Study design

A stepped-wedge cluster randomized controlled design was applied, with 12 public hospitals in Nepal as the clusters [34]. The study period followed the Nepali calendar and was initiated on April 14, 2017 (2073/1/1), and ran for 18 months up until October 17, 2018 (2075/6/31). Thus, according to published study protocol, the QI package was introduced at three hospitals at the same time with a three-month interval between each wedge [34]. Four of the hospitals were high-volume (>8,000 deliveries a year), four medium-volume (>3,000 deliveries a year), and the remaining four low-volume (>1,000 deliveries a year) hospitals, and each wedge included a hospital from these three categories. Each cluster had a control and an intervention period. The first three months of the study constituted a baseline period, when no intervention activities took place at any of the 12 hospitals (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Stepped-wedge design and time line of the NePeriQIP trial.

The period within bold lines represents the baseline period. The shaded period constitutes the intervention period. Each wedge contains three hospitals (clusters). NePeriQIP, Nepal Perinatal Quality Improvement Package.

https://doi.org/10.1371/journal.pmed.1002900.g001

Study setting

Hospitals were selected based on the criteria of having more than 1,000 deliveries per year and a referral center for maternal and newborn care. All the hospitals provided normal vaginal, assisted vaginal, and cesarean section delivery series. The small-sized hospitals (Bardiya, Pyuthan, Nuwakot, and Nawalparasi) did not have specialized sick newborn care service. The high-volume hospitals (Koshi Zonal, Bharatpur, Lumbini Zonal, and Bheri Zonal) provided specialized sick newborn care service. The medium-volume hospitals (Western Regional, Rapti Sub-Regional, Mid-Western Regional, and Seti Zonal) provided specialized newborn care services. All the hospitals, despite mostly being in the flatlands, were different in terms of service coverage and diverse in relation to ethnicity, language, and religion. Mid-Western Regional and Seti Zonal Hospitals were located in the most disadvantaged regions in Nepal in terms of literacy, access to services, and life expectancy. Bheri Zonal Hospital in Nepalgunj had a large number of minority (Muslim) communities, while in Bharatpur, some of the most fringe communities come for maternal and sick newborn services. The hospitals vary in terms of service delivery to ethnic minorities, population age groups, and cultural practices. The labor unit in each hospital was led by skilled birth attendants and had access to neonatal resuscitation services at birth. Pediatricians led these neonatal care units. In the low-volume hospitals, sick newborns were managed at the pediatric unit, which was led by medical doctors. During the baseline period, the intrapartum-related mortality rate ranged from 9 to 31 per 1,000 births at the 12 hospitals, with a mean of 13 per 1,000 births.

Participants

Women at 22 weeks of gestation or more admitted in the labor room, whose fetal heart sound was heard at the time of admission, were eligible for inclusion. Any women who had antepartum stillbirth or no fetal heart sound at admission were excluded.

Intervention

At each hospital, the QI package was introduced in a similar manner in the following steps. Firstly, an orientation of the hospital management committee was arranged in order to strengthen leadership and accountability for perinatal care (meso level). Together with the management, an assessment of service readiness and availability of intrapartum care, as well as a bottleneck analysis, was conducted. Based on the findings, a plan for improvement of quality of care was developed. Secondly, in-hospital facilitators were appointed by the management and trained to lead the QI process and facilitate delivery staff (meso- and microlevel) [30]. This was done through capacity building of the components of the QI process and based on the PDSA methodology to be used at biweekly peer review meetings [30]. Thirdly, health staff were introduced to the QI process, which consisted of (1) an introduction of and training in neonatal resuscitation standards and Essential Newborn Care (ENC) and (2) daily drills to practice neonatal resuscitation skills on a mannequin. Finally, a quality metric score card for the quality of neonatal resuscitation and outcome of care was established (microlevel). Training neonatal resuscitation and ENC was carried out by trainers who received training from trainers from the team from the American Academy of Pediatrics on basic neonatal resuscitation, Helping Babies Breathe (HBB 1.0) (S2 Text) [35].

Outcome variables

Power calculation

Power calculations based on an estimated primary outcome level of 20/1,000 births and an annual delivery rate of 60,000 at the 12 hospitals combined would allow us to demonstrate a significant reduction of intrapartum-related mortality of 14/1,000 births or more (alfa 0.05, beta 0.80). Power calculations were performed with R package cluster randomized trial (CRT)size. Reductions in intrapartum-related mortality at each of the 12 hospitals followed a before-after design, with variable length of control and intervention periods depending on the allocation to different wedges.

Randomization

Using block randomization, the high-, medium-, and low-volume hospitals were randomly allocated into one of the wedges. The last author generated the random sequence. Because this is implementation research, the allocation of treatment/intervention could not be concealed. The research and implementation team knew the outcome expected from the study. The randomization sequence was generated prior to the study start. The hospital did not have prior knowledge of when the intervention would be rolled out. Information on the rolling out of the intervention in the hospital was provided a month ahead of rollout.

The data collection and management team was not aware of when the hospital was in a control or intervention period. The independent data collection and management team was not engaged in the planning and implementation of the intervention.

Data collection and management

For the collection of data on primary and secondary outcomes, an independent data collection team was established at each hospital. In all hospitals, the mortality outcomes were collected from the women’s case notes, including the clinical events (mode of delivery, birth weight, gestational age, and neonatal resuscitation practices at birth). The demographic characteristics of the women (age, ethnicity, parity, and education) were collected through semi-structured exit interviews upon discharge. The data collection was carried out in a paper-based format. Following the completion of the forms, data collectors at each site sent the completed forms in a sealed envelope weekly via courier to the central research office in Kathmandu. At the central research office, the forms were rechecked for completeness and were entered into a Census and Survey Processing System (CS-Pro) database. All the entered forms were indexed according to respective hospital. The data cleaning of the data in the CS-Pro was done on a monthly basis. A standard operating protocol (SOP) was developed for data collection and management (S3 Text). A quarterly data quality assurance assessment was conducted by the central level investigation team to assess adherence to the SOP. Feedback was provided to the data collection team on adherence to the SOP.

Data analysis

Descriptive analyses of the demographic, obstetric, and neonatal characteristics of the population groups in control and intervention periods were performed to assess cluster variation. We calculated the proportion and CI of the characteristics of the populations in control and intervention periods using the STATA I/C 12.1. Using the software R (version 3.4.0), we analyzed change in mortality outcomes between the intervention and control periods by generalized linear mixed model (GLMM). Analyses were performed with R package lme4. Intra-cluster correlation coefficients (ICCs) were calculated for each outcome. For the subgroup analysis on the intrapartum-related mortality incidence change by wedge and size of hospitals during control and intervention period treatments, Pearson chi-squared test and binary logistic regression were used. Forward modeling was applied, adding obstetric and infant characteristic variables one at a time using Statistical Package for the Social Sciences (SPSS) 25.0. If the odds ratio (OR) was affected by 10% in either direction, the variable was maintained for the wedge and size of the hospital. Missing values in each variable were excluded from analysis. All analyses were carried out as per intention to treat (ITT).

Ethical considerations

Ethical approval was obtained from the Nepal Health Research Council (ref 26–2017). The delay in obtaining the approval of the QI package document led to a delay in clinical trial registration. The stepped-wedge design allowed all included hospitals to receive outcome information regarding the intervention, which has previously been proven beneficial. Written Informed consent was obtained from women upon admission and before the exit interview.

Methodological considerations

The study design, stepped-wedge cluster randomized trial, has many advantages. By mimicking a classical randomized control trial (RCT), this pragmatic design allowed all clusters to be exposed to the intervention by the end of the study. At the beginning of the study, there will be an initial period with no cluster exposed to intervention. At a three-month interval, one cluster was randomized to cross from control to intervention. At the end of the study, for a period all clusters were exposed to the intervention. [41,42,43]. The study design also allows for the practical implementation of the intervention that would have been very resource demanding if it had been carried out at the same time at all hospitals.

NePeriQIP was a hospital-based study in district and regional hospitals to improve the quality of neonatal resuscitation care. The selection of hospitals was representative of different levels of the public health system of the country. The range of annual delivery, human resources, infrastructure, and equipment strengthened the generalizability of results, both for similar settings in Nepal as well as in other low- and middle-income settings.

Another strength of the current study is the large sample size and the diverse settings it covers. Most importantly, by combining intrapartum stillbirths and first day mortality into the primary outcome of intrapartum-related mortality, we reduced the risk of misclassification. Early neonatal mortality risk was observed to be 11% lower in the intervention period than in the control period, but there was only moderate statistical evidence to conclude that the risk was truly different between the time periods.

There are, however, several limitations to the study that are worth mentioning. Although a large sample size of almost 90,000 women–infant pairs were enrolled in the study, the clustering reduces the power of the study. Consequently, we did not detect a reduction in first day mortality, as it was not sufficiently powered to detect the minimal change with the number of clusters provided. The delay in the start of intervention could have biased the results; however, intrapartum-related mortality subgroup analysis as per the protocol showed similar results (S1 Table, S2 Table, S3 Table). A stepped-wedge design involving clusters joining in a predetermined sequence, with each preceding cluster as intervention and the successive cluster as control, is subject to a threat of internal validity [44]. The successive clusters were exposed to the diffusion of treatment effect, as the intervention was widely disseminated in the national forum and the single hospital–based QI study had received widespread attention. Hospitals in successive clusters may also have engaged in competitive behavior and increased the effort to improve the resuscitation practice while waiting to start the intervention. Stepped-wedge trials (SWTs) randomly allocate clusters to control groups that cross over the intervention at different crossover points. The secular trend in the mortality outcome may impact the reporting of the outcome [45]. We reported the secular trend in Fig 2. To adjust the secular trend of mortality, time trends were entered into the model as fixed effects in GLMM, with the assumption that the trend was similar in all clusters.

The analysis for our main outcome is based on ITT. The intervention started at most hospitals, however, varied due to societal and organizational factors. At one hospital, there was a long and unforeseen strike, and factors like this paired with logistical considerations made it difficult to start at the same—and intended—dates of each wedge. These unforeseen events are, however, the nature of implementation research, and most likely, the effect of the intervention would have increased if all hospitals had received the intervention when intended. Finally, this was a large-scale trial recording a large number of deliveries over a relatively short period. A rigorous follow-up and control of data completeness were carried out continuously to minimize data loss. Data implausibility was detected at a late stage in relation to birth weights and gestational age at three hospitals. Data on these variables were thus discarded from these hospitals.

This study demonstrates a reduction in intrapartum-related mortality after implementation of a scaled-up QI intervention for neonatal resuscitation in public hospitals in Nepal. Using the learning from the implementation of the QI package, implementation of the package in other settings can have impact on intrapartum-related mortality. Further assessments of the sustainability of the QI process and improved quality of care in the hospital and labor units are a priority area of research.