The cost-effectiveness of a program to reduce intrapartum and neonatal mortality in a referral hospital in Ghana

Objective To evaluate the cost-effectiveness of a program intended to reduce intrapartum and neonatal mortality in Accra, Ghana. Design Quasi-experimental, time-sequence intervention, retrospective cost-effectiveness analysis. Methods A program integrating leadership development, clinical skills and quality improvement training was piloted at the Greater Accra Regional Hospital from 2013 to 2016. The number of intrapartum and neonatal deaths prevented were estimated using the hospital’s 2012 stillbirth and neonatal mortality rates as a steady-state assumption. The cost-effectiveness of the intervention was calculated as cost per disability-adjusted life year (DALY) averted. In order to test the assumptions included in this analysis, it was subjected to probabilistic and one-way sensitivity analyses. Main outcome measures Incremental cost-effectiveness ratio (ICER), which measures the cost per disability-adjusted life-year averted by the intervention compared to status quo. Results From 2012 to 2016, there were 45,495 births at the Greater Accra Regional Hospital, of whom 5,734 were admitted to the newborn intensive care unit. The budget for the systems strengthening program was US $1,716,976. Based on program estimates, 307 (±82) neonatal deaths and 84 (±35) stillbirths were prevented, amounting to 12,342 DALYs averted. The systems strengthening intervention was found to be highly cost effective with an ICER of US $139 (±$44), an amount significantly lower than the established threshold of cost-effectiveness of the per capita gross domestic product, which averaged US $1,649 between 2012–2016. The results were found to be sensitive to the following parameters: DALYs averted, number of neonatal deaths, and number of stillbirths. Conclusion An integrated approach to system strengthening in referral hospitals has the potential to reduce neonatal and intrapartum mortality in low resource settings and is likely to be cost-effective. Sustained change can be achieved by building organizational capacity through leadership and clinical training. Kybele and GHS maintain a long-standing relationship, beginning in 2007 with efforts to reduce maternal, fetal, and neonatal mortality through a quality improvement initiative [20, 21]. A previous analysis determined the cost-effectiveness of Kybele-GHS efforts to reduce maternal and fetal mortality between 2007 and 2011 [14]. The current initiative was undertaken at the Greater Accra Regional Hospital (GARH) as a component of MEBCI, a five-year collaboration between the Ghana Health Service (GHS), Kybele and PATH to improve newborn outcomes through government engagement and provider training across four regions of Ghana. MEBCI sought to strengthen the skills of healthcare personnel to improve newborn care through a multifaceted approach that included training in Helping Babies Breathe [6], Essential Care for Every Baby [22] and infection prevention; accessibility of resuscitation devices; and advocacy to enhance stakeholder relationships and national leadership [21]. For regional hospitals, these interventions were intended to be reinforced through a broader set of systems strengthening activities, but the scope of work was altered by the funder and the intervention analyzed in this study, and described in detail elsewhere [9], was only implemented at the GARH. Among Ghana’s regional referral hospitals, GARH has the highest volume obstetric unit with 70% of deliveries comprised of high-risk antenatal or peri-partum referrals [21]. The neonatal intensive care unit (NICU) was enlarged in 2013; however, funds were not available to significantly increase the clinical workforce during the intervention period. To support the capacity building approach at GARH, expert practitioners in obstetrics, neonatology, anesthesiology, midwifery and quality improvement from the United States and England made tri-annual visits to Ghana. During these visits, volunteer practitioners provided coaching and mentorship to GARH providers which were focused on introducing new clinical skill sets and optimizing care delivery. The principles of Kybele’s partnership model, and the robust drivers of the successful partnership, have been described elsewhere [23]. Motivated frontline healthcare workers were selected to serve as clinical champions to facilitate learning for colleagues and to monitor data. Key performance gaps in each clinical care area were identified through the analysis of processes and baseline outcome data. Training modules were developed to address content-specific gaps in each clinical area, while foundational training and coaching was provided on QI and leadership to enable GARH staff to test, adapt and implement solutions (Table 1) that included workflow redesign, institution of compassionate care and improved communication [9, 24, 25]. An electronic database (Microsoft Access, Version 15.0, Redmond, WA) and local data sources were used to collect information on neonatal outcomes and their drivers. The Institutional Review Board (IRB) approval was granted by Wake Forest University and the GHS to conduct this work, and informed consent was waived as part of the approval. Program costs were collected in real time to account for multiple sources and types of costs incurred; researchers kept a detailed budget during the intervention, and costs were reported in US $. This analysis includes costs incurred in 2012, as they were directly related to planning for the implementation of this intervention. All costs were standardized to 2019 US $. The program costs incurred by Kybele, along with external aid, accounted for 62% of the total cost, and the estimated cost of professional time accounted for 26%. The calculated program cost was US $1,590,276. Given that this analysis is from the perspective of the non-governmental organization, it does not consider any changes to the costs of delivering care. The calculated program costs have been adjusted for inflation and standardized to 2019 USD using the Consumer Price Index (CPI). The CPI for all items was used to adjust the program related costs, the physician services component of medical care category was used to adjust the costs of professional value time, and the medical care commodities component to adjust the costs of medical equipment and supplies [27]. Additionally, a majority of the funding for this project came from international sources; thus, this funding is not subject to purchasing-power parity (PPP) adjustments. All costs incurred in Ghanaian currency, including compensated time from Ghanaian providers and costs covered by GHS, were adjusted to account for PPP by using PPP exchange rates that factor price levels in different countries based on a standard basket of goods and services [28]. While the price levels for health-related services may differ from the basket used to calculate the PPP exchange rate, the fact that GHS investments covered both capital expenditures and provider time allows an adequate approximation. The inflation-adjusted total program cost is US $1,716,976 (Table 2); operational and infrastructure program costs are also presented (Table 3). a Funder costs included travel and administrative expenses, accommodations, training, and other direct costs b Participant costs included travel expenses of participants c Value time indicates the calculated value time for volunteers d GHS costs included food expenses, and the construction of a new triage pavilion e Other, third-party costs included donated medical equipment; costs associated with the renovation of the NICU (4,980.28 US $) were omitted from the present analysis, but these costs would amount to 0.3% of total costs and would not impact findings or conclusions. a Operational costs include travel expenses, administrative expenses, value time, training, incidentals, and other direct costs; it is important to note that operational costs associated with value time account for salaries from international staff, and these costs would be reduced were the program to be run by GHS b Infrastructure costs include equipment, donated items, and construction The disability-adjusted life year (DALY) is the most commonly used summary measure to quantify the burden of disease within a given population in LMIC [28–31]. The DALY metric relies on the assumption that the most appropriate measure of the effects of a chronic illness is time, both time lost due to premature death and time spent disabled by disease [31]. Therefore, DALYs are calculated by summing the adjusted number of years lived with disability (YLDs) and the number of years of life lost due to premature mortality (YLL) [32]. YLD = Number of cases x duration until remission or death x disability weight YLL = Number of deaths x life expectancy at the age of death DALY = YLD + YLL The Global Burden of Disease (GBD) project provides guidance on methods considered best practice for calculating DALYs. The major philosophical and methodological aspects of the DALY calculation have been described and debated [32–34]. The recent GBD does not discount future DALYs [35], which removes the assumption that current years of life lost are valued at a higher rate than future years of life lost. The current GBD does not apply age weighting, or the concept that the value of years lost varies with age. Historically, WHO has used both age weighting and discounting future DALYs when calculating YLL [36, 37], and this analysis takes both methodologies into account. Researchers have described methods associated with discounting DALYs elsewhere [38]. This study uses standard values for age weighting and discounting [39]. Although typical DALY calculations rely on years of life lost to both death and disability, it is not common for cost-effectiveness analyses of neonatal health interventions in LMICs to include YLD estimates [38–40]. Due to the challenges associated with accurately estimating the long-term impact of disabilities, the current study relies on YLL to estimate DALYs [39]. The discount rate (r) was set at 3% according to the WHO standard for economic evaluation of health interventions in LMICs [37]. The YLL due to premature death were calculated using the average of Ghana-specific life expectancy at birth for male and female, as local life expectancy is recommended as a good approximation of life expectancy (L) [38]. Early neonatal deaths account for 76% of neonatal deaths globally and were assumed to be the age at the event for the calculation of YLL (a) [41]. a = age at death, in years r = discount rate β = age-weighting constant K = age-weighting modulation factor C = constant from age-weighting function L = standard life expectancy at birth, in years Additionally, researchers are engaged in discussion around the inclusion of stillbirths in DALY calculations [42, 43]. The current study does not attempt to assign a value to life lost in utero prior to the onset of labor; the authors are conscious that the loss of a fetus places a great cost on families. In the 2013 Global Health Estimates, the WHO published recommendations around the inclusion of stillbirths as years of life lost and based the value on life expectancy at birth [44]. The present study included stillbirths in the base and sensitivity analysis, but considered only fresh stillbirths due to intrapartum complications, that is fetuses that arrived at the hospital with a heart beat but were born dead. To estimate the number of deaths and DALYs averted due to the partnership, this study compares the number of neonatal deaths avoided to a “no-intervention” counterfactual. This counterfactual was not observed, but rather estimated as a steady-state scenario that would have occurred had the intervention not been implemented. The quasi-experimental pre- and post-program evaluation, which was used to inform the estimation of the counterfactual, relied on data collected annually from non-random sites. In this method, a baseline NMR has been used to predict the number of neonatal deaths that would have occurred if the intervention had not taken place. An electronic project database was developed to collect project outcome indicators; primary data sources for outcome indicators included the Delivery Register in the Maternity Ward and the Newborn Admission and Discharge Register in the NICU, which are routinely collected following a patient encounter. The NMR from GARH in 2012 has been used as baseline NMR, with training starting January 2013. Thus, to calculate the estimated number of neonatal deaths under the steady-state assumption, the number of reported babies delivered at GARH in a given year was multiplied by the hospital’s 2012 NMR of 3.11%. Compared to this estimated baseline, the authors determined that any reduction in neonatal deaths would be seen as an improvement. However, this approach of estimating neonatal deaths averted through a steady state assumption is likely to be an over-estimate of the number of neonatal deaths averted by the intervention. Additionally, it is difficult to attribute causality to the intervention, as NMR may be impacted by existing demand- and supply-side factors. Similarly, the 2012 stillbirth rate (SBR) was used to make steady-state assumptions in order to estimate the number of stillbirth deaths averted in subsequent years. Following calculation of the deaths prevented, the DALYs averted were calculated using the same assumptions discussed above. The incremental cost-effectiveness ratio (ICER) is a metric used to determine the cost-effectiveness of a program or interventions. For this study, an ICER shows the program cost-effectiveness as measured in estimated DALYs averted by the program compared with a null hypothesis of no change. Estimates of costs, health effects, and ICERs provide clear guidance to policymakers only when an explicit threshold standard or threshold has been specified among other conditions [43]. In the absence of an explicit standard or threshold by policymakers, it would be difficult to make an objective recommendation. The ceiling ratio (λ), or decision rule, is an important component of cost-effectiveness analysis (CEA) and represents the decision makers’ valuation of a unit of health gain or the relative value against which the acceptability of ICERs is judged [45]. While explicit cost-effectiveness thresholds are available for US and UK, the selection of λ for interventions affecting LMICs has been left to the discretion of the analyst [45, 46]. Within LMIC settings, researchers most commonly use a cost-effectiveness threshold based on per capita gross domestic product (GDP). This approach has been promoted by the WHO-CHOICE project to define cost-effectiveness of an intervention [47, 48]. If the cost of averting one DALY is less than three times the national annual GDP per capita then an intervention is deemed cost-effective, and if it is less than once the country-specific GDP per capita it is considered highly cost-effective [41, 49]. Cost-effectiveness research throughout sub-Saharan Africa has widely utilized a threshold determined by GDP [50–53]. The league table approach, derived from the work of World Bank, recommends US $150 per DALY as ‘attractive’ cost-effectiveness, US $25 per DALY as ‘highly attractive’ for low-income countries and US $500 and US $100 per DALY, respectively, for middle-income countries [45, 46]. Each of the approaches has advantages and disadvantages; for the purpose of this study, the authors present results using multiple approaches to determine the cost-effectiveness of the systems strengthening intervention. To test assumptions made in the analysis, the authors subjected the data to a probabilistic sensitivity analysis using Monte Carlo simulations, run in Crystal Ball (Oracle, Redwood Shores, CA) as an add-in program to Microsoft Excel (Microsoft, Redmond, WA). All assumptions were varied simultaneously according to pre-specified distributions. The distributions were assigned according to the inherent characteristics of each parameter and according to accepted conventions and based on a similar CEA conducted for maternal mortality [14]. In order to calculate DALYs using discounting, the following parameters were applied (Table 4): the assumptions for average age and life expectancy were distributed uniformly around high and low estimates; the value of professional time was varied at 25%; and the number of neonatal deaths and stillbirths were varied around a normal distribution. Using the Monte Carlo simulation, the researchers modeled uncertainty in the program estimates. SE- Standard error for sample; NMR- Neonatal Mortality Rate