs to improve its continuity and quality. Trial registration number ISRCTN90618993.â This was a cluster randomised controlled trial using the effectiveness-implementation hybrid design registered in ISRCTN (90618993).26 In order to enhance the generalisability of study findings in rural setting of Ghana, we selected three rural sites: Navrongo (northern), Kintampo (central) and Dodowa (southern) which had diverse socioeconomic and ecological background and health systems challenges. Additionally, these study sites had Health Research Centers (HRCs) under the Ghana Health Service, and these HRCs operated the Health and Demographic Surveillance System. Such research infrastructure could be beneficial for the quality control of the intervention and surveys. Each study site covered two districts and consisted of 36 subdistricts. We included 32 subdistricts in this study (Navrongo, 12; Kintampo, 12 and Dodowa, 8) and excluded four subdistricts because of other projects implemented or planned during our intervention period. We used subdistrict as a cluster unit as it was the primary unit of the health system. In the preintervention facility assessment, the percentage of healthcare facilities with at least one midwife was 47% in Navrongo, 36% in Dodowa and 21% in Kintampo. We made 16 pairs of the clusters (Navrongo, 6; Kintampo 6 and Dodowa, 4), taking into account the population, the volume of delivery and the number of midwives in each cluster. Then, we randomly assigned the clusters within a pair to the intervention or the control groups. A data analyst who was not a member of the study team randomly allocated the paired clusters using computer-generated random sequences. However, we assigned three clusters with a district hospital to the intervention group as majority of the childbirths took place in these hospitals. We informed about the implementation of the intervention to the community people and healthcare providers in the intervention group only. However, complete blinding was not feasible; we implemented the intervention in the intention-to-treat design, which did not control for women’s choice and access to healthcare facilities across a cluster boundary. Women who were aged between 15 and 49 years old and delivered between 1 October 2014 and 30 September 2015 in the intervention group were eligible for study enrolment.26 We implemented the intervention for 12 months (1 October 2014 to 30 September 2015) as initially planned in the protocol. The details of the intervention were described previously.26 Women were enrolled to the intervention when they had contacts with healthcare providers anytime from pregnancy to the postpartum period. The intervention package was composed of four interventions. First, healthcare providers underwent reorientation about CoC. Second, healthcare providers distributed the CoC card to women, which contains the schedule and actual dates of contacts with healthcare providers, information on essential care and birth preparedness and the presence of danger signs. Healthcare providers and women used the CoC card in every contact. Third, healthcare providers retained women and their newborns in the healthcare facility for the first 24 hours postpartum to provide the first PNC. Fourth, healthcare providers made home visits to provide PNC to women and their newborns within the first 48 hours if they missed the first postnatal contact by 24 hours postpartum. We emphasised to implement the intervention using the existing health systems and resource; all intervention facilities in the three sites had reorientation of healthcare providers and implemented all or a part of the intervention package depending on availability of resource and infrastructure. In addition, district health management teams conducted monthly supervision in healthcare facilities, monitored the performance of the interventions and had a monthly meeting to report the progress and discuss the challenges in collaboration with research teams. In the control group, women and their newborns received the standard care recommended by the Ghana National Safe Motherhood Service Protocol.27 During the trial period, we did not observe any harms or unintended events in the intervention or the control groups. We conducted the baseline survey from July to September 2014, with a sample of 1500 women who delivered between 1 September 2012 and 30 June 2014, and the follow-up survey was performed from October to December 2015, with a sample of 1500 women who received care during the intervention period. We calculated the required sample size based on an expected increase in four antenatal contacts from 86.6% to 95.0% according to the finding of our formative study.4 We considered a 95% CI, 80% power, an intraclass correlation coefficient of 0.02675 and 10% attrition in the sample size calculation.26 We performed two-stage random sampling to select 500 eligible women from each study site for the baseline and follow-up surveys. For the first stage, we defined subdistricts as a cluster unit. A subdistrict is composed of several administrative community units. We used the administrative community units as a primary sampling unit and randomly selected primary sampling units from each subdistrict that corresponded to the probability proportionate to the population. For the second stage, we randomly selected 10 women per primary sampling unit. Trained research assistants performed the survey by visiting the households of the eligible women who had no knowledge about the cluster allocation and conducting face-to-face interviews with them. The structured questionnaire included women’s sociodemographic characteristics; frequency and timing of contacts with healthcare providers; contents of care that women and their newborn received during ANC, PPC and PNC and whether they received the CoC card. The frequency and timing of contacts and contents of care corresponded to the recommendation of the Ghana National Safe Motherhood Service Protocol.27 We defined adequate contacts based on the frequency and timing of contacts with healthcare providers as follows: at least four contacts with healthcare providers during pregnancy, delivery with assistance of skilled healthcare providers at a healthcare facility and three contacts with healthcare providers within 48 hours, at 1 week (3–10 days) and at 6 weeks (36–48 days) postpartum (table 1). Definitions of the study outcome ANC, antenatal care; PNC, postnatal care; PPC, peripartum care. We measured the quality of care based on the contents of care received by the women and their newborns during ANC, PPC and PNC (table 1). The process-of-care dimension in Donabedian’s framework was employed.7 We created quality of care indexes that consisted of 6 care items for ANC, 3 for PPC and 14 for PNC. High-quality care was defined as receiving all care items during ANC, PPC and PNC. Having adequate contacts with healthcare providers and high-quality care was considered as the primary study outcome. The variable was composed of three categories: inadequate contacts regardless of care quality, adequate contacts with low-quality care and adequate contacts with high-quality care (ie, quality-adjusted adequate contacts). We considered the following variables as potential confounders: study site, living in a subdistrict with a district hospital, age, education, marital status, parity, religion, wealth quintile index and the status of national health insurance membership. Of these variables, age and parity were initially continuous variables and converted to categorical variables: age (≤19, 20–34 and 35–49), parity (primipara and multipara). The variable of wealth quintile index was generated by performing principal component analysis of 13 household assets. We calculated the distributions of the basic characteristics of the women, the percentage of each care item received by women and their newborns and the percentage of having adequate contacts with healthcare providers and high-quality care. We evaluated the effect of the intervention on adequate contacts with high-quality care during ANC, PPC and PNC. However, the effect of the intervention could be biased because of imbalanced cluster allocation; the effect could appear greater as three clusters with district hospitals were assigned to the intervention group. Moreover, women in the control group could access district hospitals in the intervention area, which in turn lead to a potential contamination that could make the effect of the intervention smaller. Thus, to control for these potential biases, we used the difference-in-difference (DiD) method with four groups including the intervention (n=863) and control (n=617) groups in the baseline survey and the intervention (n=870) and control (n=602) groups in the follow-up survey. Before performing the DiD analysis, we assessed two assumptions. First, no time-varying difference existed between the intervention and the control groups.28 We did not observe any specific changes that might have affected the study outcome in both groups during the trial period. Second, the outcome trend should be equal in the intervention and the control groups in the absence of the trial.28 However, it was not feasible to measure the change that could have occurred in the intervention group in the absence of the intervention because we did not conduct any surveys before the baseline survey. Therefore, we performed the DiD analysis with cluster robust estimators of variance, controlling for individual characteristics. Robust estimators of variance is a technique used to estimate cluster robust SEs and adjust the CIs of the DiD estimators when the regression model is potentially affected by cluster correlations.29 We also considered the potential effect of contaminations. Therefore, we calculated DiD estimators in the intention-to-treat and per-protocol designs separately. The intention-to-treat design focuses whether the intervention works in the real-world setting, which shows effectiveness of the intervention.30 In the intention-to-treat analysis, we compared the percentages of the study outcomes between the intervention and the control groups corresponded to the initial cluster allocation. The results could be affected by coverage and contamination of the intervention. The per-protocol design focuses whether the intervention works in the ideal setting, which shows the efficacy of the intervention.30 In the per-protocol analysis, we treated the possession of the CoC card as actual participation in the intervention. Thus, women in the intervention group who did not receive the CoC card and women in the control group who received the CoC card were excluded from the per-protocol analysis. Finally, we performed subgroup analyses to identify factors associated with having adequate contacts with high-quality care among women in the intervention group in the follow-up survey (n=870). This analysis focused on identifying the characteristics of women who had greater chances of having adequate contacts with high-quality care in the intervention area. We used multivariable logistic regression with cluster robust SEs. The independent variables were study site, living in a subdistrict with a district hospital, age, education, marital status, parity, religion, wealth quintiles and the status of national health insurance membership. We used Stata 13 (Stata Corp, College Station, Texas, USA) for the analyses. Participants and public were not involved in the design of, the recruitment to and conduct of the study because this was a randomised controlled trial. However, community people in the intervention group were announced about the EMBRACE project at the commencement of the trial. We obtained ethical approvals from Ghana Health Service, Navrongo HRC, Kintampo HRC, Dodowa HRC and The University of Tokyo. Consent was obtained from the local health authorities and community leaders prior to conducting the intervention study. We obtained oral informed consent from participants of the intervention, whereas we obtained written informed consent from participants of the surveys. For those who were aged under 18, we requested permission from their guardians and obtained their signature on the consent form.