Study setting
Burkina Faso is a low-income country where the public health care system is organized on a pyramidal basis with three levels. The first level is the health district, which comprises basic health centers, called “Centre de Santé et de Promotion Sociale (CSPS)”, medical centers (MCs) and district hospitals. In 2016, there were 72 health districts with 1760 CSPSs, 52 MCs and 47 functioning district hospitals. In 2016, the second level comprised eight regional hospitals, and the third level comprised five University Teaching Hospitals (UTH), including an exclusively pediatric center that does not perform deliveries [18].
CSPSs provide normal deliveries that are performed by midwives, auxiliary midwives, nurses or auxiliary nurses. CSPSs with midwives also perform deliveries involving dystocia, postpartum hemorrhage management and intrauterine manual vacuum aspiration. MCs have general practitioners in addition to CSPS staff, but they are not equipped for surgery.
Obstetric emergencies (cesarean section, ectopic pregnancy, eclampsia crises, etc.) are managed at district hospital, regional hospital and UTH. Cesarean section is performed by obstetrician-gynecologists, general practitioners trained in emergency surgery and nurses with three years’ training in surgery who are called “Attachés de santé en chirurgie” in Burkina Faso.
Free care policy for women
In March 2016, a decree was adopted by the government of Burkina Faso establishing a free care policy for women. It was first implemented at the CSPSs, MCs and district hospitals in three regions (Central, Hauts-Bassins and Sahel) on April 1, 2016. On May 1, 2016, it was extended to the regional hospital in Sahel and UTH in the Central and Hauts-Bassins regions before being implemented all over the country starting on June 1, 2016.
The benefit package includes antenatal care, normal deliveries and EmONC, curative care during pregnancy and up to 42 days after delivery, treatment of obstetric fistulas, screening and in situ treatment of precancerous cervical lesions for women between 25 and 55 years old and clinical screening for breast cancer starting at age 15. Antenatal care incorporates the prevention of anemia and malaria, urine testing for albumin, blood grouping, hemoglobin electrophoresis and screening for syphilis. EmONC includes dystocia, cesarean sections, laparotomy for uterine rupture or ectopic pregnancy, pre-eclampsia or eclampsia, post-abortion care and newborn intensive care.
The covered expenses for all targeted services under the free care policy include fees for consultation or surgery, prescriptions fees, paraclinical examinations (laboratory tests and medical imaging), hospitalization expenses and the expenses of ambulance transportation between health facilities. Eligible women should not be paying for these components. In fact, the state acts as a third-party payer for health facilities.
The free care policy is fully funded by the state budget. Health facilities are paid according to a fee-for-service method with scheduled fees. However, payments are made prospectively rather than retrospectively, as it is typically the case for fee-for-service payments. Through this approach, the government avoids reimbursement delays, which is considered the main barrier to the success of cost-reduction policies in Sub-Saharan Africa. In practice, the funds are pre-deposited quarterly into hospital and district accounts based on centrally determined allocation keys based on the services that each health facility is expected to provide. This estimation considers the historical utilization over the six last months. Health facilities use this money and produce monthly reports on the services provided and their expenses to the Ministry of Health. At the beginning of the next quarter, funds are transferred again in consideration of the bank account balance.
The monthly activities and financial reports of the facilities are checked at the district, regional and central levels. A sample of health facilities is surveyed monthly by four international nongovernmental organizations (NGOs) selected by the Ministry of Health. The selection of health facilities is mainly determined by the suspect nature of their reports. This survey consists of the following: (i) checking the consistency between the data transmitted in reports and those in the health facility’s registers, (ii) conducting exit interviews of a random sample of patients to measure their satisfaction with the care they received and to ensure that the drugs they received correspond with those listed in registers, and (iii) conducting household surveys to ensure that the beneficiaries reported in the registries truly exist. Cases of fraud are submitted to the administrative authorities for sanctions according to the procedures established within public service.
Study design
This was a national cross-sectional study conducted in public health facilities from September to October 2016. It was conducted by “Institut de Recherche en Sciences de la Santé” during the annual needs assessment in reproductive health funded by the United Nations Population Fund (UNFPA). A structured questionnaire was used to collect data.
Study population and sampling
The study population comprised women who had delivered or received emergency obstetric care at a public health facility during the study period.
We used a multistage stratified sampling with facility types (hospitals, MCs and CSPS) as the strata. All of the hospitals and MCs were included in the sample because they were few in number. We then selected, by simple random sampling, one-fifth of the CSPSs in each region. At each CSPS, MC and district hospital, we chose one normal delivery without episiotomy, one with episiotomy and one case for each type of emergency obstetric care (EmOC). EmOC includes dystocia with and without episiotomy, postpartum hemorrhage, intrauterine manual vacuum aspiration, cesarean section and eclampsia. The last two services are provided only at hospitals. At the regional and university hospitals, five and ten cases of each type of care cited above were randomly selected, respectively. The services selected were those covered by the free care policy. The number of cases per type of service was higher for the regional and university hospitals to ensure significant total numbers for the statistical calculations given the relatively low number of these types of health facilities.
Women who received care at the selected health facilities during the interviewer’s visits were included in the selection of the sample. The length of stay at health facilities varied from a few hours to several days according to the type of service. The sample included only women whose care for the episode had been completed and who were still present at the health facility. This limitation was to ensure that all medical expenses for the episode were included.
The sample size of women was determined by the following formula [19]:
$$ \mathrm{n}\kern0.5em =\kern0.5em \frac{1{.96}^2\kern0.5em \times \kern0.5em \mathrm{p}\kern0.5em \times \kern0.5em \left(1-\mathrm{p}\right)\kern0.5em \times \kern0.5em \mathrm{DEFF}}{{\mathrm{d}}^2} $$
where the percentage of births attended by skilled health personnel in 2016 p = 0.809, the level of absolute precision d = 0.05 and the estimated design effect DEFF = 2. We obtained a minimal sample of 475. This sample size was to ensure that the estimated expenses were representative of skilled birth attendance.
Measure of outcomes
There were two outcome variables: OOP payments and direct health care expenses. OOP payments represented the total expenses paid by each patient for the following components: fees for consultation and/or surgical intervention, prescription fees, paraclinical examinations, hospitalization and ambulance transport between health care facilities. These expenditures could be paid inside or outside the health care facility; in particular, outside payments were made for drugs and paraclinical examinations that were not available at the facility. Unofficial payments to health professionals for drugs or care were included in the calculations. The data sources were payment receipts and patient reports. Patient reports were double-checked with the receipt except for unofficial payments (payments to health professionals, for cleaning products, etc.) that had no receipt. However, when unofficial payment practices were found to exist in a health facility, they were common and we confirmed them with other patients. Only expenses eligible for free care were included. Consequently, expenditures for food and transportation (except by ambulance) were not included in the OOP payments.
We defined the direct health care expenses of a service as the expenses covered by the free care policy. This definition includes expenses charged to the Ministry of Health by the health facility and the OOP expenses borne by the patient. The free care policy sheets that summarize the services provided for each patient and their expenses were additional data sources used. This information was cross-checked with medical prescriptions and reports obtained from the patients.
The expenditures were calculated in local currency (XOF) and then converted into US dollars using the average exchange rate for 2016 (US$1 = XOF592.912968).
Independent variables
In the study of factors associated with OOP payments, the independent variables included patient sociodemographic characteristics (age, education level, marital status, place of residence, parity at admission) and health system-related characteristics (health region, type of health facility, type of service and the service provider’s qualifications). Sociodemographic characteristics were collected from the patients, and health system-related characteristics were obtained from health professionals and by checking the registries.
Age was collected as a continuous variable and categorized into 5-year intervals. Parity was categorized into three groups: nulliparous, multiparous (1–4 deliveries) and grand multiparous (at least 5 deliveries). Health facilities were also grouped into three categories (CSPS, MCs and hospitals).
The level of education was divided into three categories: none, primary and secondary or higher level. Marital status included two categories, married and not married, whereas the service provider qualifications comprised five categories: physician, midwife, nurse, auxiliary midwife and surgeon’s assistant. The place of residence was divided into two categories, rural and urban, according to the national classification of the municipalities.
Data collection
Data were collected from September to October 2016 by physicians and medical students. In total, there were sixty-six (66) data collectors, organized into 18 teams. Each team was supervised by a team leader. There were an average of six data collectors per region. They received a two-day training and conducted a pilot test. Each team leader was responsible for ensuring that all the forms were completed. In addition, we conducted two five-day supervisions with three other teams of two supervisors during data collection to ensure that the survey was properly conducted and that data quality was maintained.
The types and quantities of drugs and consumables used and paraclinical examinations performed were reported in the questionnaire. The prices of drugs and consumables were those charged by the health facility pharmacy or those reported on the prescriptions in cases where patients had paid at private pharmacies. Consultation fees, paraclinical examinations and hospitalization prices were specific to health facilities and were specified on health tariff policy sheets or in reports from the relevant service providers (the laboratory, for example).
In cases of discrepancy between the expenses reported on the free care sheets and the prices obtained from medical prescriptions or examination reports, we checked with health professionals.
Data processing and analysis
We performed a double data entry with two trained and supervised data entry clerks using Epi Data. The data were then exported to Stata version 15.1 (Stata Corporation, Texas, USA) for quality check and analysis.
The analysis was conducted in two phases. First, we determined the direct health expenditures for normal delivery and each EmOC. For this purpose, descriptive statistics were used to describe the sample. Then, direct health care expenses were standardized by type of service, and observations with standardized values above 3.29 were deemed extreme [20]. Five such cases were noted, including four deliveries that required the administration of anti-D immunoglobulins and one case of postpartum hemorrhage. The four deliveries were excluded from the analysis because the use of anti-D immunoglobulins is a specific service, and too few patients received this service to form a separate group. Two other cases (cesarean section and dystocia) that did not meet the extreme values criteria but also required anti-D immunoglobulin were also excluded from the analysis. Then, we computed the mean and median of direct expenses for each service type at the national level and by type of health facility. We also calculated the median and mean of OOP payments separately for normal delivery and EmOC.
In the second phase, we investigated the factors associated with OOP payments. We used crosstabulation to ensure that there was no systematic relationship between certain independent variables. Women who made OOP payments were then compared according to their characteristics and the health facility characteristics using Pearson’s chi-squared test.
For the analysis of health expenditures data involving only nonnegative values, several authors recommended a two-part model approach [16, 21,22,23]. In this approach, the probability of OOP payment was first modeled, and then the amount of payment for those who paid was modeled. In general, the first part uses a probit or logit regression [21, 23]. We chose a logit regression because the interpretation of its results is quite straightforward. For the second part, the Box-Cox test (p = 0.958 for λ = 0) showed that the data are fit for use of the generalized linear model (GLM) with a link log. For the choice of family distribution, the authors recommended mainly Gamma or Inverse Gaussian family for health expenditures that exhibited skewness [21, 23]. Upon using the modified Park test, the result (p = 0.6167 for δ = 2) showed that the data are gamma distributed. Thus, the first part model may be conceptualized as follows:
$$ \mathrm{In}\kern0.5em \left[\frac{\mathrm{Prob}\left(\mathrm{y}>\left.0\right|\mathrm{x}\right)}{1\hbox{-} \mathrm{Prob}\left(\mathrm{y}>\left.0\right|\mathrm{x}\right)}\right]\kern0.5em =\kern0.5em \upalpha \kern0.5em +\kern0.5em \sum {\upbeta}_{\mathrm{i}}\kern0.2em {\mathrm{x}}_{\mathrm{i}} $$
where y represents the OOP payments, α is the constant term, xi represents a set of independent variables and βi are the estimated coefficients for these variables. Prob(y > 0| x) represents the probability that a patient will experience an OOP payment for normally free of charge service.
For the second part, the equation is the following:
$$ \mathrm{In}\kern0.5em \left[\mathrm{E}\left(\left.\mathrm{y}\right|\mathrm{x}\right)\right]\kern0.5em =\kern0.5em \upalpha \kern0.5em +\kern0.5em \sum {\upbeta}_{\mathrm{i}}\kern0.2em {\mathrm{x}}_{\mathrm{i}} $$
where E(y) represents the expected value of OOP expenses. The other notations share similar definitions as those in the first part of the model.
The two parts may be analyzed separately or together. However, analyzing the two parts together allows for prediction of the OOP expenses based simultaneously on the two models. We opted for the combined analysis but took advantage of the broad range of tools available in logistic regression and GLM to check each model individually. Observations that did not fit the models were checked and removed. For the multiple regression, we regrouped the different services into two categories, normal deliveries and EmOC, because certain EmOC services were exclusively offered in hospitals. We excluded the provider qualifications variable of the multiple regression analysis because the type of service performed depended on the provider’s qualifications. We also standardized the outcome variable OOP expenses for the second part of the analysis and excluded standardized values above 3.29 (two observations) from the analysis. The analysis was computed using the Stata command “twopm”. The statistical threshold was set at 0.05 for all statistical tests. Standard Errors were adjusted for the clustering of data at the district level. The parity variable was excluded from the final model because it did not significantly contribute to the model. Finally, we used the margins command to predict the amount of OOP payments for certain variables.
Ethical considerations
This study was part of the 2016 reproductive health needs assessment and was approved by the Health Research Ethics Committee of Burkina Faso. The patients gave informed consent. In addition, survey data confidentiality was ensured by the anonymity of the collection tools.