Impact of refugees on local health systems: A difference-in-differences analysis in Cameroon

Hosting refugees may represent a drain on local resources, particularly since external aid is frequently insufficient. Between 2004 and 2011, over 100,000 refugees settled in the eastern border of Cameroon. With little known on how refugee influx affects health services of the hosting community, we investigated the impact of refugees on mother and child health (MCH) services in the host community in Cameroon. We used Cameroon’s 2004 and 2011 Demographic and Health Surveys to evaluate changes in MCH indicators in the refugee hosting community. Our outcome variables were antenatal care (ANC) coverage, caesarean delivery rate, place of delivery and child vaccination coverage; whereas the exposure variable was residence in the refugee hosting community. We used a difference-in-differences analysis to compare indicators of the refugee hosting community to a control group selected through propensity score matching from the rest of the country. A total of 10,656 women were included in our 2004 analysis and 7.6% (n = 826) of them resided in the refugee hosting community. For 2011, 15,426 women were included and 5.8% (n = 902) of them resided in the hosting community. Between 2004 and 2011, both the proportion of women delivering outside health facilities and children not completing DPT3 vaccination in the refugee hosting community decreased by 9.0% (95% Confidence Interval (CI): 3.9-14.1%) and 9.6% (95% CI: 7.9-11.3%) respectively. However, ANC attendance and caesarean delivery did not show any significant change. Our findings demonstrate that none of the evaluated MCH service indicators deteriorated (in fact, two of them improved: delivery in health facilities and completing DPT3 vaccine) with the presence of refugees. This suggests evidence disproving the common belief that refugees always have a negative impact on their hosting community. We conducted a retrospective secondary data analysis based on two nationally-representative surveys from Cameroon to evaluate the impact of refugee presence on MCH services in a local refugee hosting community. Using data from the 2004 and 2011 Cameroon Demographic and Health Surveys (DHS), we compared MCH indicators of the local refugee hosting community with a control population selected from people living elsewhere in the country. Our control population was selected through propensity score matching. Finally, we performed a difference-in-differences analysis to estimate the effect of the presence of refugees on MCH services in the refugee hosting community. The study was reported in accordance with the STROBE (STrengthening the Reporting of OBservational studies in Epidemiology) guidelines [9]. Cameroon is a lower middle income country in sub-Saharan Africa. It has a population of 23.34 million inhabitants spread over 475,650 square kilometers. The country is relatively peaceful and stable [10]. Because of continual conflicts in the neighboring countries, a large number of refugees have been moving to Cameroon. Before 2006, only 4,000 persons of concern to the UNHCR were present in the country. After this period, most Central African Republic refugees fleeing from high level banditry and other criminal acts settled along the eastern borders of the East and Adamaoua regions of Cameroon and spread over an area of more than 50.000 square kilometers. By 2011, the number of refugees had exceeded 100,000 [11]. Given that this study involved secondary data analysis, the following guidelines were respected: Data were used in strict conformity with the DHS rules and regulations for researchers including sharing data only among registered co-authors, and making no attempt to further identify any individual in the dataset which had been anonymized before access. Data were obtained from the DHS programme. The 2004 and 2011 Cameroon surveys included 10,462 and 15,050 ordinary households selected from 467 and 580 clusters throughout the country, respectively. The final survey units (households) were selected through a multistage clustered sampling. The participants selected for our analysis were all women aged 15–49 years, together with their children under one year of age who took part in the 2004 (10,656 women; 425 infants) and 2011 (15,426 women; 501 infants) Cameroon DHS [12]. Our main outcomes of interest were ANC coverage, caesarean delivery rate, place of delivery, and infant vaccination coverage. Access to reproductive health services—including ANC coverage and a skilled birth attendant—are crucial indicators of the continuum of care for mother and child and can be used to assess the change in healthcare delivery [13]. ANC refers to the visits by a pregnant woman to a trained health worker with the goal to detect, treat and prevent pregnancy related problems; the World Health Organisation (WHO) recommends a minimum of four antenatal visits [14]. In the DHS, mothers were asked the number of times they consulted for their pregnancy and then we dichotomized as attendance of less than four ANCs or not. Caesarean delivery was defined as birth through a caesarean section or not. It is estimated that an overall low prevalence of caesarean delivery is an indication of poor quality of maternal and child health services [15]. Place of delivery was defined as delivery in a health facility or elsewhere. Typically, deliveries out of health facilities carry high risks of negative outcomes in these settings. For child care, completing the 3rd dose of DPT vaccine is considered a good indicator for evaluation of vaccination coverage and is widely used in the extended program on immunisation (EPI) coverage [16]. In the DHS, the number of DPT doses the child had received was compared to the child’s age to determine if the child had correctly completed vaccination or not. This variable was therefore defined as non-completion of DPT3 vaccine for children less than one year who had not received their due three doses of DPT. The exposure variable was the place of residence and was dichotomized as residing in the refugee hosting community or not. Participants were assigned as residing in the refugee hosting community using quantum geographical information system (QGIS) version 2.14. First, clusters in both datasets were redistributed over the national territory based on their GPS coordinates recorded during the cluster sampling phase of the survey. The refugees hosting community was then considered to be the area mapped out by the UNHCR in their 2011 country report [11]. Clusters within the mapped area were considered to be exposed (refugee hosting community). Our covariates—variables that could predict residence in the refugee zone—initially included the following variables: mother’s educational level (no education, primary, secondary or tertiary education), residence (urban or rural), wealth index (poorest, poorer, middle, richer or richest), mother’s previous birth experience (haven given birth before or not), region of residence (any of the 10 regions in Cameroon), household size, household head (male or female), number children under five years in a household, religion (animist, catholic, muslim, protestants, new religions or others) and ethnicity (any of the 40 ethnic groups reported by participants). Although all the covariates were included in an initial generalized linear model for prediction of propensity scores, we subsequently deleted variables that did not help in the prediction of propensity scores. These variables were likely to generate a difference in our study groups as suggested by the literature [17–19], but they correlated with other variables. For example, a covariate such as wealth index is generated from several other potential covariates (including household’s ownership of selected assets, such as televisions and bicycles; materials used for housing construction; and types of water access and sanitation facilities) which could easily correlate with other variables like household size and household head. Our exploratory analysis started by evaluating distributions for each numeric variable of interest as well as frequencies and percentages for each of the categorical variables. Specifically, numeric variables were evaluated for normality in their distributions while categorical variables were evaluated for near-zero variation (presence of very few observation in any class) [20]. Graphical displays were used for both univariate analysis and bivariate associations, accompanied by broader tests such as Maximal Information Coefficient [21] and Nonnegative Matrix Factorization [22] algorithms for numeric variables. Missing data were explored using a combination of graphical displays involving univariate, bivariate and multivariate methods. Propensity scores representing the probability of residing in the refugee hosting community were calculated for each individual, and a one-to-one match [23] based on propensity scores was performed between participants in the refugee hosting community and those living elsewhere using the Matching package in R statistical software [24]. Balance between groups was evaluated through a combination of plots and statistical tests (t- and chi-square-tests). Once matched controls were found, the difference in proportion of each outcome was calculated between the refugee hosting community and their selected controls using the average treatment effect on the treated (ATT). The difference-in-differences of outcomes were then calculated between 2004 and 2011. Both a significance level of p-value less than 0.05 and a 95% confidence interval was used to interpret statistical tests.