Cost-effectiveness analysis of integrated community case management delivery models utilizing drug sellers and community health workers for treatment of under-five febrile cases of malaria, pneumonia, diarrhoea in rural Uganda

The study was a cost-effectiveness analysis to compare the iCCM delivery model utilizing drug sellers against the model using CHWs. A decision tree analytic model was employed in the economic analysis to construct and structure decisions.

Effectiveness data were obtained from a quasi-experimental study conducted in Southwestern Uganda and the inSCALE cross-sectional household survey that was carried out in mid-Western Uganda. Details of these studies are described elsewhere [6, 12]. In brief: the quasi-experimental study conducted in the South West evaluated the impact of the iCCM intervention on drug seller paediatric fever management and adherence to iCCM guidelines between June 2013 to September 2015. In the intervention arm, 212 care-seeker exit interviews were conducted before, and 285 after; in the comparison arm, 216 care-seeker exit interviews were conducted before, and 268 at the end of the study period [6].

For the inSCALE study, iCCM implementation among CHWs was supported in eight districts in mid-Western Uganda by the Malaria Consortium. Between May and August 2011, a cross-sectional survey was conducted in each of the eight participating districts to assess the impact of the iCCM intervention [12]. Unit costs for items were obtained from the International Drug Price Indicator Guide (2014 edition), iCCM product selection UNICEF guide (2016 edition), Bugoye iCCM project budget, the iCCM costing report for Senegal, and other costing studies [13‐15]. International Drug Price Indicator Guide and iCCM product selection UNICEF guide contain international reference prices in US dollars, for essential medicines and medical supplies from pharmaceutical suppliers, international development organizations and government agencies. Other costs were derived from studies conducted in Uganda, close to the Ugandan setting. The micro-costing (ingredients) approach was used to cost the iCCM interventions with the aid of the Community Health Planning and Costing Tool [16]. Cost and effect data were linked using a decision tree model and analysis was carried out with Amua modelling software [17].

The iCCM intervention involved training health workers in the assessment, testing, classification, and treatment of malaria, pneumonia and diarrhoea among children under the age of 5 years. Health care workers were taught how to detect malaria using RDT, how to diagnose pneumonia using respiratory timers, and how to diagnose diarrhoea based on the number of loose stools per day.

In the ICCM delivery model utilizing CHWs, the national iCCM trainer trained health workers (mainly staff from health centres II and III) and district health office staff, as trainers of trainees (TOTs) in a six-day facilitator programme. The TOTs then spent six days training CHWs chosen by their communities. Two trainers were assigned a class of between 25 and 30 CHWs.

Two CHWswere trained in the treatment of malaria, pneumonia and diarrhoea in each village. Free of charge, the trained CHWs received iCCM kits containing rapid diagnostic tests (RDTs) for malaria, respiratory timers, and supplies of artemether/lumefantrine (20 mg/120 mg tablets), amoxicillin (125 mg dispersible tablets), low osmolarity ORS, zinc (20 mg tablets), and rectal artesunate (50 mg). Additionally, the CHWs received a job aid that included the iCCM diagnosis and treatment algorithms, as well as a register for documenting the nature and frequency of all VHT-related activities. Additionally, stock cards, medicine boxes, gloves, cotton wool, and methylated spirit were provided. CHWs were rewarded with raincoats, umbrellas, gumboots, solar lights, and hoes [18].

The iCCM trained-drug seller delivery model consisted of four different components [6]:

To improve the community care-seeking practices that affect U5 child health, CHWs delivered messages on fever, care-seeking, diagnostic testing, and treatment adherence, through community meetings, workshops, radio talk shows, announcements, and word-of-mouth [18].

The effectiveness measure used was ‘U5 case of febrile illness appropriately diagnosed and treated’. The phrase refers to any confirmed malaria, diarrhoea or suspected pneumonia case among U5s that received treatment according to treatment guidelines, or any child without malaria, diarrhoea or suspected pneumonia that was not prescribed the recommended drugs to treat malaria, diarrhoea or pneumonia.

Appropriate treatment of uncomplicated malaria entails testing a child with fever or a history of fever with RDT and administering the appropriate ACT regimen if positive. An afebrile child is not tested or prescribed ACT. Children who tested positive for mRDT were given artemether/lumefantrine 20/120 mg DT as follows: 6 tablets for children aged 4–35 months (one tablet twice daily for 3 days), 12 tablets for children aged 36–59 months (two tablets twice daily for 3 days).

A child with a cough and rapid breathing (checked using a respiratory timer to be ≥ 60 breaths per minute (bpm) for a child 0–7 days, ≥ 50 bpm for a child 2–11 months, and ≥ 40 bpm for a child 1–5 years) received an appropriate amoxicillin DT regimen. Amoxicillin DT was not prescribed to a child with a cough and normal breathing. Children with a cough and rapid breathing were given amoxicillin DT 125 mg as follows: 20 tablets for children aged 2–11 months (two tablets twice daily for 5 days); 30 tablets for children aged 12–59 months (three tablets twice daily for 5 days) [6].

Appropriate treatment for non-bloody diarrhoea included administering zinc 20 mg DT and ORS sachets to a child (3 or more loose stools with no blood seen in 24 h). Zinc sulfate was administered as follows: 5 tablets for children aged 2–6 months (half tablet once daily for 10 days), 10 tablets for children aged 7–59 months (one tablet once a day for 10 days). Each child was given two sachets of ORS, and the drug seller or CHW demonstrated how to reconstitute them to the caregiver. Each child was to drink at least half a 300-ml cup of water following each loose stool [6].

Appropriate diagnosis and treatment of U5 febrile illnesses serve as a proxy indicator of the potential for child mortality associated with these illnesses. Malaria cases can rapidly progress to complications and death if treatment is not initiated within the first 24–48 h of symptom onset [19]. Prompt treatment with a complete course of effective antibiotics is critical for pneumonia mortality reduction [20]. ORS and zinc are both effective interventions for reducing diarrhoea-related mortality [21].

The effectiveness data of the drug seller-based delivery model was determined through a quasi-experimental study conducted in Southwestern Uganda. This study is detailed elsewhere [6]. In thisstudy, the iCCM intervention utilizing drug sellers increased the appropriate treatment of uncomplicated malaria, pneumonia symptoms and non-bloody diarrhoea by 80.2% (95% CI 53.2–107.2), 65.5% (95% CI 51.6–79.4), and 31.4% (95% CI 1.6–61.2), respectively.

Effectiveness data for the CHW-based delivery model were obtained from the inSCALE cross-sectional household survey conducted in eight districts of mid-western Uganda. The Malaria Consortium supported the iCCM project through a grant from the Canadian International Development Agency (CIDA). The study's details are available elsewhere [22]. According to Soremekun et al., the coverage rates of appropriate treatment increased 80.4, 51, and 21.5% for uncomplicated confirmed malaria, suspected pneumonia, and non-bloody diarrhoea, respectively.

Economic costing was done from the societal perspective using a micro-costing approach (ingredients approach). The societal perspective is broader than the health care or government perspective and allows comparison with previous studies. Costs that were considered in the societal perspective included health sector costs, costs borne by CHWs or drug sellers, and household costs. It was essential to consider household costs because they can be significant and may deter caregivers from utilizing the iCCM service and cause poverty [23].

Data on the cost of medicines and other health supplies were collected from the International Drug Price Indicator Guide (2014 edition) and iCCM product selection UNICEF guide (2016 edition). Costs for implementation of the iCCM programme were estimated based on the budget of the iCCM project in the Bugoye sub-county and the iCCM costing report for Senegal. Household costs were obtained from costing studies [13‐15]. Overhead costs were distributed among the three illnesses based on the prevailing burden of malaria, pneumonia and diarrhoea within the community.

All costs were adjusted to the 2018 price level. For tradable resources, costs in Ugandan shillings (UGX) were exchanged to US dollars (US$) and inflated with US inflation rates. In contrast, the costs of non-tradable resources in UGX were inflated using local inflation rates and then exchanged to US$ [24]. The inflation rates were calculated based on the World Bank GDP implicit price deflators between 2010 and 2018 [25]. The average exchange was UGX3,727 = US$1 by the end of 2018 [26].

The total costs of resources for training drug sellers or CHWs, supervision, community sensitization, and management of the iCCM programmes utilizing drug sellers or CHWs were considered health sector costs because the government would need to fund these activities if any of the intervention was to be implemented at scale. The health sector costs mentioned above were estimated based on the budget for the iCCM project being implemented in the Bugoye sub-county in the Kasese district, and the iCCM costing report in Senegal [27]. The training of trainers/supervisors and initial training of drug sellers or CHWs, including the first three months of close support supervision, were considered capital items since their useful lifespan lasted more than one calendar year. These capital costs were annualized at a real discount rate of 3% assuming a useful lifespan of 5 years (annualization factor of 4.58) [28]. The less intense routine supervision and management costs incurred on an annual basis were considered recurrent costs. Costs for community sensitization were similar in both delivery models, thus eliminated.

Medicines, including artemether-lumefantrine, rectal artesunate, amoxicillin, ORS, and zinc, were supplied free of charge to CHWs, and at subsidized prices (mark-up of 50–80%) for drug sellers, and hence also considered health sector costs. Costs for medicines were obtained from the International Drug Price Indicator Guide (2014 edition). A recommended 10% was added to supplier price to cater for shipping cost (freight and insurance) [29] and the wastage rate was assumed to be 10% [30].

Other health sector costs covered mRDT, respiratory timers and health supplies such as stock cards, iCCM registers, examination gloves, cotton wool, and methylated spirit. All these items were supplied free to both drug sellers and CHWs. The prices for these previously mentioned items were obtained by the UNICEF supply catalogue (2016 edition) with 10% added to cater for freight costs [31].

In addition to these items, CHWs were also given medicine boxes, safety boxes, and incentives including raincoats, umbrellas, gumboots, solar lights, and hoes, which were valued at prevailing local prices.

Drug sellers and CHWs invested time in the iCCM programme with no formal pay. Drug sellers or CHWs spent around 30 min per consultation. The median time per household visit was also estimated to be 30 min for CHWs. It was assumed that drug sellers did not perform household follow-up visits for their clients. According to Kasteng et al., the opportunity cost attached to this volunteer time was estimated to be around US$0.34/h based on the median wage rate of alternative work for volunteering communities [32]. Drug sellers received subsidized pre-packaged drugs (ACT, amoxicillin, ORS, zinc), which were sold at a mark-up of 50–80% [7]. Other out-of-pocket costs incurred by CHWs that were not reimbursed included travel expenditure and mobile phone airtime expenditure of about US$2.4 and US$1.5 per annum, respectively [12].

Household costs incurred when visiting a drug seller or CHW were divided into direct medical, direct non-medical and indirect costs. Direct medical costs included drug costs or diagnostic costs, while costs such as transport to provider and costs for special food for a febrile child on treatment were considered as direct non-medical costs. Indirect costs related to productivity losses and referred to time spent travelling to a drug seller or CHW and time spent at or with the provider.

Concerning receiving care from CHWs, caregivers travelled an average of 20 min. The waiting times were generally low with caregivers spending only 30 min receiving care from CHWs; the overall time was about 50 [14]. It was assumed that caregivers who sought care from drug sellers spent about 30 min travelling to the provider and another 30 min receiving care from the drug seller [12]. The opportunity cost attached to the caregiver’s time was estimated to be around US$0.34/hour based on the median wage rate within the community [32]. The household costs were obtained from previously conducted household surveys [12, 14], with assumptions being made on the opportunity cost attached to the time the caregiver spent travelling and receiving care from the drug seller.

Data on the costs and effects were linked through a decision analytical approach [33]. A decision tree was used to model delivery of the iCCM intervention via drug sellers or CHWs as shown in Fig. 1. The probabilities corresponding to the individual tree branches (chance nodes) were obtained from iCCM intervention studies conducted by [6, 7, 12, 15, 34] and are shown in Additional file 1. At the end of each decision tree branch, the treatment of a child with suspected malaria, pneumonia or diarrhoea was classified as appropriate or not.

Societal costs for each decision tree branch were calculated by populating the decision tree with corresponding costs per drug seller or CHW, assuming each saw about 100 febrile U5 cases per year. The overhead costs, such as programme costs, were divided among the three illnesses based on the annual prevalence rate of malaria, pneumonia and diarrhoea for 2017/2018 [3]. The total societal costs and the number of appropriately treated U5 children were calculated by letting 100 children pass through the populated decision tree for each arm. The incremental cost was obtained by subtracting the total cost of the control (CHW) arm from that of the intervention (drug seller) arm. Incremental effect measured as the number of appropriately treated U5s was obtained by subtracting the total effect in the control arm from the total effect in the intervention arm. The incremental cost-effectiveness ratio (ICER) was then calculated as a ratio of the incremental cost (numerator) to the incremental effect (denominator). Therefore, ICER measured the extra cost per additional appropriately treated child when a drug seller is utilized to deliver iCCM instead of a CHW.

The results of this study were presented to guide policymakers. Based on the cost-effectiveness plane, the following classifications were used to explore in which circumstances it might be appropriate to support the drug seller-iCCM delivery model with public funds:

From a policymaker’s perspective, if the iCCM trained-drug seller delivery model dominated, it would justify the support of this strategy with public funds. However, this would not mean that the iCCM delivery model utilizing CHWs should be discontinued. The iCCM trained-drug seller model aims to complement the iCCM trained-CHW intervention, and not to substitute it, as a strategy to increase access to quality treatment.

If the iCCM trained-drug seller model were more costly and more effective, the decision on financing it or not would depend on the government’s willingness-to-pay (WTP), as there is no threshold to suggest how much extra money is reasonable to pay per additional case appropriately treated.

If the iCCM trained-drug seller model was less costly but less effective, it is not likely to be considered worthwhile using public funds unless the difference in effectiveness is minimal. In case the iCCM trained-drug seller model was dominated, this would suggest that the intervention should not be supported.

One-way sensitivity analyses were conducted by varying the values of individual parameters and assumptions in the decision models to observe the effect on the ICER. The parameters assessed included the probability of seeking care from an iCCM trained-drug seller or CHW, drug seller or CHW utilization of RDTs and respiratory rate timers, malaria, and pneumonia positivity rate.

Scenario analysis was performed for iCCM costs (50% increase or decrease), iCCM effectiveness (50% increase or decrease), and varying prevalence of malaria, pneumonia, and diarrhoea, to assess their impact on the ICER. Probabilistic sensitivity analysis was performed to assess the sensitivity of the ICER to simultaneous variation in the relevant model input parameters by defining probability distributions to selected parameters rather than point estimates [33]. Beta distributions were used for all branch probabilities, except the prevalence of malaria, pneumonia and diarrhoea that were assumed point estimates. Cost parameters were entered in the analysis as point estimates since there was insufficient data to derive their distribution [30]. Simultaneous selection of values from these parameter distributions and point estimates, followed by calculation of ICERs, was performed 10,000 times in Amua modelling software [17] to propagate uncertainty in the ICERs. Uncertainty surrounding the ICERs was summarized by plotting a cost-effectiveness acceptability curve (CEAC), which shows the probability that the use of iCCM delivery models to treat malaria, pneumonia and diarrhoea is cost-effective for different levels of a health policymaker’s hypothetical WTP for an appropriately treated U5 child.

Most cost [12] and effect data [6, 22] utilized were based on retrospective studies that interviewed U5 caregivers and sometimes drug sellers or CHWs, which thus might have been subjected to recall bias. Data for deriving the distribution of certain parameters like costs and effect were missing; thus, only point estimates were used in PSA.

The authors assumed no co-infection with malaria and pneumonia, malaria and diarrhoea, pneumonia and diarrhoea, or both malaria, pneumonia and diarrhoea. Thus, the possibility of a child presenting with any two or all three illnesses at the same time were not considered in the model.

Better data could improve the model’s accuracy: more effectiveness data for either iCCM trained-drug sellers or CHWs in various rural settings in Uganda based on randomized controlled trials, more data on the cost incurred by drug sellers or CHWs, and household costs in addition to iCCM programme costs.