Menu
Menu
Menu
Menu
Introduction Results-based financing (RBF) is being promoted to increase coverage and quality of maternal and perinatal healthcare in sub-Saharan Africa (SSA) countries. Evidence on the cost-effectiveness of RBF is limited. We assessed the cost-effectiveness within the context of an RBF intervention, including performance-based financing and conditional cash transfers, in rural Malawi. Methods We used a decision tree model to estimate expected costs and effects of RBF compared with status quo care during single pregnancy episodes. RBF effects on maternal case fatality rates were modelled based on data from a maternal and perinatal programme evaluation in Zambia and Uganda. We obtained complementary epidemiological information from the published literature. Service utilisation rates for normal and complicated deliveries and associated costs of care were based on the RBF intervention in Malawi. Costs were estimated from a societal perspective. We estimated incremental cost-effectiveness ratios per disability adjusted life year (DALY) averted, death averted and life-year gained (LYG) and conducted sensitivity analyses to how robust results were to variations in key model parameters. Results Relative to status quo, RBF implied incremental costs of US$1122, US$26 220 and US$987 per additional DALY averted, death averted and LYG, respectively. The share of non-RBF facilities that provide quality care, life expectancy of mothers at time of delivery and the share of births in non-RBF facilities strongly influenced cost-effectiveness values. At a willingness to pay of US$1485 (3 times Malawi gross domestic product per capita) per DALY averted, RBF has a 77% probability of being cost-effective. Conclusions At high thresholds of wiliness-to-pay, RBF is a cost-effective intervention to improve quality of maternal and perinatal healthcare and outcomes, compared with the non-RBF based approach. More RBF cost-effectiveness analyses are needed in the SSA region to complement the few published studies and narrow the uncertainties surrounding cost-effectiveness estimates.
Malawi is 1 of 30 countries in SSA implementing RBF.26 In 2013, the Malawi Ministry of Health (MoH), with financial support from the Norwegian Ministry of Foreign Affairs and the German Federal Ministry for Economic Cooperation and Development, initiated a 5-year RBF programme designed to improve coverage and quality of MNH services. Selection of facilities was based on capacity to offer emergency obstetric care (EmOC), make referrals and provide 24/7 delivery care in four districts: Balaka, Dedza, Mchinji and Ntcheu. Performance indicators focused on delivery care,27 Box 1. Implementation started in 18 out of a total of 33 EmOC facilities in April 2013 to October 2014 and then expanded to 28 facilities until 2017.28 HMIS, health management information system; MNH, maternal neonatal health; PMTCT, prevention of mother to child transmission; RBF, results-based financing. HMIS, health management information system; MNH, maternal neonatal health; PMTCT, prevention of mother to child transmission; RBF, results-based financing. HMIS, health management information system; MNH, maternal neonatal health; PMTCT, prevention of mother to child transmission; RBF, results-based financing. MNH care in Malawi is provided at EmOC facilities and includes antenatal, delivery and postnatal care. Basic EmOC (BEmOC) facilities are expected to consistently provide a set of seven key interventions known as ‘signal functions’, while two additional signal functions are to be provided by Comprehensive EmOC (CEmOC) facilities.29 Health centres providing BEmOC are supposed to be capable of managing obstetric complications and to refer emergency cases requiring more comprehensive care to CEmOC facilities. MNH and EmOC services are provided free in public health facilities and private not for profit facilities contracted by the MoH through Service Level Agreements. The Malawi RBF Initiative, designed and implemented by the MoH, aimed to improve service quality of EmOC facilities using performance-based payments to health facilities and providers based on achievements of predefined quantity and quality targets (box 1), implying that RBF facilities had additional funding on top of centralised allocations (budgets, supplies). The RBF Initiative also used CCT to incentivise pregnant women to deliver at EmOC facilities instead of non-EmOC facilities or at home. To ensure that the providers at EmOC facilities operated within environments with the required capacity to provide quality MNH care, the RBF implementation was preceded by a one-off investment in infrastructural upgrades and equipment supplies. More details of the Malawi RBF Initiative are provided elsewhere.25 28 In contrast to RBF-supported EmOC facilities, non-RBF EmOC facilities (comparator) neither received funding beyond centrally allocated budgets nor any explicit infrastructural upgrades. We used a decision tree model to calculate the expected health effects and expected costs of the RBF4MNH Initiative from a societal perspective.30 The model simulates maternal and perinatal outcomes from 28 weeks gestation until 7 days after delivery. This period is consistent with the definition of perinatal outcomes in developing settings.31 32 Importantly, it captures the majority of maternal deaths, which occur during the third trimester and the first week after birth.33 We considered two alternatives: the RBF4MNH Initiative and status quo care (comparator). We used the model to estimate deaths averted, life-years gained (LYG) and disability adjusted life years (DALYs) averted from perinatal and maternal complications, as well as the additional (incremental) costs incurred by the RBF programme. The model was populated with information on population coverage with facility-based delivery (FD), the incidence of maternal complications, cause-specific maternal case fatality rates (CFRs), time to seek care for complications and effective coverage (parameter details are explained below). Malawi-specific estimates for life expectancy at birth and life expectancy at the mean age of women of reproductive age were used to calculate LYG34 for each perinatal and maternal death averted, respectively. Future LYG and DALYs were discounted at 3%23 in the baseline scenario, while the influence of no discounting of future health was explored in sensitivity analysis. Each alternative was simultaneously fitted with associated treatment costs, including RBF costs for the intervention arm. The costs are presented in 2013 US$. We calculated incremental cost-effectiveness ratios (ICERs)35 in terms of cost per DALY and death averted and per LYG. The ICER is the difference in costs between two interventions divided by the difference in their effects.36 An intervention is considered cost-effective if its ICER in US$ per DALY averted is less than three times gross domestic product (GDP) per capita and considered ‘very cost-effective’ if its ICER is less than the per capita GDP.37 Applied to the Malawian context, the RBF intervention would be considered cost-effective as long as it costs less than US$1485 per DALY averted and very cost-effective if it costs less than US$495 per DALY averted.38 Finally, deterministic and probabilistic sensitivity analyses were conducted to assess important drivers impacting the ICERs and the robustness of the model to variations in key parameter and model assumptions. Reflecting the options decision makers face, the decision model included two arms: RBF and non-RBF (comparator). Policy makers decided which facilities received RBF while mothers decided whether to deliver in an RBF health facility, a non-RBF health facility or at home. In the model, mothers’ decisions/service use parameters were based on primary trial data. Mothers who delivered in RBF facilities benefited from the intervention (combination of PBF+CCT) while those who delivered in non-RBF facilities only received status quo care. For each delivery, we accounted for both perinatal and maternal deaths. We defined each delivery as normal or uncomplicated (not associated with any maternal complication) or complicated (associated with any maternal complication). Maternal complications included direct causes (haemorrhage, sepsis, obstruction, eclampsia) and/or any indirect causes. Each complication could lead to a maternal death or recovery. The model allowed for the fact that mothers experiencing normal deliveries may die from incidental causes. Because some maternal complications can negatively affect perinatal outcomes, perinatal survival is linked to maternal survival.33 39 After a delivery event, the model therefore considered perinatal outcomes based on mothers’ status, that is, whether the mother was alive or dead. The outcome of the newborn (ie, stillbirth, early neonatal death, alive) was then assigned conditional on the status of their mothers. Figure 1A, B gives an overview of the decision model. The full model (available on request) was constructed using TreeAgePro 2016 software. (A) Pathways of maternal events, demonstrating maternal status after delivery. (A) is linked to perinatal outcomes for live mothers. (B) is linked to perinatal outcomes for dead mothers as shown in (B). (B) Pathways of perinatal events, demonstrating conditional relationships between perinatal outcomes and maternal status after a delivery event. RBF, results-based financing. The Malawi RBF trial25 28 was not designed to collect health outcomes data including case fatality or mortality rates. This lack of health outcomes data necessitated modelling. We therefore obtained epidemiological estimates of perinatal mortality, maternal complications and CFRs from the published literature to populate the corresponding probabilities in the model, assumed to be similar for both arms at baseline. At population level, maternal/perinatal survival depends on both coverage of pregnant women with FD services and timely access to quality obstetric emergency care.40 An impact evaluation of the RBF4MNH on service use did not demonstrate any significant differences in utilisation with facility delivery rates of 82.8% and 79.8% between RBF and comparison non-RBF facilities, respectively.41 Furthermore, obstetric care utilisation rates among women who developed complications outside health facilities were 78% and 75%, respectively.42 However, RBF was associated with significantly reduced mean time to care for women experiencing complications,42 which may translate into better survival. Importantly, the RBF impact evaluation detected significant improvements in effective coverage (ie, provision of higher quality care in respect to structural and process quality indicators, such as routine use of partographs, uterotonics for active management of third stage of labour and infection control).28 41 Similar efforts to improve the quality of obstetric care were associated with 25%–30% reductions in CFRs and 19%–20% reductions in stillbirth rates in Zambia and Uganda.43 Given similarity in MNH settings across these countries,44 we adopted the mean figures of 27.5% and 19.5% reductions in CFRs and stillbirth rates with RBF, respectively. Consistent with the Zambia and Uganda studies, we assumed that RBF had no significant effect on early neonatal mortality. Details of parameters used in the model and their sources are shown in table 1. List of parameters used in results based financing compared with non-results based financing decision tree model. *Sensitivity range based on assumption. †Precludes incentive money given to health facilities. PSA, probability sensitivity analysis. Information on incidence of maternal complications and respective CFRs is needed to calculate maternal deaths. There is wide variation in reported incidences of maternal complications even though countries by principle adhere to the same version of the International Classification of Diseases.45 A WHO multicountry survey estimated that 7.3% of FDs in developing countries are associated with maternal complications.46 This is probably the most representative estimate; thus, we used it for the baseline scenario. For home births, we assumed that the percentage of complicated deliveries was 50% higher, given that most home deliveries are not assisted by skilled birth attendants in developing countries.47 There is lack of reliable data on case-specific incidences for sepsis, haemorrhage, eclampsia and obstruction.48 49 However, data on relative frequencies of these conditions among women with complications exist. In Malawi’s rural facilities, sepsis, haemorrhage, eclampsia and obstruction account for 32%, 32%, 20% and 11% of obstetric complications, respectively, while other direct/indirect causes account for the remaining 5%50 (table 1). We thus estimated cause-specific incidences indirectly, by multiplying the relative frequency of each condition with the overall incidence of maternal complications. Overall incidence of other direct/indirect causes was estimated in a similar way. Regarding maternal CFRs, wide variations exist in data for the SSA region. A review of recent estimates (year 2000 onwards)49 reported that facility based CFRs range from 3.6% to 18.0% for sepsis;51 52 2.8% to 12.3% for haemorrhage;52 53 3.4% to 18.0% for eclampsia51 52 and 2.0% to 12.7% for obstruction.51 54 We used the mean estimates as baseline estimates of CFR and tested the whole ranges in sensitivity analyses to reflect this diversity of CFRs across SSA settings. As no corresponding data exist for women experiencing indirect complications and among home births, we adopted the mean CFR (0.09%) to identify deaths due to complications from other/indirect causes and during home births (table 1). Though rare, women can have comorbidities or experience more than one complication, raising the problem of competing mortality risks.55 The model applies cause-specific incidences concurrently, on the assumption that the risk of each maternal complication is independent from the risk of other complications. Mortality risk for non-maternal causes was approximated by subtracting annualised life time risk for maternal death (0.0008) for women of reproductive age (15–49 years)56 from annualised all-cause mortality risk (0.0048) for women aged 25–29 years.34 We calculated perinatal deaths by combining the risk of stillbirth and early neonatal mortality with information about the status (live or dead) of the mother after delivery. The international literature estimates stillbirth and early neonatal mortality rates at 28.4/1000 births and 19.3/1000 live births, respectively, among a population-based cohort of mothers that survive births in LMIC.39 For mothers that die soon after births, the estimated stillbirth and neonatal mortality rates are 318.8/1000 births and 89.9/1000 live births, respectively.39 As perinatal mortality risks following incidental maternal deaths are not linked to maternal complications, we assumed they are the same as for normal deliveries. We transformed these rates into corresponding probabilities57 (table 1). Individual studies in SSA report mixed results on risk of perinatal mortality by place of delivery; some studies find that the risk is lower for FDs58–60 while others report lower risks for non-facility based deliveries.61 62 We based our perinatal risk adjustment on a meta-analysis that pooled results from population-based cohort studies in SSA.63 The study reported an OR of 1.21 in perinatal deaths for non-facility based compared with facility-based births. This was transformed into corresponding relative risk.64 DALYs are the sum of years of life lost (YLL) due to premature mortality and the years lived with a disability (YLD),65 DALY=YLL + YLD. Maternal YLL were estimated as the number of maternal deaths multiplied by the life expectancy at 25 years using Malawi life tables.34 Similarly, perinatal YLL were estimated as the product of perinatal deaths and the life expectancy at birth. Maternal YLD were estimated by multiplying prevalence of maternal complications by their corresponding disability weights65 (table 1). Since disability weights for eclampsia and other maternal conditions are not available,66 we adopted the mean disability weight of haemorrhage, sepsis and obstruction (0.260) for eclampsia and 0.133 for other maternal conditions66 assuming that most of them would be related to infections given the high infectious disease prevalence in the SSA region.67 Perinatal YLD were not estimated due to lack of data. We collected health systems and RBF4MNH programme costs from four health centres. We randomly selected two districts, and within each district, randomly selected an intervention and a comparison health centre. From these, we collected cost data twice, in 2014 and in 2015, to cover two separate fiscal years. Given some differences in actual timing of RBF implementation between intervention health centres, the periods of data collection were not identical. We used the World Bank RBF toolkit to guide cost data collection.68 We defined costs as variable (changing with service volume) or fixed (not changing with service volume). For variable costs, we used an ingredient approach to estimate unit costs.23 At the health facilities, we identified and recorded quantities of resources used for maternal and new-born care. We used pharmacy and inventory records to quantify drugs and other supplies. Unit prices were obtained from Malawi Central Medical Stores catalogue or local retailers as appropriate.69 Information on utilities, building maintenance were either collected at the health facilities or respective district offices, depending on where complete records were available. RBF data on incentives, training, information and communication materials, equipment supplies, upgrades and supervision were collected from both RBF desk officers at the health facility and/or the main RBF office at the MoH Reproductive Health Unit. We used the consumer price index to convert prices into 2013 constant prices before discounting them at 3%.23 For fixed costs (building and equipment), information on useful life years and replacements costs was obtained from MoH Planning and Policy Directorate and from the National Health Accounts.70 We annualised and discounted the fixed costs at 3% rate and used a top-down approach to allocate joint or shared costs using allocation proxies.23 For instance, we used the proportion of maternity unit area relative to the area of all hospital units to allocate building costs to the maternity unit, while we used the share of maternity unit clients among all visits to allocate health worker salaries. An implicit assumption of the latter is that resource requirements of the maternity unit are equal to the average resource requirement of all facility activities. We used facility registers and human resource records to find information on births, staffing levels and cadres. Costs related to administrative support from the district offices were not collected, on the assumption that they would not substantially differ between health centres. We estimated RBF personnel costs, office rentals and other overhead costs from the central office by inflating all RBF costs by 38.8%. In this way, the overall RBF administrative costs account for 28% of total implementation costs, which is consistent with a pay for performance programme in Tanzania.22 Household costs, including direct and indirect costs, associated with care seeking were based on our earlier analyses.42 Local currency values were converted to US$ equivalents using the 2013 midyear exchange rate (US$1=MK 330). We first explored the impact of each model parameter on ICERs through one-way sensitivity analyses.71 We varied the mean of each parameter over appropriate reported ranges. We used normal approximation methods to estimate ranges for binomial parameters in cases where corresponding CIs were not provided.72 In the absence of empirical estimates, parameters were varied ±20% (table 1). The 10 parameters that influenced the ICERs most were further assessed through probabilistic sensitivity analyses after assignment of appropriate distributions. Gamma distributions were specified for costs, normal distribution for LYG, lognormal distributions for probabilities and Beta distributions for service use.30 Bounds for the parameters were derived using methods of moments.30 We conducted parametric bootstrapping based on 5000 iterations to propagate parameter uncertainty through the model and presented the results as cost–effectiveness scatter plot and acceptability curves. At any given value of willingness to pay, acceptability curves show the probability of an intervention being cost-effective relative to the comparator. We validated the model by comparing baseline perinatal mortality rates with estimates from the published literature (internal validity) and by inspecting that all parameters influence the model according to expectations (face validity). It was not appropriate or possible to involve patients or the public in the design, or conduct, or reporting, or dissemination plans of our research.
N/A
