Is the closest health facility the one used in pregnancy care-seeking? A cross-sectional comparative analysis of self-reported and modelled geographical access to maternal care in Mozambique, India and Pakistan

• Main Authors: Liberty Makacha, Prestige Tatenda Makanga, Yolisa Prudence Dube, Jeffrey Bone, Khátia Munguambe, Geetanjali Katageri, Sumedha Sharma, Marianne Vidler, Esperança Sevene, Umesh Ramadurg, Umesh Charantimath, Amit Revankar, Peter von Dadelszen
• Format: Article
• Language: English
• Published: BMC 2020-02-01
• Series: International Journal of Health Geographics
• Subjects: Potential access; Realised access; Bland–Altman Index; Fixed bias; Limits of agreement; Proportional bias
• Online Access: https://doi.org/10.1186/s12942-020-0197-5

Abstract Background Travel time to care is known to influence uptake of health services. Generally, pregnant women who take longer to transit to health facilities are the least likely to deliver in facilities. It is not clear if modelled access predicts fairly the vulnerability in women seeking maternal care across different spatial settings. Objectives This cross-sectional analysis aimed to (i) compare travel times to care as modelled in a GIS environment with self-reported travel times by women seeking maternal care in Community Level Interventions for Pre-eclampsia: Mozambique, India and Pakistan; and (ii) investigate the assumption that women would seek care at the closest health facility. Methods Women were interviewed to obtain estimated travel times to health facilities (R). Travel time to the closest facility was also modelled (P) (closest facility tool (ArcGIS)) and time to facility where care was sought estimated (A) (route network layer finder (ArcGIS)). Bland–Altman analysis compared spatial variation in differences between modelled and self-reported travel times. Variations between travel times to the nearest facility (P) with modelled travel times to the actual facilities accessed (A) were analysed. Log-transformed data comparison graphs for medians, with box plots superimposed distributions were used. Results Modelled geographical access (P) is generally lower than self-reported access (R), but there is a geography to this relationship. In India and Pakistan, potential access (P) compared fairly with self-reported travel times (R) [P (H0: Mean difference = 0)] < .001, limits of agreement: [− 273.81; 56.40] and [− 264.10; 94.25] respectively. In Mozambique, mean differences between the two measures of access were significantly different from 0 [P (H0: Mean difference = 0) = 0.31, limits of agreement: [− 187.26; 199.96]]. Conclusion Modelling access successfully predict potential vulnerability in populations. Differences between modelled (P) and self-reported travel times (R) are partially a result of women not seeking care at their closest facilities. Modelling access should not be viewed through a geographically static lens. Modelling assumptions are likely modified by spatio-temporal and/or socio-cultural settings. Geographical stratification of access reveals disproportionate variations in differences emphasizing the varied nature of assumptions across spatial settings. Trial registration ClinicalTrials.gov, NCT01911494. Registered 30 July 2013, https://clinicaltrials.gov/ct2/show/NCT01911494