Menu
Menu
Menu
Menu
Introduction Longer travel times are associated with increased adverse maternal and perinatal outcomes. Geospatial modelling has been increasingly used to estimate geographic proximity in emergency obstetric care. In this study, we aimed to assess the correlation between modelled and patient-reported travel times and to evaluate its clinical relevance. Methods Women who delivered by caesarean section in nine hospitals were followed up with home visits at 1 month and 1 year. Travel times between the location before the delivery and the facility where caesarean section was performed were estimated, based on two models (model I Ouma et al; model II Munoz et al). Patient-reported and modelled travel times were compared applying a univariable linear regression analysis, and the relation between travel time and perinatal mortality was assessed. Results The median reported travel time was 60 min, compared with 13 and 34 min estimated by the two models, respectively. The 2-hour access threshold correlated with a patient-reported travel time of 5.7 hours for model I and 1.8 hours for model II. Longer travel times were associated with transport by boat and ambulance, visiting one or two facilities before reaching the final facility, lower education and poverty. Lower perinatal mortality was found both in the group with a reported travel time of 2 hours or less (193 vs 308 per 1000 births, p<0.001) and a modelled travel time of 2 hours or less (model I: 209 vs 344 per 1000 births, p=0.003; model II: 181 vs 319 per 1000 births, p<0.001). Conclusion The standard model, used to estimate geographical proximity, consistently underestimated the travel time. However, the conservative travel time model corresponded better to patient-reported travel times. The 2-hour threshold as determined by the Lancet Commission on Global Surgery, is clinically relevant with respect to reducing perinatal death, not a clear cut-off.
Sierra Leone, in West Africa, reports some of the world’s worst maternal and perinatal health outcomes.17 More than half (54.4%) of the deliveries take place in a health facility,18 either in one of the 1160 primary healthcare units, or 24 public or 30 private hospitals.19 The national population caesarean section rate is 2.9%,20 far below the suggested threshold of 10%–19%,21 22 and reflects limited access to emergency obstetric services in the country. Delayed and substandard care results in a perinatal mortality rate after caesarean section of 190 per 1000 births,18 much higher than the national perinatal mortality rate of 39 per 1000 pregnancies.23 This study was part of a prospective multicentre audit comparing outcome of caesarean sections performed by medical doctors and associate clinicians in nine hospitals in Sierra Leone.23 24 The study facilities consisted of four district hospitals, one regional hospital, the national maternity referral hospital and three private non-profit hospitals, located in all geographical regions of Sierra Leone. Women who underwent a planned caesarean section were excluded from the analysis. In each of the participating hospitals, anaesthesia team members were trained to enrol patients and do the in-hospital data collection. Data collection was supervised and reviewed by the primary investigator, during hospital visits at 1–3 weeks intervals, throughout the whole study period. Data were entered into a Microsoft Excel 2016 database in the study facilities and inconsistent or missing data were supplemented from operation logbooks or patient files. During admission, the following data was collected: the patient’s address before coming to the hospital; patient-reported estimated time from the place of stay before the delivery to the facility where the caesarean section was performed; other health facilities visited en route to the hospital; and clinical process and outcome data. Except for clinical data, information was provided by patients and their relatives. Follow-up home visits were conducted at 1 month and 1 year after discharge by four research nurses. During the home visits, data collected while admitted was verified, information regarding the means of transport to the hospital was collected and geolocations were recorded. OpenStreetMap was used to review all geospatial data regarding location before coming to the hospital.25 Two previously published geospatial models were used to create travel time maps. The first (model I) was based on the methods described by Ouma et al11, which overestimated geographical access compared with patient-reported travel time in a recent study.16 Several more conservative national models have been published from Rwanda, Ghana, Tanzania and Zambia.12–15 As a sensitivity analysis, the model from Rwanda (Huerta Munoz et al12, walking and public transport scenario) was applied to our data set as it presented the most conservative travel time estimates. Minor adjustments were made to both models to increase the reproducibility (table 1). Comparison of two geospatial models Comparison of two geospatial models based on the methodology described by Ouma et al11 and “scenario 3” Munoz et al.12 *Walking and public transport scenario. †In this study we extracted the road network from OpenStreetMap while Ouma et al combined the road network from OpenStreetmap and Google Map Maker Project and Munoz et al. obtained the road network from Centre for Geographical Information Systems – National University of Rwanda. ‡In this study we applied a spatial grid of 94 m compared to a spatial grid of respectively 100 m and 90 m in the original articles. DEM, digital elevation model; N/A, not assigned. Based on the two models, two maps were generated for each of the nine study hospitals (see online supplemental figure 1), using the open-source WHO tool AccessMod V.5.6.0,26 freely available geospatial data and geographical information systems (GIS) desktop software (QGIS V.3.12, Open Source Geospatial Foundation Project). The input layers for each map analysis were a Digital Elevation Model (DEM) with a resolution of 94 m and a vegetation map extracted from the 2016 Africa land cover (Contains modified Copernicus data (2015/2016), ESA Climate Change Initiative-Land Cover project 2017). Rivers and road networks, with primary (including trunk roads), secondary and tertiary roads (including unclassified roads for model II), were retrieved from OpenStreetMap.25 The models were based on the assumption that patients would reach the nearest road at walking speed and then continue with other means of transport. Road and non-road speeds varied between the models. For model II, an anisotropic analysis using the DEM was incorporated, implying that travel speeds were affected by the slope of the terrain. For each patient, travel time to the facility was extracted from the facility-specific maps with QGIS V.3.12 software. bmjgh-2020-003943supp001.pdf Univariable linear regression analysis was used to compare patient-reported and geospatial modelled travel times. For each patient, a conversion factor was calculated by dividing the modelled travel time by the patient-reported travel time. Median and IQRs were used to present the non-normally distributed variables travel time and conversion factor. Travel time comparisons were presented using scatter plot with regression line and analysed with correlation coefficients. Statistical analyses were performed with STATA V.16.0 (StataCorp). P values of less than 0.05 were considered statistically significant. For patients that used multiple means of transport, the main mode of transport was presented. Maps were produced with QGIS V.3.12 to illustrate patient-reported and geospatial modelled travel times, using the raw output of the AccessMod tool and an interpolated surface of patient-reported travel times. The interpolation was produced using inverse distance weighting. Written informed consent was obtained from each woman by signature or thumbprint, either before or as soon as possible after surgery. The study is registered in the international clinical trial register (ISRCTN16157971). Patients were involved in the selection of the content of the health promotion packages provided as an incentive during home visits and provided feedback on the burden of the questionnaires.
N/A
