Introduction Most of the literature on terrorist attacks’ health impacts has focused on direct victims rather than on distal consequences in the overall population. There is limited knowledge on how terrorist attacks can be detrimental to access to healthcare services. The objective of this study is to assess the impact of terrorist attacks on the utilisation of maternal healthcare services by examining the case of Burkina Faso. Methods This longitudinal quasi-experimental study uses multiple interrupted time series analysis. Utilisation of healthcare services data was extracted from the National Health Information System in Burkina Faso. Data span the period of January 2013-December 2018 and include all public primary healthcare centres and district hospitals. Terrorist attack data were extracted from the Armed Conflict Location and Event Data project. Negative binomial regression models were fitted with fixed effects to isolate the immediate and long-term effects of terrorist attacks on three outcomes (antenatal care visits, of facility deliveries and of cesarean sections). Results During the next month of an attack, the incidence of assisted deliveries in healthcare facilities is significantly reduced by 3.8% (95% CI 1.3 to 6.3). Multiple attacks have immediate effects more pronounced than single attacks. Longitudinal analysis show that the incremental number of terrorist attacks is associated with a decrease of the three outcomes. For every additional attack in a commune, the incidence of cesarean sections is reduced by 7.7% (95% CI 4.7 to 10.7) while, for assisted deliveries, it is reduced by 2.5% (95% CI 1.9 to 3.1) and, for antenatal care visits, by 1.8% (95% CI 1.2 to 2.5). Conclusion Terrorist attacks constitute a new barrier to access of maternal healthcare in Burkina Faso. The exponential increase in terrorist activities in West Africa is expected to have negative effects on maternal health in the entire region.
Burkina Faso is a landlocked country of ~20 million inhabitants located in West Africa, and surrounded by Mali, Niger, Benin, Togo, Ghana and Côte d’Ivoire. Between 1987 and 2014, the Republic was governed by Blaise Compaoré, a former military man who seized power in a coup d’état. Throughout this period, Burkina Faso was considered to be a relatively secure country despite human rights violations and sporadic tensions and clashes between ethnic or religious groups. However, the security situation changed rapidly in the mid-2010s. After mounting pressure against his attempt to modify the Constitution in order to remain in power, Compaoré was forced to resign and flee the country.32 Presidential elections were organised in 2015, but not before the failure of a 1-week-long contre-coup. During this short period of unrest, approximately 15 people were killed and over 300 were wounded according to press releases.33 Meanwhile, the security situation had dramatically deteriorated in the neighbouring countries of Mali, Niger and (Northern) Nigeria, where jihadist groups—sometimes allied with rebel movements with territorial claims—carried out regular attacks against both the population and military forces.34 With these groups moving across borders and pursuing regional ambitions, the exact reasons that Burkina Faso remained relatively free of terrorist attacks remain unclear. Nevertheless, its government agreed in 2014 to enter the G5 Sahel Joint Force, along with Mauritania, Mali, Niger and Chad, to coordinate a regional response to the terrorist threat. Since then, several jihadist groups have escalated their attacks throughout the country, most notably Ansarul Islam, Islamic State in the Greater Sahara, and the Group to Support Islam and Muslims (known by its Arabic acronym JNIM). As a member of the G5 Sahel Joint Force, Burkina Faso’s military and police are supported in the field by Operation Barkhane, a French-led military force of approximately 5000 soldiers. This is a longitudinal quasi-experimental study that used multiple (pooled) interrupted time-series analysis to evaluate the effects of terrorist attacks on access to maternal healthcare services at the level of the lowest administrative unit (ie, the commune).35 Immediate effects were defined as level changes in the month of or the month following an attack. Longitudinal effects of repeated attacks were examined by defining segments based on the incremental number of attacks in a commune over time and by measuring level change between segments. All communes of the national database were included in the analysis. The study period spanned from January 2013 to December 2018, totalling 72 time points of observation. This study has three outcome indicators: (1) the total number of ANC visits per commune per month; (2) the number of facility-based deliveries per commune per month; (3) the number of cesarean sections per commune per month. These outcomes were selected because they are key indicators of accessibility to maternal healthcare in low-income and middle-income countries and they are routinely collected in the facilities at the primary care level, including cesarean sections performed in district hospitals.36 In communes with several health facilities, the outcomes refer to the total number per commune per month. Models were adjusted for the proportion of missing data.37 38 Exposure was operationalised differently according to the objective. To evaluate the average immediate effects of a terrorist attack, communes that recorded at least one attack were defined as being exposed for that particular month and the following one, in order to cover a 30-day period after the attack. Therefore, the first exposure variable is categorical (no attack, single attack, multiple attacks) and reflects immediate exposure to an attack. Three categories were defined (rather than two, that is, absence/presence) to verify the presence of a dose–response relationship since it is hypothesised that more attacks will generate more insecurity and further reduce visits to health facilities. To evaluate the longitudinal effects of the incremental levels of insecurity, exposure was defined based on the cumulative number of attacks in a given commune over time. Exposure variable is therefore numeric and reflects the shift into a new ‘phase’ characterised by one additional attack. The duration of these phases (segments) vary since they last until a new attack occurs. For both objectives, a terrorist attack was defined as an act involving a jihadist group in which one of the protagonists used violence (ie, battle, explosion/remote violence, looting/property destruction and violence against civilians). Attacks involving ‘unidentified armed groups’ were included in terrorist attacks. Two secondary sources of data were used. First, the utilisation of healthcare services data was extracted from the National Health Information System in Burkina Faso. Data were available from January 2013 to December 2018, which constitutes a reliable time series of 72 points of observation. All public facilities at the primary care level were considered in the analysis, that is, primary healthcare centres (‘Centres de santé et de promotion sociale’) and district hospitals (‘Centres médicaux avec antenne chirurgicale’). Every month, facilities review their record books and complete a form that is sent to the Health District, which compiles data from all the facilities in its catchment area. Data quality is assessed in each district before being transmitted to the Director of Health Statistics at the Ministry of Health, where data from all health districts are compiled. The Ministry of Health performs regular supervision visits and audits in the field. The data collection instruments (record books, monthly reports, national database structure) remained constant during 2013–2018. Data from the passive surveillance system in Burkina Faso have been proven reliable in previous studies.39 40 Second, terrorist attack data were extracted from the Armed Conflict Location and Event Data (ACLED) project. The ACLED project collects data on violent events within States, which includes armed conflicts and terrorist activities with or without fatalities. Data are disaggregated by date, location and actors. This spatial scale is relevant for the purpose of the present study since its hypothesis is that terrorist attacks reduce access to the surrounding primary care facilities, rather than at the national level.41 For those violent events with fatalities, ACLED data were cross-checked and completed by using the Uppsala Conflict Data Program Georeferenced Event Dataset (UCDP-GEP).42 Based on the GPS coordinates of the events, communes were identified by using the database of Global Administrative Areas (GADM). Finally, the ACLED and passive surveillance datasets were merged at the commune-month level of aggregation. The unit of all analyses was the commune-month. To explore the attacks’ effects, three separate regression models (corresponding to the three outcomes) were fitted using the exact same set of variables and parameters. Even if the outcomes were all count variables, negative binomial regression was preferred over Poisson because of overdispersion. In order to best isolate the effect of attacks, the commune unit was entered as fixed effects while using unconditional maximum-likelihood estimation. This allows for control for any stable characteristic of the communes, whether observed or not.43 The underlying equation of the basic fixed effect level can be expressed as yit=μt + βxit + αi + εit with i=1, … n (communes) and t=1, … t (time) where μ is a constant term, yit is the response value for the commune i at time t, x is a vector of time-variant variables, αi are commune-specific intercepts that capture heterogeneity between communes and ε are residual errors. Four time-varying variables were entered in the models: the monthly variation (calendar month), the baseline trend (time units since January 2013), the trend since occurrence of the first attack (time units since the month of the first attack in a commune) and the percentage of missing observations. The linearity of the relationship between the outcome and continuous covariate was assessed by adding quadratic terms. Multicollinearity was ruled out by using the Collin package (StataCorp, College Station, Texas) and verifying that variance inflation factors did not exceed 4. Robust variance estimators (Huber/White estimator) were used throughout the analyses. Coefficients were expressed as incidence rate ratios. The threshold for statistical significance was set at 0.05 (bilateral tests). All analyses were performed in Stata V.14.0 (StataCorp). Maps were created using QGIS V.3.8 (open-source GIS software). This study only uses secondary, administrative data. GADM, ACLED and UCDP-GEP data are publicly available online (https://gadm.org/, https://acleddata.com/ and https://ucdp.uu.se/). Access to the National Health Information System data was granted by the Ministry of Health of Burkina Faso (Notice #2018-3032). Patients and members of the public were not used in the design, conduct, reporting and dissemination of this research. Utilisation of healthcare services by patients were routinely collected by health facilities providers and analysed; however, data were aggregated and individual patients cannot be identified from the reported data.
N/A