Garchitorena A, Miller AC, Cordier LF, Ramananjato R, Rabeza VR, Murray M, et al. In Madagascar, use of health care services increased when fees were removed: lessons for universal health coverage. Health Aff (Millwood). 2017;36(8). Appendix Section 1. Summary of interventions to improve health care access in Ifanadiana The district of Ifanadiana has benefited since the beginning of 2014 from several programs and interventions to increase health care access in the population. The aim of this section is to provide a comprehensive understanding of the two programs that reduced financial barriers for the targeted populations, as well as other non-financial interventions that were implemented during the same period and which could have influenced health care access. The district of Ifanadiana, with an estimated population of nearly 193,000 inhabitants in 2015, is divided in 13 communes, each with one major health center (CSB2). Six of the larger communes (population at least 15,000) have an additional, more basic, health center (CSB1). By Ministry of Health (MoH) norms, CSB1s only provide the most elementary health services, such as vaccinations and family planning, and are staffed by a nurse or a midwife, whereas CSB2s provide all types of primary care and are staffed with at least two trained practitioners, a doctor and either a nurse or a midwife. The two reimbursement programs explained below were implemented only in CSB2s: the PAUSENS program was implemented in all thirteen, whereas the PIVOT program was implemented only in four of those that were part of the initial catchment area of the organization (Exhibit A).
1. World Bank project PAUSENS The project Emergency Support to Critical Education, Health and Nutrition Services (PAUSENS), funded by the World Bank, was implemented in 5 out of 22 regions of Madagascar to provide critical services to vulnerable populations. While the project included multiple interventions to improve nutrition, education and health (full details are available at [1]), the intervention relevant to this study was a financial program aimed at improving access to health care for children under 5 and pregnant women. The program provided a basic package of services free of charge at the health facility level for these vulnerable groups to remove out-of-pocket spending, through a voucher reimbursement system managed by three local NGOs in Ifanadiana (Taratra, Maintso an Ala and Cafed/Fafed). The basic package of services for pregnant women, from antenatal to postnatal care, included: prevention of mother-to-child transmission, treatment of syphilis, prevention and treatment of HIV/AIDS, supplementation with iron and folic acid, tetanus vaccination, intermittent preventive medication against malaria, and distribution of safe delivery kits. The basic package for children under 5 included: activities to promote good practices, breastfeeding and nutrition, Vitamin A supplementation, vaccinations, distribution of bed nets, treatment of diarrhea, prevention and treatment of malaria, and support to integrated management of childhood illness (IMCI), covering the full cost of treatments. The project also included trainings, support for child vaccination in remote areas, and donations to health centers to improve health care quality such as furniture, medical equipment, cellphones or solar-powered refrigerators. Implementation of the program in Ifanadiana started in February 2014 and covered all thirteen CSB2s in the district. In each of these CSB2s, a program voucher agent would be present, and every woman attending the health center for antenatal, delivery or postnatal care (first 6 weeks), or for any children under five illness would receive a voucher from the agent. After consultation at the health center, medicines prescribed by MoH health staff that were included in the program would then be provided free of charge through presentation of the voucher both at the dispensary and to the program agent. The patient would not disburse any money out-of-pocket for these medicines, since PAUSENS would directly reimburse the dispensary on behalf of the patients on a periodic basis. By December 2015, 55,979 patients of the target population had benefited from the program at an average cost per patient of $ 0.55. 2. PIVOT reimbursement program The non-governmental organization PIVOT started activities in early 2014 in Ifanadiana, as part of a new partnership with the MoH to strengthen the existing health system in the district. Although
some of the activities spanned the entire district, such as staffing of health centers and trainings, the initial catchment area where most programs were implemented covered 4 out of 13 communes. In the CSB2s of these four communes, PIVOT implemented a reimbursement program, complementary to PAUSENS, aimed at improving access to health care for all patients. The program started in October 2014 and reimburses the health center for the costs of 40 essential drugs and 20 consumables (Exhibit B) on behalf of patients of all ages that visited the health center for outpatient care. No voucher is needed, the patient obtains at the dispensary the medicines and consumables used or prescribed by the health care provider, and a record is kept at the dispensary for reimbursement through the nongovernmental partner on a monthly basis. By December 2015, 34,221 patients of the target population had benefited from this program at an average cost per patient of $ 0.69. The combination of the two financial programs in these four CSB2s virtually eliminated point-of-service payments for all patients seeking any type of primary care at the health care center.
Exhibit B. List of essential drugs and consumables available for free Medicines Albendazole (400 mg tab) Gentamicin (40 mg/ml inj, 0.3% collyrium) Aminophylline (100 mg tab, 25 mg/ml inj) Amoxicillin (250 mg/500 mg gel, 250 mg/5 ml oral susp) Amoxicillin/Clavulanic acid (125 mg/500 mg tab) Ampicillin (500 mg gel, 500 mg inj, 1 g inj) Hydrochlorothiazide (25 mg tab) Hydrocortisone (100 mg inj) Consumables Tongue depressor 70% ethanol Elastic, non-adhesive bandages (10cm x 4m) Ibuprofen (200 mg tab) Catheter (G18, G20, G22) Lactated Ringer's solution (500 ml) Sterile gauze pads (10cm x 10cm) Artesunate (10 mg/ml inj) Lidocaine hydrochloride (2% inj) Hydrophilic cotton (50g) Atropine sulfate (0.25/1 mg/ml inj) Benzathine benzylpenicillin (2.4 MUI inj) Metoclopramide (10 mg tab, 5 mg/ml inj) Metronidazole (250 mg tab, 5 mg/ml inj) Non-absorbable suture (polyester) Absorbable suture (polyglactine) Fortified penicillin procaine (1 MUI inj) Nifedipine (10 mg tab) Non-sterile latex exam gloves Calcium gluconate (100 mg/ml inj) Nystatin (100,000 UI tab, ointment, 30 ml oral susp) Blood-sampling tourniquet Captopril (25 mg tab) Paracetamol (500 mg tab, 125 mg susp) Hydro alcoholic gel Ceftriaxone (1 g inj) Phenobarbital (50 mg tab, 40 mg/ml inj) Surgical knife blade Chlorpheniramine (4 mg tab) Praziquantel (600 mg tab) IV infusion set Ciprofloxacine (500 mg tab) Quinine (300 mg tab, 300 mg/ml inj) Urine Drain Bag (2L) Cotrimoxazole (120 mg tab, 400-80 mg tab, 200-40 mg/5ml oral susp) Salbutamol aerosol (100µgr/Dose, 0.5 mg inj) Povidone-iodine 10% (500 ml) Diazepam (5 mg tab, 5 mg/ml inj) Oral rehydration solution (sachet) Feeding syringe (50ml) Sterile water for injection (5 ml, 10 ml) 5% glucose solution (500 ml) Hypodermic syringe (2 ml, 5 ml, 10 ml) Erythromycin (250 mg tab, 125 mg syrup powder) 10% glucose solution (500 ml) Foley vesical catheter (CH16 CH20) Iron-folic acid (200-40 mg tab) Normal saline solution (500 ml) Adhesive tape (10cmx5m) Furosemide (40 mg tab, 10 mg/ml inj) Tetracycline (1% ointment, 250 mg tab) tab = tablet ; inj = injectable ; susp = suspension ; 3. PIVOT non-financial programs Strengthening the district s referral program: PIVOT set up a referral system in Ifanadiana district in February 2014, allowing patients to be referred by a network of ambulances from one of the district health centers to seek higher level of care (district hospital or higher). To this day, more than 2,000 patients have been referred free of transport cost, in an equipped ambulance with medical assistance. Patients referred by PIVOT are then covered for and accompanied throughout their hospitalization to further reduce social and financial barriers to care. Reinforcing infrastructure at health facilities: PIVOT has carried out major renovations and improvements in the infrastructure of the four PIVOT-supported health centers. Besides the external
restoration of the building, renovations ensure the availability of clean facilities, waste management, beds, reliable power, lighting, clean water, examination and delivery tables, sterilization equipment, and basic diagnostic equipment. Key dates include: Renovation of the CSB2 Ranomafana maternity ward (June 2014), followed by full renovation of the health center (March 2015); full renovation of CSB2 Ifanadiana (November 2015); full renovation of CSB2 Tsaratanana by external partner FID (November 2014); Building of temporary OPD and IPD consultations rooms at CSB2 Kelilalina (October 2014), followed by Full renovation of the health center (ongoing) Strengthening the health workforce: PIVOT has a joint recruitment program with the Ministry of Health (MoH) to ensure that all health centers have at least two trained health practitioners, including a midwife and a nurse, in order to bring the health workforce in the Ifanadiana district up to MoH norms. This was done in support of the district s rapid response initiative to reopen all closed health centers. By December 2015, nine health centers in the district that were not up to MoH norms have benefited from this joint recruitment, including the CSB1 in Ambodiara and the CSB2s in Ranomafana, Kelilalina, Androrangavola, Tsaratanana, Ambohimiera, Atsindra, Ifanadiana, and Antaretra.
Exhibit D. Timeline for major financial and non-financial programs taking place in Ifanadiana Health Centers between January 2013 and December 2015. Health Center PAUSENS PIVOT Infrastructure Ambulance fee-exemptions fee-exemptions renovations 1 referral program 2 CSB2 Ambohimanga du Sud February 2014 - - May 2014 (71) CSB1 Ambodimanga Nord - - - - CSB2 Ambohimiera February 2014 - - January 2015 (17) CSB1 Maromanana - - - - CSB2 Analampasina February 2014 - - September 2015 (1) CSB2 Androrangavola February 2014 - - April 2014 (69) CSB1 Mahasoa - - - - CSB2 Antaretra February 2014 - - April 2014 (160) CSB2 Atsindra February 2014 - - August 2015 (2) CSB2 Fasintsara February 2014 - - November 2015 (1) CSB1 Ambodiara Sud - - - September 2015 (2) CSB2 Ifanadiana February 2014 October 2014 November 2015 April 2014 (364) CSB1 Ekar Ifanadiana - - - - CSB2 Kelilalina February 2014 October 2014 October 2014 February 2014 (182) CSB2 Maroharatra February 2014 - - November 2015 (1) CSB1 Ambalavolo - - - - CSB2 Marotoko February 2014 - - October 2014 (20) CSB2 Ranomafana February 2014 October 2014 June 2014 March 2014 (253) CSB2 Tsaratanana February 2014 October 2014 November 2014 February 2014 (299) 1 Date of completion of the first major renovation, as explained above; 2 Date of first referral by CSB, and total number of referrals by December 2015 (in brackets) Section 2. Variable construction and transformation 1. Analysis of health care access at baseline based on household surveys a. Response variables Access to health care: we defined access to health care as the probability of seeking medical care at a health care facility (health center or hospital) when an individual was in need due to reported illness or pregnancy. We constructed three indicators to explore health care access in different populations: all household members, children under five and pregnant women. All three indicators were constructed as dummy variables to be used in binomial regression models, reflecting whether households or individuals sought treatment at health care facilities or not (Exhibit E). For the household model, we aggregated information from the household questionnaire on number of household members reporting being sick in the previous 4 weeks, and number of reported household outpatient medical consultations in the same period. For children under five, we aggregated information from the women questionnaire reporting on their children s illnesses in the previous two weeks (fever, diarrhea or ARI) and health seeking behavior for these different illnesses. For maternal care, we aggregated information from the women questionnaire on health seeking behavior during the last pregnancy for prenatal care, delivery and postnatal care, for those women who reported having a live birth in the previous 5 years. For each pregnancy, we defined as a success
having sought care at a health care facility for at least one prenatal visit, for delivery and for at least one postnatal visit within 2 days. Exhibit E. Construction of outcome variables of health care access for the survey models Model Health Care Need Health Care Access 1. Household Did at least one household member report being sick in the past 4 weeks? No = 0 Did the household report at least one outpatient visit at a health center? NA Yes = 1 No = 0 2. Children <5 Did the child have diarrhea or fever or respiratory symptoms in the past 2 weeks? No = 0 Yes = 1 Yes = 1 No = 0 3. Maternal Did the woman had a live birth in the past 5 years? No = 0 Did the parents sought treatment at a health care facility for their child illness? NA Yes = 1 Yes = 1 Any No = 0 Did she seek treatment at a health care facility for at least one antenatal visit, delivery and at least one postnatal visit? NA All Yes = 1 b. Explanatory variables Socio-economic variables: per capita household income was measured as the combined annual income reported by all adults in the household ( 15 years) from primary and secondary activities divided by the number of household members. A wealth score was constructed for the population in Ifanadiana following standard methods used in the DHS. Briefly, household physical assets were included in a Principal Component Analysis (PCA) and then the scores from the first principal component were used. Wealth scores were then normalized between 0 and 100. Average household education was estimated as the total number of years of schooling for adults ( 15 years) averaged over the total number of adults in the household. We also included the years of education of the head of the household (household model only) and the years of education of the mother (children and maternal model only). Geographic variables: since the best spatial resolution was the cluster level due to data confidentiality, we estimated the centroid for each cluster and calculated the shortest Euclidian distance from each cluster to the health center, to the main paved road and to secondary non-paved roads, using R packages maptools and spdep [2,3]. The distance values for each cluster (in kilometers) was allocated to all 20 households within the cluster. We also constructed dummy variables representing whether or not the clusters distance to each of these geographical features was higher to 5 km. Demographic variables: we used different indicators in each model to control for the demographic characteristics of the individuals accessing care or ill (age, sex), which could influence health-seeking
behaviors. For the household model, we used dummy variables to reflect whether, among sick household members, there were children under 5, elder over 50, or predominance of males. For the children model, we used the age and sex of the ill child, and the age of the mother. For the maternal model, we used the age of the mother and the order of the child (first child, second, etc.). Severity variables: we used different indicators in each model to control for factors related to the health-related episode that could influence seeking behaviors, such as severity and complications during the reported illness, or an atypical pregnancy. For the household model, we estimated total number of days lost to disease by all household members in the last 4 weeks and the total number of household members that reported being sick in the last 4 weeks. For the children model, we constructed a dummy variable reflecting whether the child presented visible complications during the reported illness episode (blood in feces during diarrhea, tachypnea during respiratory illness). For the maternal model, we included dummy variables for mothers reporting that their child was smaller than the average, which could be a sign of prematurity and influence both delivery and postnatal care seeking behavior. c. Data analysis Three sets of analyses were carried out in order to study separately access to health care in the general population (all household members) and the two vulnerable populations, children under 5 and pregnant women. For each, exploratory analyses were first carried out, outliers were removed and skewed variables were log-transformed (transformed variables are specified in Exhibit H). Access to health care was modelled using logistic regressions in generalized linear models. First, the effect of socio-economic and geographical factors were studied through univariate analyses. Demographic factors as well as factors related to the severity of the illness episode or pregnancy that could influence seeking behaviors (referred to as severity factors ) were also studied through univariate analyses to identify relevant confounders. Second, explanatory variables with p-values less than 0.1 in univariate analyses were included in multivariate models. Step-wise model selection was based on adjusted Akaike Information Criteria (daic) for complex surveys (22), whereby we selected the reduced model with the lowest daic among all potential candidates. Model assumptions were verified in the final models, including multicollinearity (estimation of variance inflated factors for each explanatory variable), violations to homogeneity (plots of residuals vs. fitted values and residuals vs. explanatory variables) and independence of residuals (spatial correlograms to study spatial autocorrelation of residuals). Analyses were performed with R software using the survey procedures available in R package survey, which account for complex survey designs in variance estimation.
d. Ethics, consent and permissions for survey data All French and Malagasy questionnaires, data collection and analysis methods were standardized and had been reviewed and approved the year prior to the study by the Madagascar National Ethics Committee and re-reviewed through internal processes at INSTAT. The study was also reviewed by Harvard Medical School IRB. All adults ( 15 years) provided verbal consent for the in-person interview and anthropometric measurements. Parents or guardians provided consent for children 5 years of age. INSTAT provided survey data to the investigators with all individual identifiers removed and with geographical information at the cluster level. 2. Analysis of trends in health care utilization based on health system data a. Response variables Utilization rates: we defined health care utilization rate as the number of new individual consultations per month at each public health center. Consistently with the analyses of baseline population data, three indicators were constructed to reflect utilization rates for the whole population, for children under five and for maternal care (pregnant women). Utilization rates for maternal care were estimated as the sum of consultations per month at each public health center for antenatal care (first visit), delivery and postnatal care. b. Explanatory variables Time-specific variables: we introduced a time variable to account for the linear trend in utilization rates in the absence of interventions. For each health center, we created a sequence where each unit was the month since the beginning of the study, from 1 (January 2013) to 36 (December 2015). We also introduced a seasonal effect to account for seasonal differences in utilization rates (i.e. due to malaria incidence) and to avoid temporal autocorrelation in our models. Seasonality was included in the model by transforming the month of the year with a sine function of annual period: f t = sin 2π t + θ 12 Where t is the month of the year and θ is the horizontal shift for the sine function, as previously described [4]. To select the appropriate seasonal effect for each model, sine functions with different horizontal shifts were fitted to the observed utilization dynamics at district health centers in univariate GLMMs, and the one that best fitted the data (lowest AIC) was retained for inclusion in the multivariate model. A different seasonal effect was retained for outpatient care of all patients and children under 5 (θ = 1) than for maternal care (θ = 5). Health center type: we introduced two dummy variables to study baseline differences in utilization rates between the health centers, independently of the programs. One variable discriminated
between CSB2s, the reference health center for each commune, with at least a medical doctor and nurse or midwife (these benefited at least from PAUSENS program), and CSB1s, a health center with only a nurse or midwife that provides more basic care (these did not benefit from any reimbursement program). In addition, we constructed a second variable to further discriminate the four PIVOT-supported CSB2s from the rest (these benefited from both PAUSENS and PIVOT programs). Financial and non-financial programs: we modeled the level of change in utilization rates associated with the two reimbursement programs with dummy variables reflecting when and where each program (PAUSENS or PIVOT) was in place. We also introduced two time variables to account for the change in slope in utilization rates after each program was in place. For this, we created a sequence where each unit was the month, from 1 at the beginning of each program (February 2014 for PAUSENS; October 2014 for PIVOT) to N (December 2015). We also included two non-financial variables to control for changes in infrastructure and medical workforce at each health center, which could have an impact on utilization. For workforce, we used the number of health care staff (doctors, nurses and midwives) in each health center per month, which was obtained through MoH official data and completed through interviews with staff at every health center. For infrastructure, we introduced a dummy variable reflecting completion of major renovations at each PIVOTsupported health center (see Exhibit D). c. Data analysis Statistical models were developed for each of the three outcome variables of health care utilization explained above: outpatient care for all patients (excluding maternal care), outpatient care for children under 5, and maternal care (including all prenatal, delivery and postnatal care visits). A powerful statistical technique to model longitudinal data with repeated measures is the use of mixed effects models, in which subjects or groups (i.e. health center) are treated as random effects to adjust for the dependency of the observations within groups [5]. The dataset available for our study was composed of 19 health centers followed for 36 months, resulting in a sample size of 684 observations (616-628 after accounting for missing data and lagged utilization), which suggested that mixed effects models were appropriate for our study [5]. Utilization rates at each health center were modelled using Negative Binomial regressions in generalized linear mixed models (GLMMs), with a random intercept introduced for each health center. This model was selected to correct for overdispersion observed in poison models that were initially tested. All explanatory variables in this model were introduced as fixed effects without interactions. Each explanatory variable described above was studied through univariate analyses, and those with p-values bellow 0.1 were included in multivariate models. Model selection was performed through step-wise procedures based on AIC, by selecting the reduced model with the
lowest AIC. Model assumptions in the final model were verified, including violations to homogeneity and independence of residuals. Variables responsible of substantial multicollinearity (variance inflation factor > 3) were removed. We introduced a 1-month utilization lag in the final models to remove significant temporal autocorrelation in the residuals. To facilitate interpretation of results, we report exponentiated model coefficients, which reflect the ratio of change in utilization rates associated with each explanatory variable. All analyses were performed with R software [6] and R package lme4 [7]. Section 3. Detailed results 1. Barriers to health care access in the population at baseline Exhibit F. Spatial distribution of household health-seeking behavior and wealth in Ifanadiana District, 2014. Left map displays the mean proportion of households seeking care at a health center when a member of the household reported illness in the previous 4 weeks. Right map displays the average wealth score of households (score normalized between 0-1), classified by Jenks natural breaks. Besides clear geographical barriers, an important spatial overlap can be observed between the district s poorest areas (average wealth score < 0.07) and areas of low health care access (<15%). Maps were developed from 2014 survey data with ArcMap, version 10.2.2, by performing natural neighbor interpolation based on the values of 80 spatial clusters, which include 20 households each. Interpolation could only be performed for areas situated between at least two surveyed villages, leaving some parts of the district blank (notably near the edges of the district boundary).
Exhibit G. Summary statistics for variables (outcome and explanatory) included in the models of health care access from household survey data Household model Children model Maternal model Variable Mean (Std dev.) N observ. Mean (Std dev.) N observ. Mean (Std dev.) N observ. Health care access HH member ill past 4 weeks (%) 87.25 (1.01) 1522 Children <5 ill past 2 weeks (%) 58.01 (1.83) 1391 Live birth past 5yrs (%) 76.19 (1.75) 1286 Of those above, % who sought medical care 27.65 (2.44) 1324 22.71 (2.67) 736 16.56 (2.95) 986 Socio-Economic Factors NAs NAs NAs Wealth Score (normalized) 10.82 (1.53) - 9.15 (1.01) - 9.3 (1.11) - HH per capita Income (USD) 108.11 (9.82) 17 91.59 (9.67) - 86.41 (7.77) - Education of Head of HH 1 of Mother 2,3 (years) 2.52 (0.2) 10 2.06 (0.14) - 2.15 (0.17) - HH Average Education 1 (years) 2.6 (0.19) 1 2.32 (0.14) - 2.31 (0.15) - Geographical Factors Distance from CSB (km) 5.11 (0.45) - 5.36 (0.5) - 5.51 (0.46) - Distance from CSB > 5km (%) 45.86 (6.38) - 49.19 (6.93) - 50.89 (6.42) - Distance from Main Road (km) 26.43 (2.63) - 28.88 (2.76) - 26.69 (2.5) - Distance from Main Road > 5km (%) 75.85 (5.06) - 81.12 (4.3) - 78.43 (4.59) - Distance from Secondary Road (km) 6.64 (0.6) - 7.28 (0.64) - 6.96 (0.55) - Distance from Secondary Road > 5km (%) 55.64 (6.38) - 63.08 (6.59) - 59.6 (6.17) - Demographic Factors Number of HH members 5.69 (0.1) - 6.39 (0.15) - 6.42 (0.13) - More Sick Males (%) 1 Child Sex 2 40.27 (1.61) - 52.06 (2.15) - Sick Under 5 (%) 1 Child Age 2 Child Order 3 58.5 (1.66) - 1.7 (0.06) - 4.02 (0.11) - Sick Over 50 (%) 1 Mother Age 2,3 23.03 (1.35) - 27.58 (0.38) - 27.84 (0.25) - Severity Factors Days Lost 1 Complication (%) 2 Small Child (%) 3 2.25 (0.11) 43 61.96 (2.28) 147 32.51 (1.76) - Number of sick HH members 1 2.93 (0.06) - 1 Household model only; 2 Children model only; 3 Maternal model only
Exhibit H. Factors associated with health care access in Ifanadiana District (binomial GLMs with survey design, univariate and multivariate results) Household model Children model Maternal model (N=1324) (N=736) (N=986) Univariate Multivariate Univariate Multivariate Univariate Multivariate Variable OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI) Base Probability (Intercept) 0.13 (0.06-0.24)*** 0.38 (0.26-0.51). 0.34 (0.1-0.71) Socio-Economic Factors Wealth Score (log 10 ) 1.91 (1.23-2.98)** 2.7 (1.34-5.46)** 14.93 (6.69-33.3)*** 3.51 (1.27-9.72)* HH per capita Income (log 10 ) 1.96 (1.5-2.56)*** 1.49 (1.12-1.98)** 1.88 (1.11-3.21)* 3.79 (2.29-6.28)*** Education of Head of HH 1 of Mother 2,3 (log 10 ) 1.1 (1.05-1.15)*** 2.86 (1.38-5.94)** 2.11 (1.02-4.39)* 13.7 (7.01-26.77)*** 3.26 (1.3-8.16)* HH Average Education (log 10 ) 1 1.12 (1.05-1.19)*** Geographical Factors Distance from CSB (km) 0.9 (0.83-0.97)** 0.8 (0.74-0.86)*** 0.81 (0.75-0.88)*** 0.65 (0.57-0.75)*** 0.74 (0.64-0.86)*** Distance from CSB > 5km 0.5 (0.33-0.74)*** 0.64 (0.4-1.03). 0.3 (0.19-0.49)*** 0.14 (0.07-0.27)*** Distance from Main Road (km) 0.99 (0.98-1.01) 0.99 (0.97-1.01) 0.97 (0.94-1.01) Distance from Main Road > 5km 0.45 (0.28-0.72)** 0.5 (0.28-0.88)* 0.39 (0.23-0.69)** 0.19 (0.09-0.39)*** Distance from Sec. Road (km) 0.99 (0.93-1.04) 0.98 (0.91-1.05) 0.96 (0.86-1.08) Distance from Sec. Road > 5km 0.83 (0.52-1.33) 0.81 (0.43-1.53) 0.66 (0.3-1.44) Demographic Factors Number of HH members (log 10 ) 1 (0.95-1.05) 0.37 (0.11-1.22) 0.12 (0.03-0.45)** 0.15 (0.04-0.56)** More Sick Males (yes/no) 1 Child Sex 2 0.88 (0.64-1.2) 1.2 (0.77-1.87) Sick Under 5 (yes/no) 1 Child Age 2 Child Order (log 10 ) 3 1.28 (1.03-1.6)* 1.53 (1.19-1.98)** 1.02 (0.92-1.13) 0.29 (0.15-0.53)*** Sick Over 50 (yes/no) 1 2,3 Mother Age (log 10 ) 1 (0.74-1.36) 0.98 (0.17-5.74) 0.8 (0.17-3.69) Severity Factors Days Lost (log 10 ) 1 Complication 2 Small Child (yes/no) 3 4.83 (3.11-7.5)*** 4.91 (3.02-7.98)*** 1.32 (0.82-2.13) 0.75 (0.45-1.24) Number of sick HH members 1 1.12 (1.04-1.21)**. p<0.1; * p<0.05; ** p<0.01; ***p<0.001; 1 Household model only; 2 Children model only; 3 Maternal model only
Impact of user-fee elimination on health care utilization Exhibit I. Monthly utilization rates before and after removal of financial barriers in all 19 health centers in Ifanadiana District, 2013-2015. Utilization rates are shown separately for all outpatient consultations (top), outpatient consultations of children under five (middle) and maternal care consultations (prenatal, delivery and postnatal care visits; bottom). Left panels show average number of consultations per month in the four health centers benefiting from both interventions (blue lines); in the nine health centers benefiting only from PAUSENS intervention (red lines), and in the six health centers not benefiting from any intervention (green lines). Right panels show model predictions from multivariate results shown in Table 3, averaged for each group of health centers. Solid lines represent predictions when interventions are in place, and dotted lines represent predictions in the hypothetic case where each of the two interventions had not been implemented.
Exhibit J. Impact of reduction of financial barriers on health care utilization in Ifanadiana (negative binomial GLMMs, univariate and multivariate results) Outpatient Model (N=616) Child Model (N=628) Maternal Model (N=627) Univariate Multivariate Univariate Multivariate Univariate Multivariate Estimate (95% CI) Estimate (95% CI) Estimate (95% CI) Estimate (95% CI) Estimate (95% CI) Estimate (95% CI) Base monthly consultations (Intercept) 13.34 (9.57-18.59)*** 6.5 (4.92-8.6)*** 8.3 (5.6-12.3)*** Underlying trends Linear trend 1.06 (1-1.12). 1.11 (1.04-1.18)** 1.11 (1.08-1.15)*** Seasonal trend 1.46 (1.38-1.55)*** 1.32 (1.25-1.39)*** 1.52 (1.43-1.63)*** 1.35 (1.28-1.43)*** 0.96 (0.93-1). 0.94 (0.91-0.98)** Lagged trend (1-month lag) 1.8 (1.7-1.91)*** 1.5 (1.41-1.59)*** 1.89 (1.78-2.01)*** 1.55 (1.46-1.65)*** 1.44 (1.34-1.55)*** 1.27 (1.18-1.37)*** Differences between CSBs All CSBs 2 (vs. CSB1s) 3.57 (2.08-6.11)*** 1.6 (1.23-2.07)*** 3.55 (2.21-5.71)*** 1.44 (1.12-1.84)** 3.76 (2.36-6)*** PIVOT CSBs 2 (vs. all others) 4.5 (2.38-8.49)*** 3.58 (1.85-6.95)*** 2.89 (1.36-6.16)** Total catchment population 1.13 (1.07-1.2)*** 1.12 (1.06-1.18)*** 1.16 (1.11-1.22)*** 1.09 (1.06-1.13)*** Density of catchment pop. 1.03 (1.02-1.05)*** 1.03 (1.01-1.04)*** 1.04 (1.02-1.05)*** Impact of PAUSENS financial program Immediate (level of change) 1.49 (1.33-1.67)*** 1.13 (1.02-1.24)* 1.79 (1.58-2.03)*** 1.29 (1.16-1.44)*** 1.32 (1.23-1.41)*** Over time (slope of change) 1.26 (1.15-1.38)*** 1.35 (1.22-1.49)*** 1.27 (1.2-1.33)*** 1.14 (1.08-1.21)*** Impact of PIVOT financial program Immediate (level of change) 2.57 (2.14-3.1)*** 1.52 (1.29-1.78)*** 2.35 (1.9-2.9)*** 1.29 (1.09-1.53)** 1.48 (1.32-1.66)*** Over time (slope of change) 2.28 (1.73-3.02)*** 2.04 (1.49-2.79)*** 1.53 (1.32-1.77)*** Impact of non-financial programs Medical staff 1.24 (1.16-1.33)*** 1.09 (1.04-1.16)** 1.25 (1.16-1.35)*** 1.09 (1.03-1.16)** 1.18 (1.13-1.23)*** 1.09 (1.04-1.13)*** Infrastructure 2.4 (1.93-2.97)*** 2.25 (1.77-2.86)*** 1.5 (1.32-1.7)*** Referral network 1.04 (1.03-1.05)*** 1.04 (1.03-1.06)*** 1.02 (1.01-1.02)*** * p<0.05; ** p<0.01; ***p<0.001; All models include a random intercept at each health center.
Section 4. Results from supervisions on health center data quality Quality of data from official health management information systems (HMIS) in low resource settings can be particularly challenging [8]. In an effort to strengthen HMIS at the district level and limit potential biases stemming from poor official reporting of data, four joint MoH-PIVOT supervisions were carried out during 2015 in six health centers of the district, including the four supported by PIVOT. During each supervision, data from the health center paper registries, containing each individual visit, was used to calculate a number of indicators per month such as total outpatient visits (all patients) and maternal care consultations (only antenatal visits and births were validated, not postnatal visits). Values for each indicator were then compared to those reported in the monthly report to the District (RMA) and the errors in the RMA estimates for each health center and month were calculated. One trimester of 2015 was validated during each supervision. The validated data available from these six health centers was used in the analyses instead of the original RMA values. Errors for outpatient consultations are only reported from January to April 2015, since official reporting of utilization rates by the MoH changed in May 2015, and all outpatient utilization data from that point on was gathered directly from the registries for all 19 health centers to obtain consistent data throughout the time series. Results from these supervisions (Exhibit K) showed that more than 70% of data reported for outpatient consultations (all patients) had errors of 0 to 25 visits. This represented an error in visits below 5% of the true value for 75% of data. However, errors were substantial in two out of 24 reports, amounting to differences of more than 20% of the true value. Reporting of maternal care visits was more accurate, with errors smaller than 10 visits in nearly 90% of reported maternal care visits. This represented an error in maternal care visits below 5% of the true value for nearly 85% of data. Exhibit K. Distribution of errors in reported data for outpatient care and maternal care indicators during 2015.
References 1. World Bank. Madagascar Emergency Support to Critical Education, Health and Nutrition Services Project [Internet]. 2015 [cited 23 Nov 2015]. Available: http://www.worldbank.org/projects/p131945/madagascar-emergency-support-criticaleducation-health-nutrition-services-project?lang=en 2. Bivand R, Lewin-Koh N. maptools: Tools for reading and handling spatial objects. R package version 0.8-30. [Internet]. 2014. Available: http://cran.r-project.org/package=maptools 3. Bivand R. spdep: Spatial dependence: weighting schemes, statistics and models. R package version 0.5-77. [Internet]. 2014. Available: http://cran.r-project.org/package=spdep 4. Stolwijk a M, Straatman H, Zielhuis G a. Studying seasonality by using sine and cosine functions in regression analysis. J Epidemiol Community Health. 1999;53: 235 8. Available: http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=1756865&tool=pmcentrez&rendert ype=abstract 5. Zuur AF, Ieno EN, Walker NJ, Saveliev AA, Smith GM. Mixed Effects Models and Extensions in ecology with R. New York: Springer Science; 2009. 6. R Development Core Team. R: A language and environment for statistical computing [Internet]. Vienna, Austria: R Foundation for Statistical Computing; 2011. Available: http://www.rproject.org/ 7. Bates D, Maechler M, Bolker B, Walker S. lme4: Linear mixed-effects models using Eigen and S4. R package [Internet]. 2013. Available: http://cran.r-project.org/package=lme4 8. Wagenaar BH, Gimbel S, Hoek R, Pfeiffer J, Michel C, Manuel JL, et al. Effects of a health information system data quality intervention on concordance in Mozambique: time-series analyses from 2009-2012. Popul Health Metr. 2015;13: 9. doi:10.1186/s12963-015-0043-3