Menu
Menu
Menu
Menu
Maternal and perinatal mortality remain huge challenges globally, particularly in low- and middle-income countries (LMICs) where >98% of these deaths occur. Emergency obstetric care (EmOC) provided by skilled health personnel is an evidence-based package of interventions effective in reducing these deaths associated with pregnancy and childbirth. Until recently, pregnant women residing in urban areas have been considered to have good access to care, including EmOC. However, emerging evidence shows that due to rapid urbanization, this so called “urban advantage” is shrinking and in some LMIC settings, it is almost non-existent. This poses a complex challenge for structuring an effective health service delivery system, which tend to have poor spatial planning especially in LMIC settings. To optimize access to EmOC and ultimately reduce preventable maternal deaths within the context of urbanization, it is imperative to accurately locate areas and population groups that are geographically marginalized. Underpinning such assessments is accurately estimating travel time to health facilities that provide EmOC. In this perspective, we discuss strengths and weaknesses of approaches commonly used to estimate travel times to EmOC in LMICs, broadly grouped as reported and modeled approaches, while contextualizing our discussion in urban areas. We then introduce the novel OnTIME project, which seeks to address some of the key limitations in these commonly used approaches by leveraging big data. The perspective concludes with a discussion on anticipated outcomes and potential policy applications of the OnTIME project.
Broadly, methods that have been used for estimating travel time to EmOC in LMICs can be grouped into reported and modeled approaches. Reported approaches entail asking health workers or women to estimate their travel times to health facilities. Some concerns with this approach have been raised. First, since health workers themselves did not make the journeys, their estimates are at best “guestimates” of the journeys that women might have undertaken to reach the health facility. In cases where women are asked to report their travel time, issues of recall bias have been highlighted by researchers, especially as they traveled in an emergency (15, 16). On the other hand, modeled approaches are commonly used to estimate travel time to health facilities in LMICs (17). They range from simple approaches such as Euclidean model, to sophisticated methods that include network analysis, cost distance analysis and gravity models as summarized in Figure 1. These methods have been detailed by Ouma et al. (17). Briefly, Euclidean distances are the simplest to compute and assume straight line of travel from residence to EmOC locations, however, they ignore the influence of transport variables such as travel barriers, road network and slope. Gravity models combines availability and accessibility across defined spatial units to overcome this limitation of Euclidean approaches. However, the method may suffer from the modifiable areal unit problem, is dependent on the availability of population at very fine geographic units and service provider capacity data which are not always available in resource limited settings. Network analysis entails computing travel time along existing travel routes to a specified health facility. The method relies on a well-mapped transportation network and settlements, assumes travel can only occur along the roads and it is computationally intensive. On the other hand, cost distance analysis relies on travel speeds across land covers, road network and elevation, to define the least time needed to get to a health facility from residences. A common problem underlying all these approaches is reliance of empirical data (which is rarely available) to parametrise a model that represents the journey between where a need is triggered and the location of the service provider (18). Common methods for estimating travel time to health facilities in LMICs. An illustration of common approaches used to compute geographic access to health service providers in LMICs including (a), Euclidean distance, (b) network distance, (c) least cost path distance, and (d) gravity models. Despite the widespread application of modeled approaches, they have a range of known limitations, some of which are accentuated by the intrinsic dynamics and variability of conditions that typify urban contexts (19). One fundamental limitation is that it is hard to create accurate models that replicate actual journeys. This is often due to inadequate data on where the journey was initiated, the health facility visited, the route used, its condition at the time of travel (traffic, weather, accidents), the mode(s) of transport and the speed of travel. As a result, empirical models tend to make assumptions about travel speeds and mode of transport, rarely accounting for the dynamism of traffic conditions, weather conditions, and unforeseen travel circumstances such as waiting time, police checkpoints, or impassability of roads. As regards traffic in particular, urbanization and the expansion of the middle class in urban LMIC areas have resulted in a rapid increase in vehicular traffic, leading to significant traffic congestion (20). For example, commuters in Lagos, the largest megacity in sub-Saharan Africa spend an average of 30 h a week (equivalent to 75% of a 40-h working week) in traffic, with some taking up to 3 h to travel 10 km (21). In Asia, three megacities of India—Bengaluru, Mumbai, and New Delhi—are part of the top 10 highly ranked cities with populations over eight million and with the highest levels of traffic congestion across the globe (22). Previous research which compared modeled travel time estimates with those collected from replication of travel journeys made by pregnant women in Lagos showed that while the median replicated drive time was 50–52 min, mean errors of >45 min were reported for the cost-friction surface approach and Open Street Route Mapping (23). Ignoring variability in traffic conditions results in as much as three-fold overestimation of geographic coverage and masks intra-urban inequities in accessibility to emergency care (19). Another limitation of modeled approaches, as they have been commonly used, relates to establishing the travel destination. Majority of the modeled approaches estimate travel time to the nearest health facility. Yet, it is well established that even in emergencies, pregnant women may bypass the nearest health facility for a myriad of reasons including trust, cost, and the real or perceived quality of care. Women might also be referred from one facility to others. When this occurs, their journeys are typically a lot more complex, harder to model and does not always follow the path of the least resistance a common approach to modeled approaches (24–27). These limitations can result in underestimated time to access care, with significant implications for underserved populations that require targeted policies and action (28). The constraints reported by researchers regarding pushing the frontier to reflect closer-to-reality travel time estimates relate to capacity to accurately parametrise a model that mimics the dynamics of the journey between the residence and service provider (29, 30). Data required for improved model parameterization include residential location of service users, location of the utilized facility providing EmOC, route used, mode of transport, traffic and weather variables, travel speed and transport barriers, among other travel dynamics (31). However, collecting such data is time-consuming, expensive, and probably impractical especially in low resource settings where there are many competing needs for resources. Also, the dynamics change dramatically, so data from last year or even last month may become less useful for understanding travel of mothers in an emergency. To move forward, such data needs to be real-time or at least close to real time.
N/A
