Effect of a labor triage checklist and ultrasound on obstetric referral at three primary health centers in Eastern Uganda

Abstract Objective To test whether introduction of a midwife‐performed triage checklist and focused ultrasound improves diagnosis and referral for obstetric conditions, including multiple gestation, placenta previa, oligohydramnios, preterm birth, malpresentation, and abnormal fetal heart rate. Methods We implemented an intake log (Phase 1), a checklist (Phase 2), and a checklist plus ultrasound scan (Phase 3) at three primary health centers in Eastern Uganda for women presenting in labor. Intake diagnoses, referral status, and delivery outcomes were assessed, as well as sensitivity and positive predictive value (PPV). Results Between February 2018 and July 2019, 1155, 961, and 603 women were enrolled across the three phases (n=2719); 2339 had outcome data. Incidence of any outcome‐confirmed condition was 8.8%, 7.9%, and 7.1% (P=0.526) for each phase, respectively. The proportion of referred women with a condition did not change between Phases 1 and 2 (7.8% versus 8.6%, P=0.855), but increased in Phase 3 (48.4%, P<0.001). Sensitivity improved with each intervention; PPV decreased with ultrasound. Conclusion Use of ultrasound plus checklist increased referrals and sensitivity for high‐risk conditions, with decreased PPV. The checklist alone improved correct diagnosis, but not referral. Further evaluation of these triage interventions to maximize diagnostic accuracy, referral decisions, and outcomes are warranted.


| INTRODUC TI ON
Nearly three-quarters of the estimated 295 000 annual maternal deaths are due to direct obstetric complications and half of 2.6 million third-trimester stillbirths occur during labor and delivery. 1,2 Improved quality of care, particularly during the intrapartum period, can reduce preventable maternal and perinatal mortality. In many low-and middle-income countries, failure to receive adequate care when a facility is reached, the third delay, 3 is exacerbated by lack of supplies, personnel shortages, long waiting times, and weak referral protocols. 4 Effective referral from primary health centers (PHCs) to hospitals offering comprehensive emergency obstetric and neonatal care is a critical component of quality maternity care. 5 Inter-facility referral relies on a confluence of factors, such as timely arrival of a woman in labor, appropriate identification of high-risk conditions by providers, emergency transportation, and communication between the referring and receiving facilities. 6 However, inconsistent understanding of clinical criteria for referral, guideline non-compliance, inadequate clinical skills, lack of confidence in decision-making, and absence of transportation remain critical gaps at PHCs. [7][8][9] These barriers are exacerbated by weak communication across the health system, and further delays when a referral hospital is reached.
Labor triage, when a woman first presents to the PHC maternity unit in labor, is an opportunity to screen for conditions that may warrant referral, especially in contexts where prenatal care screening is poor. Although the WHO Safe Childbirth Checklist begins with "Does mother need referral?", standardized assessments before decision to admit, refer, or send home are not included. 10 Triage practices at higher-level referral facilities in resource constrained settings have been explored, such as implementation of a Traffic Light System, interactive training programs, and standardized documentation 11,12 ; however, these interventions have not been introduced at the PHC level.
This study aimed to evaluate if triage interventions-a checklist, focused ultrasound scan and referral transportation support-improved the ability of PHC midwives to correctly diagnose high-risk conditions and appropriately initiate referral to the district hospital (DH).

| MATERIAL S AND ME THODS
We examined the phased implementation of triage interventions at three PHCs in Busoga region, Uganda, between February 2018 and July 2019. In 2016, 60% of Ugandan women received four prenatal care visits and the median length of pregnancy at entry to prenatal care was 4.7 months. 13 The three study PHCs provide 24-hour delivery services without cesarean delivery capacity, conduct 60-75 monthly deliveries, and are located 11, 25, and 41 km from the DH. On average, each PHC has five midwives on staff with each shift covered by one midwife in the labor room and another in prenatal care. The standard of care guidance is to refer to higher care for the conditions of interest (detailed below) unless delivery is imminent. Other conditions including obstructed/prolonged labor, previous cesarean section, pre-eclamptic toxemia, and antepartum hemorrhage, also warrant referral.
Ambulances are accessible, but patients pay money for fuel or rely on their own means to reach the DH. Ultrasounds were not available before the study.
The study's primary outcome was the proportion of women with one or more of six high-risk conditions confirmed at birth who were referred upon initial PHC presentation. The conditions (preterm birth, multiple gestation, oligohydramnios, placenta previa, malpresentation, and abnormal fetal heart rate) were combined into one composite variable for the primary analysis. The following criteria were used to confirm presence of a complication at outcome: multiple gestation, more than one fetus present; preterm birth, gestational age by Ballard examination; oligohydramnios, reduced amniotic fluid at birth without rupture of membranes; placenta previa, if reported by vaginal examination or cesarean section; malpresentation, non-cephalic presenting fetal part; abnormal fetal heart rate, 1-minute Apgar scores less than 7 or infant born without signs of life.
The study interventions are described in Table 1. The study evaluated the effect of Phase 2 and Phase 3 interventions on the primary T A B L E 1 Description of triage interventions introduced at the three primary health centers (PHCs) during each study phase.

Intervention description
How it was implemented Phases Standardized documentation at intake and outcome The checklists used in Phases 2 and 3 are provided in Appendix S1. outcome using Phase 1 as the baseline comparison. Phase 1 introduced a triage intake log and outcome form. In Phase 2, standardized documentation was supplemented with a triage checklist and referral support. Ultrasound (Mindray DP-10, Mindray, Shenzhen, China) was added in Phase 3. Phase 2 and Phase 3 checklists are provided in the Appendix S1. Documentation, checklist, and ultrasound were also introduced at the referral DH, as part of a concurrent study that will be described elsewhere. The ultrasound curriulum and quality assurance activites are published elsewhere. 14 Women who presented with labor-like pains after 28 weeks of pregnancy were eligible. Women were excluded if they were not in labor or required immediate intervention, such as those with severe antepartum hemorrhage, eclamptic seizure, or imminent delivery.
We designed a balanced study with an equal number of women per phase. We estimated that 4% of all parturient women were referred for one of the six conditions based on baseline assessment of register data. Given a two-tailed test, α of 0.05, 80% power, and a relative effect of 100% (from 4% to 8%), the study required 601 women across the three PHCs per phase (Fleiss continuity correction applied). The sample size was increased by 20%, to 721 per phase, to account for loss to follow up, refusal, and missing data.
For each phase, three midwives and one study research nurse from each PHC were trained in study procedures. One study-trained midwife covered each shift with support from the research nurse.
Tools were piloted and revised before implementation.
Study-trained midwives filled out paper-based study tools and related clinical data sources (i.e., medical charts, register).
Research nurses identified eligible women, obtained informed written consent, and entered data using tablets into Open Data Kit.
They verified data completeness and consistency before entering data electronically. The study data manager performed biweekly data quality spot-checks, transferred data to a secure server, and obtained monthly counts of admissions and deliveries to estimate enrollment rates.
Paper forms were kept in secure cabinets. All devices were encrypted and password protected, and all electronic data were kept on secure systems. Data access was limited to designated study staff, including the Open Data Kit server, which was hosted by University The incidence of any maternal or fetal condition, as defined by post-delivery outcomes, was 8.8%, 7.9%, and 7.1% across phases (P=0.526, Table 3). Preterm birth, malpresentation, and abnormal fetal heart rate were most common. We observed no differences in adverse maternal or newborn outcomes between phases though the numbers were small and the study was not powered to assess this (Table S1).
Among women who had any outcome-confirmed maternal or fetal condition, the proportion who were referred or re-admitted as the result of failed referral in Phases 1 and 2 were similar (7.8% versus 8.6%, P=0.855, Fig. 2  respectively. Hence, checklists may be useful cognitive aids to standardize triage practices. 19 Although our findings suggest that focused, midwife-performed ultrasound at PHC triage can increase appropriate inter-facility referral for complicated cases, several limitations exist. First, follow up among those referred varied across phases, reducing complete assessment of the primary outcome, appropriate referral for correctly identified conditions. As the number of referrals with known outcomes was smaller in Phase 1 than subsequent phases, the baseline rate could have been higher. Moreover, for Phases 2 and 3, if condition-related referrals with missed outcome had the condition confirmed (true positive), the effect would be larger; conversely, incorrect diagnosis at outcome (false positive) would have decreased the effect. We attempted to link referrals from the PHCs to the DH and identify final diagnoses from maternity registers, but the data were not collected in the registers in a standardized manner. Additionally, many referred women self-reported going to private facilities or traditional birth attendants, which hindered follow up and acquisition of provider-confirmed delivery data. Improved capacity for follow up within the health system is needed to better understand maternal and perinatal mortality and morbidity data, and opportunities for intervention.
Other study limitations include potential suspicion bias, given that the midwives identified both intake and outcome diagnoses, and inaccuracy of the outcome diagnoses used to determine sensitivity and PPV. Under-reporting of the conditions at outcome, whether due to misdiagnosis or data quality, could have compromised the study results. To this end, we conducted post-hoc analyses excluding oligohydramnios from the primary and secondary outcomes, given the TA B L E 4 Sensitivity and positive predictive value for conditions of interest among all women who have an outcome (N=2339), inclusive of those admitted/delivered at the primary health center (n=2271) and those referred and have outcome (n=68). subjective nature of oligohydramnios assessment post-delivery without the use of ultrasound, 20,21 as well as the inconsistent assessment or documentation of premature rupture of membranes. Trends in referral remained the same upon exclusion of oligohydramnios (Table   S5A), but Phase 3 PPV was no longer significantly lower for any maternal or fetal condition compared to Phase 1 when oligohydramnios was excluded ( Nonetheless, through a phased intervention approach, we assessed multiple triage interventions with robust statistical methods.
Logistic regression analyses adjusted for phase differences to minimize the design limitation of a pre-post study. We used a composite variable comprising several conditions of varying prevalence to assess the potential holistic benefit of these interventions. We also conservatively presented composite variables to avoid over-interpretation of individual conditions' results and secondary outcomes.

It is worthwhile testing these interventions in other populations
where prevalence of these conditions may vary.
Although this study solely evaluated PHC interventions that might improve referral, we concurrently implemented a similar study at the DH, which will yield important insights about the interventions' utility in higher-level care settings (manuscript submitted).
Complementarity of interventions across the care continuum and health system tiers to improve identification of high-risk obstetric conditions and referral is critical.
In conclusion, ultrasound use substantially increased referrals from PHCs to the DH for both correct and incorrect diagnoses of the six conditions. Its use in labor triage could benefit from more nuanced consideration of costs, facility resources, and implementation challenges. The PHC triage checklist and standardized documentation were less beneficial in terms of referral, but increased correct identification. Therefore, a future study examining the impact of both interventions on maternal and newborn outcomes is warranted, as well as evaluation of a more focused or targeted scan for select women at labor triage to maximize diagnostic accuracy and referral decisions.

CO N FLI C T S O F I NTE R E S T
The authors of this study have no financial conflicts of interest to report.

D E TA I L S O F E TH I C S A PPROVA L
Ethics approval for this study was obtained from the Institutional Review Board at the University of California San Francisco IRB (#17-23310) and the Higher Degrees, Research and Ethics Committee at Makerere University in Uganda (#515). Initial approvals were obtained October 28, 2017 and November 7, 2017, respectively, and renewed annually.

S U PP O RTI N G I N FO R M ATI O N
Additional supporting information may be found online in the Supporting Information section at the end of the article. Table S1. Maternal and newborn outcomes (n=2339). Table S2. Maternal disposition at intake among those with an outcome-defined condition (n=2339). Table S3. Maternal disposition at intake among study population including those with or without outcome data (n=2719).   Appendix S1. Checklists for Phases 2 and 3.