Menu
Menu
Menu
Menu
Objective: The aim of this study was to evaluate the effect of a midwife-performed checklist and limited obstetric ultrasound on sensitivity and positive predictive value for a composite outcome comprising multiple gestation, placenta praevia, oligohydramnios, preterm birth, malpresentation, abnormal foetal heart rate. Design: Quasi-experimental pre-post intervention study. Setting: Maternity unit at a district hospital in Eastern Uganda. Interventions: Interventions were implemented in a phased approach: standardised labour triage documentation (Phase 1), a triage checklist (Phase 2), and checklist plus limited obstetric ultrasound (Phase 3). Participants: Consenting women presenting to labour triage for admission after 28 weeks of gestation between February 2018 and June 2019 were eligible. Women not in labour or those requiring immediate care were excluded. 3,865 women and 3,937 newborns with similar sample sizes per phase were included in the analysis. Measurement and findings: Outcome data after birth were used to determine true presence of a complication, while intake and checklist data were used to inform diagnosis before birth. Compared to Phase 1, Phase 2 and 3 interventions improved sensitivity (Phase 1: 47%, Phase 2: 68.8%, Phase 3: 73.5%; p ≤ 0.001) and reduced positive predictive value (65.9%, 55%, 48.7%, p ≤ 0.001) for the composite outcome. No phase differences in adverse maternal or foetal outcomes were observed. Conclusion: Both a triage checklist and a checklist plus limited obstetric ultrasound improved accurate identification of cases with some increase in false positive diagnosis. These interventions may be beneficial in a resource-limited maternity triage setting to improve midwives’ diagnoses and clinical decision-making.
This quasi-experimental study compared the phased implementation of three interventions between February 2018 and June 2019: Phase 1 introduced standardised intake/outcome documentation; Phase 2 included standardised documentation plus a triage checklist; Phase 3 included standardised documentation, a triage checklist plus limited obstetric ultrasound. The phased-in design assessed the added value of incremental package components (Phase 2 and 3) compared to the referent period (Phase 1). The study was conducted at a district hospital in Eastern Uganda which has an estimated 7600 deliveries annually. It serves a catchment area of six districts (approximately 0.5 million people) and receives referrals from public health centres in the area. The maternity ward has 17 midwives on staff; each shift is staffed with approximately four midwives, one obstetrician/gynaecologist and one medical officer. The hospital’s Caesarean rate was 25% in 2017 according to District Health Information Software (DHIS-2) data (Uganda Ministry of Health, 2017). Our primary objective was to determine if a clinical assessment checklist alone (Phase 2) or checklist with limited obstetric ultrasound (Phase 3) at labour triage could improve diagnostic accuracy of a composite variable comprising preterm labour, multiple gestation, oligohydramnios, placenta praevia, malpresentation, and abnormal FHR. We also assessed any maternal condition (multiple gestation, oligohydramnios, placenta praevia and preterm birth) and any foetal condition (malpresentation, abnormal FHR). For diagnostic measures, we ascertained sensitivity, the proportion of outcome-defined cases who were identified at intake correctly, and positive predictive value (PPV), the proportion of those screened positive who were true cases. Outcome data collected after birth were used to classify the true presence of a complication, while intake and checklist data were used to inform intake diagnosis before birth. An example of correct identification is if twins were born and the mother was diagnosed with a multiple gestation on admission. Each intervention phase began with a training and pilot period. All clinical activities were conducted by midwives who were trained in study procedures since the goal was to determine if a checklist and/or limited obstetric ultrasound could be performed without the addition of extraneous personnel. In most cases, the midwife who conducted the clinical intake assessment was involved in the clinical decision-making for birth, consulting the hospital’s obstetrician/gynaecologist or medical officers when needed. Non-clinical activities, such as consenting procedures and data abstraction, were done by study-hired research nurses. Each shift was covered by at least one study-trained midwife and one research nurse. Phase 1 (February – May 2018) introduced an intake log with a single record for each eligible patient assessed in triage. This documented participant demographic information, gestational age at presentation, and clinical diagnoses made before decision to admit, send home or refer. Phase 1 captured baseline data and supplemented the existing intake system information. The intake log did not include any clinical prompts related to the conditions of interest, as this phase was meant to capture midwives’ clinical assessments using the existing standard of care. During a 5-day orientation workshop and piloting phase, three midwives and two research study nurses were trained on Phase 1 study procedures and protocols. Phase 2 (June – September 2018) introduced a triage checklist. This checklist was created by the study team which included two U.S.-based obstetricians and the study hospital’s obstetrician/ gynaecologist. Those trained during Phase 1 plus 2 additional midwives participated in a 5-day workshop and pilot period, after which revisions to the checklist were made prior to implementation. The checklist prompted specific assessments and associated reminders at the point of labour triage, such as identification of abnormal FHR, presentation, evidence of leaking membranes, maternal vital signs and gestational age assessment using last normal menstrual period (LNMP) and fundal height. It described cardinal clinical signs and symptoms to increase suspicion of high-risk conditions so that the midwife was guided to an appropriate management plan. Supplemental information contains the Phase 2 checklist, including the diagnostic criteria used to raise suspicion of these conditions. Midwives were asked to use the checklist when assessing the mother upon presentation, then complete the intake log to document diagnoses before birth. For Phase 3 (October 2018 – June 2019), following clinical assessments of the checklist, midwives additionally performed limited obstetric ultrasound to assess FHR, head position, placenta location, multiple gestation, deepest vertical fluid pocket, and biometry measures (biparietal diameter, head circumference, and femur length). The Phase 3 checklist, including the clinical and ultrasound diagnostic criteria related to the conditions of interest, is provided in Supplemental information. Midwives were asked to use the checklist and conduct the scan when assessing the mother, then complete the intake log to document intake diagnoses. Although ultrasound can be used to detect other conditions, such as foetal abnormalities, nuchal cord, foetal sex, etc., we focused on these measures because they can be identified after proper training and can assist in clinical decision-making during labour in resource-limited settings (Shah et al., 2011). As such, we refer to this triage scan protocol as limited obstetric ultrasound because it focused on a set of specific conditions that were of particular interest for this study. The ultrasound training curriculum was created by a team comprising in-country stakeholders and a U.S.-based expert in training ultrasonography in LMICs. A total of 7 midwives, 2 doctors and 2 research nurses received training over a 2-week period in October 2018. Robust quality assurance included 25 required proctored scans, requirement to pass an observed structured clinical exam, and all images produced 3 months post-training underwent blinded review to evaluate image quality and to flag common errors. These findings, as well as data regarding midwives’ skill acquisition, confidence and perceptions of the ultrasound training course, are described elsewhere (Shah et al., 2020). Because of software defects in the ultrasound machines initially procured (SonoScape A5), data collected from October to December 2018 were excluded from the analysis due to inaccurate image measurements and interpretation. New ultrasound machines (Mindray DP-10) and a 1-hour refresher course were implemented in January 2019. Outcomes after birth were determined in each phase by a standard outcome form, which relied on data abstraction from existing hospital data sources (the maternity register and patient medical charts). Midwives were also trained to complete a newborn assessment tool. For the conditions of interest, midwives used the following criteria to confirm presence of a complication at outcome: Multiple gestation was confirmed when more than one foetus was present. For preterm birth less than 37 weeks, midwives identify the infant in the maternity register as term or preterm based on standard practice clinical assessment. We also provided additional training to conduct the New Ballard exam and collect postnatal anthropometric measurements, such as birth weight, birth length and head circumference to help inform this designation (Ballard et al., 1991; Battaglia and Lubchenco, 1967). Providers reported oligohydramnios if there was reduced amniotic fluid at birth. Placenta praevia was diagnosed if vaginal bleeding or Caesarean section reported placenta praevia. Malpresentation was discerned if the presenting foetal part was non-cephalic (e.g. breech, transverse, oblique). Midwives assessed Apgar and breathing at birth as indications of probable abnormal FHR prior to birth. Intrauterine demise or intrapartum stillbirth was included in this outcome category because it was possible that a foetus under distress died in utero after intake assessment, for example, due to delayed care. Women who presented to the labour unit after 28 weeks of gestation with regular, intermittent cramping pain were eligible. We excluded women with conditions that required immediate clinical intervention, for whom there would be no time to complete the study interventions; for example, women with an eclamptic seizure, severe antepartum haemorrhage, or imminent birth. We also excluded mothers who were admitted but not in labour (e.g., malaria in pregnancy), and those who did not consent for participation. Using data extracted from the maternity register at the same hospital for a concurrent study20 from March 2016 to March 2017, we estimated that 20% of women presenting for labour had one or more of the conditions of interest. We conservatively assumed a complication prevalence of 17.5% and that 75% of the time, providers were already correctly identifying the conditions of interest. Using an α error of 0.05, 70% statistical power, and a 1-tailed test, to test a 25% increased rate of accurate detection compared to the baseline rate (i.e., 13.3% in Phase 1 vs. 16.4% in subsequent phases), we required 1225 births with known birth outcomes in each phase. We increased the sample to account for 5% missing data (e.g., missing outcomes, refusal to consent or data quality issues). A research nurse identified eligible women upon arrival to the maternity ward and obtained informed written consent. Information, including maternal and foetal diagnoses at intake, was recorded in the intake log across the phases. Participants were assessed at a single timepoint by study-trained midwives to determine maternal and foetal well-being. Triage assessment and interventions used were dependant on the phase in which the woman was enrolled. Specifically, during Phase 1, existing standard of care assessments were used; in Phase 2, the triage checklist was added to the triage assessment procedures; in Phase 3, the checklist with limited obstetric ultrasound scan was implemented. For outcome data for both singleton and multiple gestation births, a research nurse extracted data from the woman’s medical chart, the facility maternity register, and the newborn assessment tool. These outcome-related data sources were completed by a clinical provider. The intake log, checklists and outcome form were in English, while consent forms for women were translated into Lusoga, the local language. Data were collected on paper forms, kept in locked file cabinets in a secure location, and transferred via tablet to Open Data Kit (ODK), an open-source software designed for collecting and managing data in resource-limited contexts. Before data entry, study research nurses ensured consistent and non-duplicative study identification numbers. The study data manager also spot-checked data consistency between paper forms and ODK. Attainment of intended sample size was monitored by the study team during each phase. Specifically, every two weeks, the team counted the number of outcome forms that could be linked to intake data, as well as the woman’s consent form. Lastly, aggregate counts of facility admissions, deliveries by Caesarean, live births, birthweight <2500 g and stillbirths were abstracted monthly from the maternity register to better understand enrolment rates and facility trends. Data were converted into SPSS Version 25.02 (Armonk, NY: IBM Corp.) for cleaning, range and logic checks prior to analysis. All devices used for data entry or analyses were encrypted and password protected. All electronic data were maintained on secure systems with access limited to designated study staff, including the ODK server which is securely hosted by UC San Francisco. All individual-level data from the intake log, Phase 2 checklist, Phase 3 checklist and limited obstetric ultrasound, and outcome forms were linked through an individual study identification number and inpatient number. Our primary analysis was intent-to-treat (all women who consented at intake and had an outcome), as well as a per-protocol analysis (those who received all study components relevant to their study phase) for the composite variable (i.e. any condition). To evaluate the comparability of study participants' sociodemographic and reproductive health characteristics, bivariate analyses stratified by phase were conducted using chi-square tests (with Fisher's exact statistics when any cell had less than 5 observations) for categorical data and Student's t-test for continuous data. Logistic regression with robust variance estimation analyses with adjustment for maternal age, education level, fuel source, attendance to 4 ANC, gestational age at intake, nulliparity, history of Caesarean/stillbirth/neonatal death, and infant sex were conducted. To ensure that changes in diagnostic accuracy were not detrimental, we also evaluated phase differences in the rates of adverse birth outcomes, including 5-minute Apgar scores <7, intrapartum stillbirth, pre-discharge mortality and maternal mortality. The study was originally powered using a one-tailed test. However, because the effect on the primary outcome was considerably larger than expected, we conservatively used and present two-tailed test statistics. All participants provided written informed consent. Ethical approvals were obtained from the Institutional Review Board at the University of California San Francisco (#17-23310), the Higher Degrees, Research and Ethics Committee at Makerere University in Uganda (#515) and the Uganda National Council for Science and Technology (#HS 2347).
