MORES Intervention:
Following baseline collection of referral data, a two-day interactive obstetric triage and MORES referral training was conducted with the nurses and midwives at 20 RHFs and two district hospitals in Bong County, Liberia. The training focused on recognition and management of acutely ill pregnant woman, managing obstetric emergencies, and prioritization of care. The healthcare providers were provided with an overview of MORES, assigned unique ID, and messaging templates for RHF referrals to the district hospitals (Figure 1) and templates for received hospital referrals back to the RHF provider were designed to ensure bi-directional communication (Figure 2). Templates for referrals from providers at RHFs included information such as patient initials, reasons for referral, time of referral, and type of transportation. Templates for providers at the district hospitals to communicate back to RHFs providers included information such as unique ID, patient initials, referral received confirmation, patient’s arrival time and maternal and newborn outcomes. Prior to the implementation of the MORES intervention, referrals were handwritten at the RHF and carried by the women’s family to the district hospital. Furthermore, these handwritten referrals would often get misplaced and forgotten, causing the necessary information to never reach the hospital.
Data collection and procedures :
Research assistants (RAs) from the Bong County health team obtained permission from the RHFs and district hospitals to conduct chart reviews and collect transfer data. A retrospective review of the hospital labor and delivery logs for baseline referrals sent from the 20 RHFs to the two district hospitals was conducted prior to implementation of the MORES intervention. Data collection was repeated at endline, during the final six months of the study. Identifying data was collected to link data from hospital labor and delivery log and patient charts to data from RHF logs. The linking document, and identifiable data, were destroyed after clinical data were collected. Deidentified data collected and used for analysis included date and time of departure from RHF and hospital arrival, type of delivery, and maternal and newborn outcomes. The hospital logs contained more referral records than those recorded in the RHF logs, as not all women referred arrived at the district hospital. Thus, data were sorted as complete (hospital data linked to RHF data) or incomplete (hospital records only).
Data management and analysis :
Referral data were collected via paper and pen, entered into Excel and exported into Stata 17 (StataCorp, College Station, TX, USA) for analysis. All paper copies were stored in a locked cabinet at the county health office in Bong County. All digital data were stored in an encrypted DropBox folder only accessible to the research team.
Descriptive analyses were conducted for all baseline and endline data and analyzed as separate cohorts (complete and incomplete). Frequency and percentage were tabulated for RHFs, referrals to district hospitals, reasons for referral, and outcome variables including mode of delivery, maternal outcome, newborn outcome, and transfer time from RHF to district hospital. Transfer time was further presented as mean with (SD), as well as the median time, and categorized into two hours or less, between two hours and twelve hours, and more than 12 hours. The two hours or less category was included per The Lancet’sdefinition of geographic accessibility, with access to a health facility with the capacity to provide essential surgical and anesthesia services, including CS, within two hours (14). Furthermore, the 12 hours or less and more than 12 hours were included because despite a referral from a RHF, woman often go back home before going to the referred hospital.
Binary logistic regression models and linear regression models were fit to assess the relationship between the timepoint of the data (baseline, endline), mode of delivery, maternal outcome, newborn outcome, and transfer time. All models accounted for the clustering of individuals within hospital and RHF. All logistic regression models provided odds ratios (ORs) and 95% confidence intervals (95% CIs) and linear regression models provided coefficient and standard error (SE).
Ethics :
Ethical approval for the study was obtained from the Institutional Review Boards of the University of Michigan and University of Liberia. The study used retrospective medical records, the data were fully anonymized before analysis, and the ethics committee waived the requirement for the informed consent.