Menu
Menu
Menu
Menu
Background Community based breastfeeding promotion programmes have been shown to be effective in increasing breastfeeding prevalence. However, there is limited data on the cost-effectiveness of these programmes in sub-Saharan Africa. This paper evaluates the cost-effectiveness of a breastfeeding promotion intervention targetingmothers and their 0 to 6 month old children. Methods Data were obtained from a community randomized trial conducted in Uganda between 2006- 2008, and supplemented with evidence from several studies in sub-Saharan Africa. In the trial, peer counselling was offered to women in intervention clusters. In the control and intervention clusters, women could access standard health facility breastfeeding promotion services (HFP). Thus, twomethods of breastfeeding promotion were compared: community based peer counselling (in addition to HFP) and standard HFP alone. A Markov model was used to calculate incremental cost-effectiveness ratios between the two strategies. The model estimated changes in breastfeeding prevalence and disability adjusted life years. Costs were estimated from a provider perspective. Uncertainty around the results was characterized using one-way sensitivity analyses and a probabilistic sensitivity analysis. Findings Peer counselling more than doubled the breastfeeding prevalence as reported by mothers, but there was no observable impact on diarrhoea prevalence. Estimated incremental cost-effectiveness ratios were US$68 per month of exclusive or predominant breastfeeding and U$11,353 per disability adjusted life year (DALY) averted. The findings were robust to parameter variations in the sensitivity analyses Conclusions Our strategy to promote community based peer counselling is unlikely to be cost-effective in reducing diarrhoea prevalence and mortality in Uganda, because its cost per DALY averted far exceeds the commonly assumed willingness-to-pay threshold of three times Uganda’s GDP per capita (US$1653). However, since the intervention significantly increases prevalence of exclusive or predominant breastfeeding, it could be adopted in Uganda if benefits other than reducing the occurrence of diarrhoea are believed to be important.
An economic model based on the PROMISE-EBF trial was created in order to estimate the cost-effectiveness of using peer counselling to promote exclusive breastfeeding. Because some of the data needed to populate the model, such as hospitalisation rates and disease costs, were not captured during the trial, this was supplemented by evidence from the literature. The CHEERS checklist was used to guide reporting (see the checklist in S1 Checklist). The economic evaluation compares two breastfeeding promotion strategies, based on the control and intervention groups in the trial: 1) community-based peer counselling conducted alongside breastfeeding promotion in facility-based maternal and child health services, including antenatal and postnatal services (herein after referred to as peer counselling); and 2) breastfeeding promotion in facility-based maternal and child health services only (herein after referred to as health facility breastfeeding promotion—HFP). This followed the set-up in the PROMISE-EBF trial, where peer counselling was offered to pregnant women and mothers of newborns, while in both control and intervention clusters, women could access standard health facility services. Facility-based breastfeeding promotion is often done at antenatal and postnatal services. Many facility-based maternal and child health packages provide health advice and other services for improved maternal and child health. Women are encouraged to attend antenatal clinics at least once during pregnancy and other under-five services after birth. Over 90% of pregnant women in Uganda access antenatal care (ANC) at least once before or after delivery, and over 50% attend postnatal services [7]. PROMISE-EBF did not assess the quantity or quality of facility-based services, but parallel research within our study group has described current practices in the study setting [26]. A decision tree with a state Markov model was used to depict breastfeeding promotion and the associated feeding patterns in the first six months of life (Fig 2). The model predicts feeding patterns and children can have one of two options: exclusive or predominant breastfeeding (EBF/PBF) and mixed or replacement feeding (MF/RF), see Box 1 for definitions. The merging of EBF and PBF was done because few studies indicate clear differences in health outcomes between EBF and PBF [11, 27]. MF and RF were merged because only 0.3% of the children in the PROMISE-EBF trial received replacement feeding. In the model, children were allowed to move both ways between feeding states. This was done to reflect the findings of the PROMISE-EBF trial, which showed that mothers tended to periodically switch between EBF/PBF and MF [28]. This might be the case, since the definition of EBF/PBF was based on shorter periods (24-hours and 1-week recall), which may allow for more frequent fluctuations and may under-represent usual practice [29]. Exclusive breastfeeding is when only breast milk is given to the child, except for medicines, vitamins or mineral supplements. Predominant breastfeeding is when breast milk is nutritionally dominant while given in addition to water-based fluids including fruit juices, tea without milk or oral rehydration salts. Complementary feeding including breast milk (partial breastfeeding or mixed feeding): These terms are used to describe when non-human milk, semi-solids or other solids are given to the child in addition to breast milk. The term mixed feeding does normally refer to the feeding practice specified above during the first half of infancy (under 6 months old). Replacement feeding is defined as the feeding strategy when breastfeeding has been stopped, or if the child never has been given any breast milk. Exclusive replacement feeding was defined as when never having given any breast milk. At each feeding state a decision tree model depicts three health scenarios: sick (diarrhoea), well (no diarrhoea) and dead. The diarrhoea state could either be severe, in which case the child would be hospitalised, moderate–requiring outpatient treatment, or mild–requiring no treatment. Probabilities of diarrhoea and mortality were assumed to depend on the feeding practice, i.e. whether a child belonged to the EBF/PBF or MF/RF pattern. The model had monthly cycles, terminating at six months, based on the PROMISE-EBF trial. It was based on the hypothesis that additional breastfeeding promotion increases EBF/PBF prevalence, which in turn could affect morbidity and mortality. In the PROMISE-EBF trial, mothers were followed up for six months, and a questionnaire was administered at 3, 6, 12 and 24 weeks postpartum. These data were used to estimate feeding and diarrhoea state transition probabilities. Feeding state transition probabilities were estimated using parametric survival models [30] of EBF/PBF duration (Table 1) [28]. A multiple events model was used to estimate survival. The event was cessation of EBF/PBF, a time dependent variable indicating the age at which EBF/PBF was stopped. Cases were censored if respondents were lost to follow-up, missed the last valid interview, or continued EBF/PBF beyond six months. Source: Chola et al. 2013. Infant feeding transitions among Ugandan children from the cluster-randomised trial PROMISE-EBF. α 1 = Probability of remaining in EBF/PBF; α 2 = Probability of transitioning from EBF/PBF to MF/RF; α 3 = Probability of remaining in MF/RF; α 4 = Probability of transitioning from MF/RF to EBF/PBF. Probabilities of diarrhoea events (sick or well) were estimated using longitudinal/panel methods in STATA. The aim was to assess diarrhoea morbidity in infants during the six month follow-up. We estimated four transition probabilities between two states (sick and well) as shown in Table 2. The events were dependent on a child’s feeding state. β 1 = Probability to be sick with diarrhoea in respective visit and continued diarrhoea in next visit; β 2 = Probability of diarrhoea in respective visit and changed to well (no diarrhoea) in next visit; β 3 = Probability of being well in respective visit, and continue in this state in next visit; β 4 = Probability of well in respective visit and sick in next visit. The PROMISE-EBF trial did not collect data on diarrhoea hospitalisation. We based our assumptions on diarrhoea hospitalisation on a systematic review of breastfeeding and the risk of diarrhoea and diarrheal death [31]. This study reviewed literature published from 1980 to 2009 on suboptimal breastfeeding and the associated diarrhoea morbidity and mortality. Random effects meta-analyses were conducted on 18 studies, to generate pooled relative risks of diarrhoea hospitalisation and death. The estimated relative risk (95% confidence interval) of diarrhoea hospitalisation was 2.3 (0.08–6.55) for predominantly, 4.4 (1.75–13.84) for partially and 19.5 (6.13–33.86) for not breastfed infants 0–5 months of age, compared to those exclusively breastfed (Table 3). CI = Confidence interval; RR = Relative risk; SA = Sensitivity analysis. Costs were assessed from a provider perspective (Table 3), mainly because the cost analysis done on the PROMISE-EBF trial was retrospective and only costs from project accounts could be obtained [15]. Costs of peer counselling were estimated in the PROMISE-EBF trial. The structure of the trial could be broken in four basic functions: administration; peer supervision and coordination; peer support; and recruitment (done by recruiters, who recruited mothers into the trial. These were not peer supporters). We thus used an ingredients approach to costing, in order to capture the different elements of the trial. Costs included capital items such as motor vehicles, furniture and computers; and recurrent items such as salaries, fuel and rentals. Staff costs were based on the local rates paid to employees in the trial. Peer supporters and recruiters were not permanent members of staff. They were offered a token US$20 every month for their participation, a figure that was arrived at in a meeting with peer supporters, where they agreed on this as adequate compensation for their time. Intervention start-up costs, as well as costs of training and retraining of peer supporters were also included in the analysis. Detailed descriptions and analyses of the costs of the PROMISE-EBF trial in Uganda are given elsewhere [15]. Annual intervention costs were estimated to be US$56,308, and the cost per mother/infant pair (the output required in the economic model) was US$139 per year. The cost per visit was US$26 and the cost per week of EBF was estimated to be US$15 at 12 weeks post-partum. Costs of HFP were not collected in the PROMISE-EBF trial, but were averaged from results of two studies assessing maternal health services. Orach et al.[32] estimated an average cost of antenatal and postnatal services ranging from US$2 to US$6 per child in three rural districts in Uganda. Levin et al. [33] estimated an average antenatal and postnatal services cost between US$2 to US$4 per child in Uganda. The average cost of US$3.6 was used in the model. Diarrhoea treatment costs were also not collected in the trial. These were estimated form ad hoc and PubMed searches for articles in English undertaken in the last decade on treatment costs of diarrhoeal disease among infants in sub-Saharan Africa. The search words were costs, cost analysis and diarrhoea. The criteria for selection were that studies presented results from a provider perspective and provided costs for children less than five years. Studies were considered if costs were provided for outpatient, inpatient care or both. The search yielded 321 articles, whose abstracts were individually scrutinised. Fifteen articles were then selected for full text assessment, out of which 3 were finally included from Kenya [34], Ghana [35] and Zambia [36] (Table 3). All costs were converted to 2007 prices using the local (specific country) consumer price index, and then converted to US$ at the estimated exchange rate in 2007. Two measures of effectiveness were calculated as follows: Parameter uncertainty was assessed in one way sensitivity analyses. Life expectancy was varied from 38 years (males in Sierra Leone) to 77 years (females in Libya) [39]. Discount rates for both costs and outcomes were varied from 0% to 6% as recommended in the literature [40]. Disease costs were varied as follows: US$3.8 to US$26 for non-hospitalised and US$65 to US$134 for hospitalised diarrhoea cases, based on the maximum and minimum ranges of costs found in literature [34–36]. Diarrhoea hospitalisation rates were varied according to the upper and lower limits of the confidence intervals of the respective relative risk [31]. Costs of peer counselling were varied from US$74 to US$233. The low value was obtained from a scenario analysis performed in the costing study [15] and the high value from a similar study estimating the costs of peer counselling in Zambia (data available on request). A range of US$2.2 to US$6.4 was adopted for costs of HFP, representing the lowest and highest costs identified in the literature [32, 33]. Probabilistic sensitivity analyses were also undertaken. Beta distributions were fitted for probability parameters using point estimates and standard errors. Lognormal distributions were fitted for relative risks using point estimates and their confidence intervals. The gamma distribution was used for costs and DALYs, with mean costs also representing the standard error [30]. We used Microsoft Excel to summarise cost data, STATA version 11 for statistical analyses and TreeAge Pro 2012 Suite to model cost effectiveness.
