Cost-effectiveness and economies of scale of a mass radio campaign to promote household life-saving practices in Burkina Faso

Introduction Child health promotion through mass media has not been rigorously evaluated for cost-effectiveness in low-income and middle-income countries. We assessed the cost-effectiveness of a mass radio campaign on health-seeking behaviours for child survival within a trial in Burkina Faso and at national scale. Methods We collected provider cost data prospectively alongside a 35-month cluster randomised trial in rural Burkina Faso in 2012-2015. Out-of-pocket costs of care-seeking were estimated through a household survey. We modelled intervention effects on child survival based on increased care-seeking and estimated the intervention’s incremental cost-effectiveness ratio (ICER) in terms of the cost per disability-adjusted life year (DALY) averted versus current practice. Model uncertainty was gauged using one-way and probabilistic sensitivity analyses. We projected the ICER of national-scale implementation in five sub-Saharan countries with differing media structures. All costs are in 2015 USD. results The provider cost of the campaign was $7 749 128 ($9 146 101 including household costs). The campaign broadcast radio spots 74 480 times and 4610 2-hour shows through seven local radio stations, reaching approximately 2.4 million people including 620 000 direct beneficiaries (pregnant women and children under five). It resulted in an average estimated 24% increase in care-seeking for children under five and a 7% reduction in child mortality per year. The ICER was estimated at $94 ($111 including household costs (95% CI −38 to 320)). The projected provider cost per DALY averted of a national level campaign in Burkina Faso, Burundi, Malawi, Mozambique and Niger in 2018-2020, varied between $7 in Malawi to $27 in Burundi. Conclusion This study suggests that mass-media campaigns can be very cost-effective in improving child survival in areas with high media penetration and can potentially benefit from considerable economies of scale. Trial registration number NCT01517230; Results.

In 2012, 1 in 10 children born in Burkina Faso died before the age of five. Malaria, preterm birth complications, pneumonia and diarrhoea were the main causes of mortality.12 Institutional delivery rates were 65% and despite reductions in maternal mortality, the maternal mortality ratio remained high (400 deaths per 100 000 live births).13 Seventy per cent of the population resided in rural areas with predominantly public healthcare provision.14 15 In 2002, user fees were removed for antenatal care in public facilities, and since 2006 childbirth and emergency obstetric care have been subsidised by the government.16 Child health services remained subject to user fees until mid-2016.17 Burkina Faso has a media environment dominated by local FM radio stations with a broadcasting range of 50–100 km and limited penetration of national media, which facilitated a cluster randomised controlled trial.9 A mass radio campaign addressing the main causes of postneonatal child mortality was broadcast over a 35-month period between 2012 and 2014 by seven community radio stations in the country with high radio listenership.10 18 The campaign covered a population of approximately 2.4 million and comprised around 480 000 children under 5 years.14 The campaign consisted of 1-min radio spots, with a new spot each week broadcast 10 times every day, together with 2-hour long radio shows broadcast every weekday evening. Campaign messages addressed illness symptoms in children and the importance of taking the child to a health facility (or using oral rehydration salts for diarrhoea); nutrition during pregnancy and for neonates and infants; hygiene practices and antenatal care and institutional delivery (online supplementary appendix, p.1). Spots were translated into the six local languages spoken in the intervention areas. The spots and the radio show programme scripts were developed through formative research and piloting among people in rural villages and spots were recorded by a team of 17 radio professionals and actors based in Ouagadougou.19 The campaign was implemented by the Burkina Faso country office of the non-governmental organisation (NGO) Development Media International (DMI), with support from its international head office in London. Radio stations received mentoring and training, some equipment (laptops and software, solar panels) as well as a monthly cash payment to cover production costs of long format shows (payment to local actors, etc) equivalent to $1425 per station. In exchange, airtime (radio broadcasting time) was offered at no cost by the radio stations during the campaign. The intervention and theory of change are described in more detail elsewhere.15 bmjgh-2018-000809supp001.pdf We estimated the incremental financial and economic costs of the mass radio campaign compared with current practice (existing media activities) from a provider and a societal perspective. Costs incurred by the campaign implementers (provider costs) were measured during the start-up period (December 2010–February 2012) and throughout the campaign (March 2012–January 2015). Financial costs include the value of all financial transactions incurred as a result of the campaign. Economic costs value all resources at their opportunity cost, including donated resources. Research costs related to the evaluation were excluded as they did not contribute to the campaign impact. The additional costs to households from changes in care practices and care-seeking due to the intervention were quantified and added to provider (campaign) costs to measure societal costs. Facility-level costs relating to increased service uptake were not included, as primary care facilities were often underused, and drugs were paid for by patients.20–22 Although following the change to free child and maternal services in mid-2016 the costs of drugs would fall on the healthcare system, these costs are not included in the scale-up scenarios for 2018–2020 (see below) which are limited to a provider (campaign implementation) perspective. Provider costs were primarily estimated from financial accounts data. The quantities and value of resources not directly paid for were documented and valued at market prices. The value of airtime for spots and long format shows was estimated by radio stations. Household expenditure (including transport) for antenatal care, childbirth and care-seeking for children under five was estimated through a baseline survey of 5043 mothers of children aged under five in 2012.18 Provider costs were categorised by resource input: staff, travel, supplies, rent/utilities, equipment and vehicles and by activities. Start-up activities included the setup of operations, recruitment and training of staff, general administration, project coordination and contracting with radio stations. Recurring activities included general administration, project coordination, the development of radio spots and long format shows, regular support and mentoring of radio station staff and formative research (online supplementary appendix, p. 1). Time sheets were used to allocate staff time to activities and driver time sheets used to allocate vehicle costs. Capital items and start-up costs were annualised over the lifetime of the campaign.23 For economic costs, the 1 min spots were valued at market prices for bulk purchasing of advertising airtime. The long format shows were seen as providing a net contribution to radio stations due to the investment made by DMI in improving stations’ programming capacity, developing staff skills and were not costed in the base case analysis, but these costs were included in the sensitivity analysis. Additional care-seeking costs to households due to the campaign were calculated from the mean costs of care-seeking for each service type reported in the baseline household survey. The care-seeking costs were then multiplied by the additional number of visits attributable to the intervention. We present annual and total intervention costs over the 35 months of implementation. Provider costs were analysed in Excel and household costs were analysed in Stata. The cost per radio spot and long format show were estimated based on the share of staff time working on each, applying the same proportional allocation to divide the other activity costs between the two. All costs are presented in constant 2015 USD (exchange rates: 501 XOF/USD, 0.63 GBP/USD)24 (0.71 GBP/USD for the prospective 2018–2020 analyses) and were discounted at a 3% rate. The effects of the mass radio campaign were measured on all-cause postneonatal under-five mortality (primary outcome) and under-five mortality (secondary outcome) in a cluster-randomised controlled trial.10 Fourteen community radio stations were selected for the evaluation. Clusters around each radio station were identified using the last national census with a population of about 40 000 inhabitants per cluster. We included villages located around the selected community radio station, with a good radio signal but more than 5 km away from town centres (thus less likely to be on the electricity grid, limiting access to television and making radio listening more likely). Seven clusters were randomly allocated to receive the intervention using pair-matched randomisation based on geography and radio listenership, as outlined further elsewhere.10 The trial was designed to detect a 20% reduction in the primary outcome (all-cause, postneonatal, under-five child mortality) with a power of 80% and had a 54% power to detect a 15% reduction at baseline. However, rapidly declining mortality in both arms over the study period (from 93.3 to 58.5 postneonatal under-five deaths per 1000 live births in the control group and from 125.1 to 85.1 in the intervention arm) further reduced the power of the study. The mortality reduction estimated from the endline survey showed no significant effect on mortality, but the CI was wide (risk ratio 1.00, 95% CI 0.82 to 1.22)7 and the lack of an effect on mortality appeared inconsistent with substantial increases in care-seeking observed in the intervention arm. This study therefore uses a modelling approach using the well validated Lives Saved Tool (LiST)25 to estimate the impact of the increase in health service utilisation observed during the trial period on the number of under-five lives saved and child mortality that may have been undetected by survey data. Routine health facility data from across trial clusters were analysed using interrupted time-series analyses with mixed effects Poisson regression of monthly counts of attendances per cluster, to assess the intervention effect by time period on under-five consultations, for separate diagnosis categories over the period January 2011 to December 201410. Findings showed that under-five consultations increased by 35% in year 1 (p<0.001), 20% in year 2 (p=0.003) and 16% in year 3 (p=0.049) in the intervention arm relative to the increase in the control arm; antenatal care consultations increased by 6% in year 1 (p=0.004), 9% in year 2 (p=0.026) and 8% in year 3 (p=0.129) relative to the increase in the control arm and facility-based deliveries increased by 7% in year 1 (p=0.004), 6% in year 2 (p=0.003) and 9% in year 3 (p<0.001).10 Further analysis of under-five consultations by diagnosis showed that consultations for malaria, pneumonia and diarrhoea, the three main diseases targeted by the media campaign (and the leading causes of postneonatal child mortality in Burkina Faso), also increased substantially in the intervention arm relative to the control arm.11 Consultations for malaria symptoms increased by 56% in the first year (p<0.001) of the campaign, 37% in the second year (p=0.003) and 35% in the third year (p=0.006); consultations for lower respiratory infections increased by 39% in the first year of the campaign (p<0.001), 25% in the second year (p=0.010) and 11% in the third year (p=0.525) and consultations for diarrhoea increased by 73% in the first year (p<0.001), 60% in the second year (p=0.010) and 107% in the third year (p<0.001). Consultations for other diagnoses which were not targeted by the radio campaign did not differ between intervention and control arms.11 As described in full elsewhere,11 we modelled lives saved due to augmented levels of care-seeking for maternal care and illness in children under five during the 3-year duration of the campaign using LiST.25 Briefly, increases in consultations for malaria, diarrhoea, pneumonia, antenatal care and facility deliveries were estimated using data from the trial clusters, as described above. For malaria, diarrhoea and pneumonia, we used the 2010 Demographic and Health Survey (DHS) data for rural populations in Burkina Faso to estimate the proportion of children taken to a health facility with symptoms of these conditions who received treatment. We then used these two sets of figures to estimate the increase in the proportion of children receiving appropriate treatment as result of the intervention. Baseline coverage levels (eg, the proportion of children taken to a health facility for symptoms of diarrhoea) were also estimated from DHS 2010. We took account of declining rates of under-five mortality by inputting mortality rates for each of the years of interest estimated from the trial endline survey. We used LiST to generate two sets of projections, one projecting mortality with the increased levels of consultations observed in the trial, the other with no changes other than in the underlying mortality rate. From these, we estimated that increased consultations should have resulted in 2967 under-five lives saved, representing an overall 7.1% reduction in child mortality (9.7% in year 1, 5.7% in year 2 and 5.5% in year 3).11 We estimated discounted disability-adjusted life years (DALYs) as discounted life years saved. Life years lived with disability were not included due to the lack of information on prevalence of long term serious sequelae from childhood illness or intervention impact on the duration of an illness episode in children. Life years saved per child were estimated based on the average life expectancy in Burkina Faso at age 2.5, 60 years, and discounted at 3% to 28 years26. This was multiplied by the number of lives saved to estimate total discounted life years saved. In order to determine whether interventions are cost-effective, incremental cost-effectiveness ratios (ICERs) are estimated as the ratio of the difference in cost between the intervention and an alternative and the difference in effects between the intervention and an alternative. We estimated the ICER as the incremental cost per life saved and per DALY averted in children under 5 years of age relative to current practice from a provider (campaign) and societal perspective. This was done by combining the information on costs generated as outlined above, with information on programme effectiveness generated through the LiST model. A series of one-way sensitivity analyses were carried out to examine the effects of model and parameter uncertainty on the societal ICER, including variations in care-seeking behaviours and health outcomes, cost parameters and the discount rate (online supplementary appendix, p. 2). To assess the joint effect of uncertainty in model parameters, we ran a probabilistic sensitivity analysis (PSA) using 10 000 iterations, randomly sampling input parameters. Normal distributions were used for care-seeking behaviour and the lives saved estimates, and gamma distributions, bounded by 0 with a positive skew, were used for household costs.27 A mean point estimate was calculated by dividing mean costs by mean effects. The 95% of the ICER was based on a bootstrap of 1000 iterations of the PSA, also run with 1000 iterations.28 Its distribution is also presented graphically in terms of the 2.5th and 97.5th percentiles. We considered the ICERs in relation to three willingness to pay threshold values: those proposed in a 1993 World Bank report, updated to 2015 USD values as $41 (highly attractive intervention) and $248 (attractive intervention)29 and the WHO-CHOICE proposed country-specific threshold for a ‘highly cost-effective’ intervention of the national per capita purchasing power adjusted gross domestic product (GDP), $1700 in Burkina Faso in 201615. We plotted the results as a cost-effectiveness acceptability curve. We conducted a number of scenario analyses, to project how costs would change according to the scale of implementation and with implementation in other contexts. The campaign was scaled up nationally in May 2015, covering an estimated 45% of the female population regularly listening to radio (versus an estimated 52% in the intervention area).11 As described in full elsewhere,11 the LiST model was used to estimate the impact of a nationwide campaign, using national level DHS data and assuming that a nationwide campaign would be 13% less effective as a result of lower radio listenership (ie, reflecting the percentage reduction in listenership from 52% to 45%). We projected the incremental cost-effectiveness of the campaign at national scale in Burkina Faso during the time period of the trial 2012–2014, based on the actual costs incurred during the national scale-up in 2015, to explore what the ICER might have been had the intervention been implemented at national scale from the start. For household costs, our estimates of the mean cost per visit were applied to the projected number of additional visits in the national population.14 We also modelled the cost-effectiveness of ongoing implementation at scale during the period 2018–2020 in Burkina Faso and implementation at scale in four other African countries with diverse healthcare coverage, population size and media structure. The prospective scale-up scenarios included provider (campaign) costs only. Provider costs of scale-up in other African countries were estimated by adapting the incurred expenses in Burkina Faso to country-specific conditions in terms of salary levels, characteristics of the broadcasting market, production costs, costs of airtime and travel costs. We included the start-up costs of establishing a national campaign production office, as in Burkina Faso, with the number of full-time staff ranging from 14 to 35 per country. National offices were to be responsible for the development of radio spots, including formative research to inform content and monitoring of uptake, payments and support to radio stations. In countries where TV penetration was 20% or higher, we also included the costs of producing TV spots (online supplementary appendix, p. 3). Scale-up scenarios assumed less strategic and managerial input from the international NGO in London, with the London office costs constituting 9% of overall costs compared with 38% during the trial; valued spots based on airtime costs and excluded long format shows, as there was less evidence of impact on behaviours. Indeed, dose-response analysis at midline suggested a stronger correlation between behaviour change and spots exposure relative to longer format programmes exposure (regression coefficient, 0.9 vs 0.1).18 We assumed a 10% reduction in impact due to the exclusion of long-format shows from the campaign. LiST was used to project mortality effects for 2018–2020 across the national populations of each country. The latest DHS data for each country were used to estimate the proportion of children with childhood illnesses (malaria, diarrhoea and pneumonia) who were taken to a health facility and received appropriate treatments and the increased care-seeking effects observed in the Burkina Faso trial, for each separate diagnosis category, were then applied accordingly. Coverage of antenatal care and facility deliveries were also estimated for each country, using the same approach as for the Burkina Faso projections. LiST was then used to project mortality effects for 3 year media campaigns in each country, from 2018 to 2020. Predicted campaign population exposure (penetration of used media channels) was based on national DHS estimates of radio and TV penetration in each country. We made the assumption that the number of people impacted was directly proportional to the number exposed. We therefore adjusted the mortality outcomes generated by the LiST modelling using the figure for female radio listening in Burkina Faso as a linear index.10 11

N/A

• Dima Health