Menu
Menu
Menu
Menu
Background Travel time to comprehensive emergency obstetric care (CEmOC) facilities in low-resource settings is commonly estimated using modelling approaches. Our objective was to derive and compare estimates of travel time to reach CEmOC in an African megacity using models and web-based platforms against actual replication of travel. Methods We extracted data from patient files of all 732 pregnant women who presented in emergency in the four publicly owned tertiary CEmOC facilities in Lagos, Nigeria, between August 2018 and August 2019. For a systematically selected subsample of 385, we estimated travel time from their homes to the facility using the cost-friction surface approach, Open Source Routing Machine (OSRM) and Google Maps, and compared them to travel time by two independent drivers replicating women’s journeys. We estimated the percentage of women who reached the facilities within 60 and 120 min. Results The median travel time for 385 women from the cost-friction surface approach, OSRM and Google Maps was 5, 11 and 40 min, respectively. The median actual drive time was 50-52 min. The mean errors were >45 min for the cost-friction surface approach and OSRM, and 14 min for Google Maps. The smallest differences between replicated and estimated travel times were seen for night-Time journeys at weekends; largest errors were found for night-Time journeys at weekdays and journeys above 120 min. Modelled estimates indicated that all participants were within 60 min of the destination CEmOC facility, yet journey replication showed that only 57% were, and 92% were within 120 min. Conclusions Existing modelling methods underestimate actual travel time in low-resource megacities. Significant gaps in geographical access to life-saving health services like CEmOC must be urgently addressed, including in urban areas. Leveraging tools that generate â € closer-To-reality’ estimates will be vital for service planning if universal health coverage targets are to be realised by 2030.
Lagos State, located south-west Nigeria, has a mix of different geographical terrains (land and riverine) and settlement types, including a megacity, suburbs, towns, informal settlements and slums. Our study was conducted in the Lagos megacity, which is the most populous in sub-Saharan Africa with 13.5 million inhabitants in 2018.11 With an unprecedented population growth, researchers have projected that the population of Lagos will be tripled by 2050.12 Within the megacity, the most popular mode of travel is by road. However, the road infrastructure is particularly poor in many parts of the city with numerous potholes that are sometimes as wide as the road itself. The road conditions are worsened during the rainy season with flooding, though bumper-to-bumper traffic remains a constant feature irrespective of the season (dry or rainy). Efforts at road repairs are at best stop gaps and sometimes generate even more travel disruptions.13–16 Within the Lagos megacity and its surrounding suburbs, there are 16 public CEmOC facilities, including 12 general hospitals and four tertiary referral hospitals with capacity to provide all nine signal functions 24 hours/day. For this study, we focused on the four tertiary referral hospitals: Federal Medical Centre, Ebute-Metta (FMC), Lagos Island Maternity Hospital (LIMH) and the Institute of Maternal and Child Health (IMCH, commonly referred to as Ayinke House) at the Lagos State University Teaching Hospital and Lagos University Teaching Hospital (LUTH). These four hospitals are the apex public referral facilities managing the most complex obstetric emergencies referred from other public CEmOC facilities (general hospitals), private hospitals/clinics and primary health centres. In 2018, FMC, LIMH and LUTH managed 986, 3681 and 2011 deliveries, respectively. We have no data for IMCH, as this facility was only just reopened for service on 24 April 2019 after a 9-year shutdown for renovation.17 18 To reach these facilities, most women travel on their own or accompanied by their relative(s). If referral is needed, the Lagos State Ambulance system functions to transfer patients between facilities, though the service is not always available for pregnant women and when it is, it appears to mostly transfer from hospitals to other hospitals and not from primary healthcare centres.19 Traffic congestion, lack of driver etiquette with other commuters not giving way and community disturbance are some reasons that minimise the service effectiveness for patient transfer.20 21 Compared with the national maternal mortality ratio (MMR) of 512 maternal deaths per 100 000 live births (year 2017),22 MMR in Lagos State has been estimated to be 450 (95% CI 360 to 530) per 100 000 live births.23 However, estimates as high as 1050 (95% CI 894 to 1215) per 100 000 live births were reported in slum areas.24 In this study, data were collected from review of patient records of all pregnant women who presented in obstetric emergency situations (any major pregnancy and childbirth complication) at the four CEmOC facilities between November 2018 and October 2019. Over a 6-month period, the data were extracted from the records by members of the research team supported by trained research assistants who were qualified medical doctors conversant with the patient records system in Lagos public health facilities. Records were included whether or not women were referred from another facility. Pregnant women whose complications were identified during routine antenatal visits were excluded, as their journeys did not reflect emergency situations (88 cases). In addition, we excluded cases of women who had untraceable home addresses (26 cases). In all, using a systematic sampling technique, which eliminates the risk of clustered selection,25 we sampled every second woman from the pool of 732 included patient records. Using a predesigned online data extraction form, we captured data on demographic characteristics, obstetric history, index admission history (day and month of presentation), period of day when journey to the facility commenced (morning, afternoon, evening or night), street name of women’s self-reported start location (origin), other facilities visited en route (referral points), if any, and the destination facility. We geocoded the origin, any referral and destination locations. Three methods were used for travel time estimation. For method 1, travel time was obtained using the cost-friction surface approach. The friction surfaces were derived from a variety of geospatial data sets, including landscape characteristics and the road network.8 The cost-friction surface approach associates a value that represents the generalised difficulty for trespassing (represented as speed) to each 1 km2 grid covering the study region depending on land surface condition (eg, roads and waterbodies). The travel time between two points was then obtained with an algorithm that identifies the path that requests the least total difficult (time). This approach has previously been used to estimate travel time to healthcare facilities both in sub-Sahara Africa and at the global scale.8 26 For method 2, we used the Open Source Routing Machine (OSRM), a routing engine designed to run on OpenStreetMap data,27 to find the fastest route between pairs of coordinates. For method 3, estimated travel time between origin and destination was obtained from Google Maps using the ‘typical time of travel’ tool for the time and day that the woman commenced her journey. In addition to an assumed speed, Google Maps also accounts for traffic condition at specified time. To collect travel time estimates via Google Maps for the period of the day when journey to the facility commenced, we used specific time slots (09:00, 15:00, 18:00 and 21:00 for morning, afternoon, evening or night journeys, respectively). Lastly, the journeys of the women to reach the facilities during the period of the day of travel, including any referrals in between, were replicated by professional motor vehicle drivers. One driver replicated the journey by following the route suggested by Google Maps at the period of the day that women commenced journeys as closely as possible. A second driver used native intelligence to navigate their route from origin to destination. These journeys were tracked with a mobile application, Life360 (Life360, San Francisco, USA). Both drivers were mandated to drive carefully and keep within the speed limit. For journeys in which we could not tell the time of the day that women commenced their journeys to the facility (33% of sample), we assumed that these journeys were made in the afternoon on the day of presentation for Google Maps extraction (method 3) and journey replication. The choice of this period was made because it offered a conservative travel time estimate in between the two known peak periods for travel in Lagos (06:30 and 11:30 (morning peak period) and 15:00 and 19:30 (evening peak period)).16 All journey replications were undertaken from 12 June 2020 to 7 August 2020 which were during the 270-day period of the rainy season expected in Lagos for the year 2020.28 29 Following the descriptive analysis of the sample, we compared travel time estimates obtained using the three methods with the median travel time of the two replicated journeys. The extent to which method 1 (cost-friction surface approach), method 2 (OSRM) and method 3 (Google Maps) match with the travel time estimates of journey replication was measured by the mean absolute error (MAE), root mean square error (RMSE) and the intraclass correlation coefficient (ICC) for agreement from a one-way random effects model. The MAE was used to detect bias and should be zero if the travel time estimates were unbiased. RMSE was used to measure the average magnitude of the squared error. Smaller MAE and RMSE values would indicate few errors and more ‘accurate’ estimates. ICC was used to indicate the absolute agreement between different measures. Negative ICC values suggest very appealing agreement, and positive ICC values range from 0 to 1, with greater values indicating between agreement. In addition to reporting overall MAE, RMSE and ICC, we also reported MAE, RMSE and ICC disaggregated by day in the week and time of the day, participants’ referral status and total time of journey. Lastly, we reported the percentage of women living within 60 and 120 min of travel to the CEmOC facilities they attended, based on the estimates and the journey replication. Analysis was carried out using the ‘osrm’ and ‘gdistance’ packages in R V.4.0.2 (R Development Core Team, Auckland, New Zealand); all maps were drawn with the ‘ggmap’ package, including the tile server for Stamen Maps. There were no missing data. Patients and/or the public were not involved in the design, conduct, reporting or dissemination of this research.
N/A
