The free caesareans policy in low-income settings: An interrupted time series analysis in Mali (2003-2012)

Introduction: Several countries have instituted fee exemptions for caesareans to reduce maternal and newborn mortality. Objectives: To evaluate the effect of fee exemptions for caesareans on population caesarean rates taking into account different levels of accessibility. Methods: The observation period was from January 2003 to May 2012 in one Region and covered 11.7 million person-years. Exemption fees for caesareans were adopted on June 26, 2005. Data were obtained from a registration system implemented in 2003 that tracks all obstetrical emergencies and interventions including caesareans. The pre-intervention period was 30 months and the post-intervention period was 83 months. We used an interrupted time series to evaluate the trend before and after the policy adoption and the overall tendency. Findings: During the study period, the caesarean rate increased from 0.25 to 1.5% for the entire population. For women living in cities with district hospitals that provided caesareans, the rate increased from 1.7% before the policy was enforced to 5.7% 83 months later. No significant change in trends was observed among women living in villages with a healthcare centre or those in villages with no healthcare facility. For the latter, the caesarean rate increased from 0.4 to 1%. Conclusions: After nine years of implementation policy in Mali, the caesarean rate achieved in cities with a district hospital reached the full beneficial effect of this measure, whereas for women living elsewhere this policy did not increase the caesarean rate to a level that could contribute effectively to reduce their risk of maternal death. Only universal access to this essential intervention could reduce the inequities and increase the effectiveness of this policy. © 2014 Fournier et al. Of the seven districts of the Kayes region, two were not included in this study. The first was the Kayes district, where the regional hospital is located. The mainly urban population of the Kayes district does not encounter the same problems of geographic access and financial constraints that arise in the other districts, whose populations are mostly rural. In that district there are specialists (obstetricians, anesthesiologists, and pediatricians), whereas in district hospitals caesareans are done by general practitioners with surgical training. The second district excluded was Kenieba, where the Ref-Syst was implemented after the Free-CSec policy. To determine whether caesarean rates varied according to area of residence, we considered three zones: (1) cities with district hospitals; (2) villages with primary healthcare centres; (3) villages without a healthcare facility. In addition to providing an approximate correspondence to geographic accessibility, these zones represent different levels of organisational accessibility. In (1), services are directly accessible. In (2), the Ref-Syst facilitates transportation (ambulance availability) and covers part of the costs of transportation and services, although it is not guaranteed to be always fully operational. For (3), women and their families must first find transportation to go to the primary healthcare centre or directly to the district hospital, which entails travel over distances up to 100 km and involves considerable effort and cost. The policy of fee exemptions for caesareans was announced on June 26 2005. To evaluate its effects, we used a time series beginning on January 2003 and ending in May 2012. This series includes a pre-intervention period of 30 months (January 1st 2003 to June 30th 2005) and a post intervention period of 83 months (July 1st 2005 to May 31st 2012). The assessed outcome is the estimated monthly caesarean rate (number of caesarean deliveries/total number of deliveries). The numerator is the number of caesarean cases identified from the system GESYRE (Gestion du Système de Référence Evacuation) and the denominator is the estimate of the total number of births. GESYRE, implemented in 2003 as part of a collaborative research program with the health and social authorities of the Kayes region, is a registration system for all obstetrical emergencies and interventions including caesareans [25]. The denominator was determined using data from the 1998 and 2009 censuses at the village level. The total number of births for each commune was estimated from the 1998 and 2009 censuses and the annual crude birth rates as follows: Population from the 69 communes included in the study was estimated from 2003 to 2012 using their specific growth rate observed between the two censuses. The expected number of deliveries in one specific commune (a) is the population in year (i) x crude birth rate for year (i). (For more details see Text S1: Population and deliveries estimates). Thus, the population caesarean rate for a given geographic area and time period is the number of caesareans carried out in that period among women living in that geographic area divided by the number of births expected in that area for that period. Healthcare policies that are applied to the population at large inherently cannot be studied using experimental designs, and interrupted times series are appropriate alternatives to randomised trials [26]. Their use is recommended for clinical evaluations [27] and for measuring the effects of public policies on health [28]–[29]. Since monthly data were collected over time they were suitable for interrupted time series analysis. We used segmented linear regression models to estimate the change in the caesarean rate before and after the Free-CSec policy, immediately and over time. The Durbinalt test, a Durbin-Watson’s alternative test for serial correlation in data, showed a moderate serial autocorrelation, such as the partial autocorrelation plot. Failing to correct for autocorrelation in longitudinal data may lead to underestimated standard errors and overestimated significance of the effects of an intervention. Therefore we used an extended ordinary least-squares (OLS) regression model divided into pre and post-intervention segments while adjusting the variance estimation by the Newey-West standard errors method that corrected for serial correlation in residuals. The maximum lag to be considered in the autocorrelation structure was determined by visual inspection and with confidence intervals calculated using a standard error of 1/sqrt(n). The segmented regression model best fits an OLS regression line as follows [30]: β0 estimates the baseline caesarean rate at the beginning of the pre Free-CSec period. β1 estimates the change in rate that occurs with each month before the Free-CSec policy. β2 estimates the change in the caesarean rate immediately after the Free-CSec policy. β3 estimates the change in the trend of the rate of the post-Free-CSec period compared to the pre-Free-CSec period. To describe a clinically meaningful absolute reduction, the absolute effect (β2+β3 * Number of months after intervention) was estimated by difference between the estimated outcome at a certain time after the intervention and the outcome at that time if the intervention not taken place Its standard error was calculated including the covariance of level and slope terms [30]–[31]. The statistical significance for parameter estimation was set at α = 5% (p≤0.05). All analyses were performed using Stata 11 [32]. This research was approved by the Ethics Committees of the University of Montreal Hospital Research Centre (Canada) and the Faculty of Medicine, Pharmacy and Odonto-Stomatology of the University of Bamako (Mali). No written consent was obtained from participants for using their clinical records, but patient records/information was anonymized and de-identified prior to analysis.