We developed a decision tree that begins with ambulatory patients presenting with fever to health facilities in rural sub-Saharan Africa (Fig. 1 , Fig. 2 , Fig. 3 , Fig. 4 ), and proceeds through diagnosis and treatment to disease outcomes according to the sensitivity and specificity of each diagnostic strategy, the patient's age and malaria prevalence among patients. Typical facilities would include health centres and dispensaries staffed by nurses and perhaps clinical officers, and outpatient departments of district hospitals. Once given first-line treatment, patients were assumed to face the same probabilities, health outcomes and costs regardless of diagnostic method. Parameter estimates for initial diagnosis and treatment were extracted from recently published data. Parameters describing treatment seeking patterns, costs for programme implementation and secondary treatment, and duration of disease were based mainly on those used in previous models.12 ,13 (link) Expert opinion was relied on for probabilities of disease progression and mortality without appropriate treatment where reliable published data do not exist. Parameter values, sources, best estimates and probability distributions representing parameter uncertainty are available at: http://www.wpro.who.int/sites/rdt .
We assumed that health workers used the diagnostic test result in their clinical decision-making and that patients diagnosed positive for malaria received ACT and patients negative for malaria received an antibiotic such as amoxicillin. The proportion receiving antibiotics was varied in the sensitivity analysis. Best (most likely) estimates for drug efficacy were set at 85% for ACT in cases of malaria and 75% for antibiotics in bacterial disease. We assumed that antibiotics were not efficacious for malaria or viral illness, and that antimalarials did not cure bacterial disease. We assumed no coinfection between malaria and bacterial infections. Presumptive treatment on the basis of a history of fever was assumed to have perfect sensitivity and zero specificity. For RDTs we assumed a test detecting histidine-rich protein-2 (HRP-2) specific for P. falciparum, as 90% of malaria in sub-Saharan Africa is P. falciparum, with best estimates for RDT sensitivity and specificity of 96% and 95%, respectively.14 (link)-19 Microscopic diagnosis was based on best standard practice of district-hospital and health-centre general laboratories in sub-Saharan Africa, and assumed best estimates for sensitivity and specificity of 82% and 85%, respectively.20 (link),21 (link) We made comparisons according to all possible levels of endemicity of malaria expressed in terms of prevalence of parasitaemia in febrile outpatients presenting at facilities.
The chances of a febrile episode being fatal are far higher if associated with HIV infection.9 (link),22 (link),23 (link) Very high HIV prevalence would affect the decision tree parameters. To avoid a very complex decision tree structure, parameter values were set assuming that HIV prevalence was relatively low (about 10% of people five years old or over), which is typical outside southern Africa.
We calculated the incremental cost in US dollars (2002 prices) of changing from one diagnostic approach to another from the joint perspective of providers and patients, using the ingredients approach to calculate diagnosis costs, first-line drug costs and variable costs of second-line treatment.24 The costs of microscopy diagnosis included materials, staff time, training and supervision. RDT diagnosis included the unit cost of the test; diagnosis according to presumptive treatment was assumed to cost nothing. We assumed drug cost per adult dose to be US$ 1–2.4 for ACT and US$ 0.61–0.93 for antibiotics. We set the cost of RDT kits at US$ 0.6–1 and that of microscopy at US$ 0.32–1.27. Microscopy costs are dependent on workload and were based on a range of 1000 to 6800 or more diagnoses per year. For simplicity we assumed that microscopy was used only for malaria diagnosis, not for other diseases. All other costs of first-line treatment were excluded as they were assumed to be the same across diagnostic strategies. We included variable costs to providers and patients of any second-line treatment (drugs, reagents, food), but excluded fixed costs (buildings, equipment, supervision and most staff costs) as these would not change with numbers of patients. We assumed that unresolved uncomplicated malaria was treated with a second-line drug of the same price and efficacy as the first-line antimalarial. We assumed that secondary treatment for severe bacterial infection was an alternative antibiotic costing twice as much as first-line treatment. Costs associated with the management of neurological sequelae were excluded.
We measured health outcomes in terms of disability-adjusted life years (DALYs) averted, calculated according to the methods of Lopez et al. without age weights.25 We based life expectancies on a west African life table with a life expectancy at birth of 50 years.
The causes of non-malarial febrile infection vary from region to region and encompass diseases such as acute respiratory infections and bacterial meningitis. For simplicity, disability weights and durations for uncomplicated and severe non-malarial febrile illnesses were assumed to be the same as those for malaria. We assumed that bacterial illness was more likely than malaria to become severe, but that only 5–15% of these infections had bacterial causes, with the rest being self-limiting viral infections.
We did probabilistic sensitivity analysis with Monte-Carlo simulations (Palisade@Risk add-in tool to Microsoft Excel), and cost and health outcomes were generated stochastically (10 000 simulations). Policy-makers will wish to identify interventions that are less costly than the comparator and have better health outcomes, called dominant, and rule out those that are more costly and less effective, termed dominated. More costly and more effective interventions may be selected if they are thought to be good value for money. An intervention was defined as cost-effective if it was dominant or had an incremental cost per DALY averted under US$ 150. The value of US$ 150 was chosen in the base case, to represent a decision-maker's valuation of a healthy year of life. This was based on recommendations of the Ad Hoc Committee on Health Research Priorities, which stated that any intervention costing less than US$ 150 per DALY averted should be considered attractive in low-income countries.26
Additional sensitivity analyses were done by varying the parameter of interest and malaria prevalence according to the ranges inTable 1 . A full report of all results is available at: http://www.wpro.who.int/sites/rdt , where customized results specific to local settings can be generated online using an interactive model.
We assumed that health workers used the diagnostic test result in their clinical decision-making and that patients diagnosed positive for malaria received ACT and patients negative for malaria received an antibiotic such as amoxicillin. The proportion receiving antibiotics was varied in the sensitivity analysis. Best (most likely) estimates for drug efficacy were set at 85% for ACT in cases of malaria and 75% for antibiotics in bacterial disease. We assumed that antibiotics were not efficacious for malaria or viral illness, and that antimalarials did not cure bacterial disease. We assumed no coinfection between malaria and bacterial infections. Presumptive treatment on the basis of a history of fever was assumed to have perfect sensitivity and zero specificity. For RDTs we assumed a test detecting histidine-rich protein-2 (HRP-2) specific for P. falciparum, as 90% of malaria in sub-Saharan Africa is P. falciparum, with best estimates for RDT sensitivity and specificity of 96% and 95%, respectively.14 (link)-19 Microscopic diagnosis was based on best standard practice of district-hospital and health-centre general laboratories in sub-Saharan Africa, and assumed best estimates for sensitivity and specificity of 82% and 85%, respectively.20 (link),21 (link) We made comparisons according to all possible levels of endemicity of malaria expressed in terms of prevalence of parasitaemia in febrile outpatients presenting at facilities.
The chances of a febrile episode being fatal are far higher if associated with HIV infection.9 (link),22 (link),23 (link) Very high HIV prevalence would affect the decision tree parameters. To avoid a very complex decision tree structure, parameter values were set assuming that HIV prevalence was relatively low (about 10% of people five years old or over), which is typical outside southern Africa.
We calculated the incremental cost in US dollars (2002 prices) of changing from one diagnostic approach to another from the joint perspective of providers and patients, using the ingredients approach to calculate diagnosis costs, first-line drug costs and variable costs of second-line treatment.24 The costs of microscopy diagnosis included materials, staff time, training and supervision. RDT diagnosis included the unit cost of the test; diagnosis according to presumptive treatment was assumed to cost nothing. We assumed drug cost per adult dose to be US$ 1–2.4 for ACT and US$ 0.61–0.93 for antibiotics. We set the cost of RDT kits at US$ 0.6–1 and that of microscopy at US$ 0.32–1.27. Microscopy costs are dependent on workload and were based on a range of 1000 to 6800 or more diagnoses per year. For simplicity we assumed that microscopy was used only for malaria diagnosis, not for other diseases. All other costs of first-line treatment were excluded as they were assumed to be the same across diagnostic strategies. We included variable costs to providers and patients of any second-line treatment (drugs, reagents, food), but excluded fixed costs (buildings, equipment, supervision and most staff costs) as these would not change with numbers of patients. We assumed that unresolved uncomplicated malaria was treated with a second-line drug of the same price and efficacy as the first-line antimalarial. We assumed that secondary treatment for severe bacterial infection was an alternative antibiotic costing twice as much as first-line treatment. Costs associated with the management of neurological sequelae were excluded.
We measured health outcomes in terms of disability-adjusted life years (DALYs) averted, calculated according to the methods of Lopez et al. without age weights.25 We based life expectancies on a west African life table with a life expectancy at birth of 50 years.
The causes of non-malarial febrile infection vary from region to region and encompass diseases such as acute respiratory infections and bacterial meningitis. For simplicity, disability weights and durations for uncomplicated and severe non-malarial febrile illnesses were assumed to be the same as those for malaria. We assumed that bacterial illness was more likely than malaria to become severe, but that only 5–15% of these infections had bacterial causes, with the rest being self-limiting viral infections.
We did probabilistic sensitivity analysis with Monte-Carlo simulations (Palisade@Risk add-in tool to Microsoft Excel), and cost and health outcomes were generated stochastically (10 000 simulations). Policy-makers will wish to identify interventions that are less costly than the comparator and have better health outcomes, called dominant, and rule out those that are more costly and less effective, termed dominated. More costly and more effective interventions may be selected if they are thought to be good value for money. An intervention was defined as cost-effective if it was dominant or had an incremental cost per DALY averted under US$ 150. The value of US$ 150 was chosen in the base case, to represent a decision-maker's valuation of a healthy year of life. This was based on recommendations of the Ad Hoc Committee on Health Research Priorities, which stated that any intervention costing less than US$ 150 per DALY averted should be considered attractive in low-income countries.26
Additional sensitivity analyses were done by varying the parameter of interest and malaria prevalence according to the ranges in