Prevention and treatment of cardiovascular disease in Ethiopia: a cost-effectiveness analysis

Fifteen single interventions and sixteen integrated intervention packages were assessed. Interventions target individuals without a history of established CVD but at risk of developing a CVD event; those with an acute CVD event; and those with a history of established CVD event. Interventions were selected based on the recommendations of WHO and local experts and scientific evidence of effectiveness. Full description of the interventions is outlined in Table 1.

For primary prevention, individual-based drug regimens based on either the level of systolic blood pressure (SBP), the level of total serum cholesterol or the absolute risk of developing a CVD event over the next 10 years were assessed. Absolute risk is determined based on well-known CVD risk factors (age, gender, SBP, smoking status, body mass index and total serum cholesterol level) [18, 19]. The distribution of mean risk factor levels and smoking status in the population was stratified by age and gender based on the estimates from WHO’s Comparative Risk Assessment project for East Africa region. Estimates of relative risk of developing a CVD event per unit increase in the level of risk factors was then applied to estimate the individual level relative risk of developing a CVD event which is then used to extrapolate the absolute risk of CVD event at population level [19, 31]. The drug regimens are to be delivered on an outpatient basis at health centers and constitute: (a) a beta-blocker and a diuretic at SBP of >140 mmHg or >160 mmHg; (b) statin treatment at serum cholesterol level of >5.7 mmol/l or >6.2 mmol/l; and (c) a combination of aspirin, beta-blocker, diuretic and statin-based on the absolute risk of a CVD event for four thresholds (>5, >15, >25 or >35 %) respectively.

Interventions for acute myocardial infarction (AMI) constitute treatment with aspirin, streptokinase, clopidogrel, beta-blocker, ACE-inhibitor and surgical revascularization with percutaneous coronary intervention (PCI) on an inpatient basis. Aspirin is used for acute treatment of ischemic stroke; and beta-blocker, aspirin, ACE-inhibitor and statin for secondary prevention of IHD and stroke. Interventions were first assessed individually; clinically relevant packages were then formed, building on the intervention with the lowest cost-effectiveness ratio.

Given the current low coverage of interventions—less than 5 %, based on experts’ recommendations—we set modest target coverage of 20 % for all of the interventions.

In the absence of local evidence, efficacy estimates were drawn from previous randomized controlled trials and meta-analyses performed elsewhere (Table 2) [32‐46]. Efficacy estimates were adjusted by target coverage and patient adherence level [47‐49].

The WHO-CHOICE’s CVD model for East Africa was used to undertake the analysis [50]. The model was updated with age and sex distribution, birth rate and background mortality rate for Ethiopia [51‐53]. In the absence of national data on the current level of incidence, prevalence and mortality rates of IHD and stroke and the distribution of CVD risk factors, the analysis used respective estimates for the East Africa region [22, 23, 31, 50, 52].

The effect of primary prevention interventions is modeled through their impact on the level of risk factors, which is used to recalculate the expected incidence rate for IHD and stroke after implementing the specific intervention. The new incidence rate is applied to estimate the reduction in mortality from the respective diseases. Interventions targeting AMI and acute stroke were modeled through the interventions’ impact on 28-day case fatality rate, while secondary prevention interventions were modeled through their impact on post–acute case fatality rate. The effect of interventions was assumed to be the same across sub-groups.

We used PopMod, a multi-state population model, to estimate the health benefits in disability adjusted life years (DALYs) averted for the Ethiopian population resulting from changes in CVD risk due to specific interventions.

The population in the model is divided into age–sex categories of one-year intervals which are further stratified into four health states: (a) those having IHD; (b) those having stroke; (c) those having both; and (d) those without any of the conditions. Transition between states is dictated by the respective incidence, case fatality and mortality rates. Disability weights for the health states were drawn from the Global Burden of Disease Study 2010 [54]. PopMod traces the changes in population size in each age–sex category over a lifetime of 100 years by standard life table methods with and without specific interventions (‘no intervention’ scenario). Interventions are implemented for 10 years, after which the epidemiologic rates are taken back to the ‘no intervention’ level. Births and background mortality are taken into account [31, 55]. The expected health benefits of the current coverage level of interventions are eliminated to create a hypothetical reference case of null scenario. The model provides removal of the benefits of current coverage of interventions, thereby allowing recalculation of the incidence, prevalence and case fatality rates for MI and stroke, assuming a scenario where the currently implemented interventions are stopped. The health benefits are reported in terms of DALYs averted, discounted at 3 % per year without age weighting. The model has been used to undertake CEAs of various interventions in multiple settings [22]; and details have been published elsewhere [23, 24, 55].

A healthcare provider perspective was used for analysis and hence only program costs, training costs and patient-related costs to the provider were taken into account. Program costs constitute the cost of development and administration of an intervention at national and sub-national levels. This includes cost of administration and planning, media and communication, law enforcement, training, monitoring and evaluation. Patient-related costs consist only of direct medical costs incurred by the provider at the point of service delivery, including hospital bed days, outpatient visits, drugs and laboratory [28]. The analysis did not include direct non-medical costs such as transportation and indirect costs to patients and care givers such as lost productivity. The ingredients costing approach was employed whereby the quantities of resources required to deliver the interventions and respective unit prices were accounted for separately (Table 3). The quantities of resources used were largely determined based on WHO-CHOICE assumptions. We updated the prices of relevant laboratory tests and imaging using pricing from two public hospitals in Addis Ababa (Tikur Anbessa teaching hospital and Zewditu hospital). Salary scale of the health workforce was based on the FMOH of Ethiopia. Equipment and material prices were based on WHO price estimates for Ethiopia for the year 2012/13 [56] and drug prices were based on the lowest supplier prices for 2012, as noted in the International Drug Price Indicator Guide [57]. WHO-CHOICE’s transport multiplier factor was applied to the drug prices. The total cost of an intervention was then calculated as the sum of the product of the quantities of resources with their respective unit prices. As recommended by WHO-CHOICE costs were discounted at an annual rate of 3 % [28] and reported in 2012 US$.

All interventions were assessed compared to ‘no intervention’ scenario first, followed by incremental analysis between mutually exclusive interventions. Average cost-effectiveness ratios (ACERs) were estimated dividing the incremental cost by incremental effects of each intervention compared with a ‘no intervention’ scenario. In order to assess the relative cost-effectiveness of mutually exclusive interventions, incremental cost-effectiveness ratios (ICERs) were estimated as the ratio of the incremental cost to incremental effects for moving from one intervention to the next more effective intervention, starting from the null scenario. Interventions that are more costly and less effective than their comparators or those having higher ICER than their more effective comparator are designated as dominated. ACERs and ICERs are reported in US$ per DALY averted for the year 2012.

A probabilistic sensitivity analysis was conducted using Monte Carlo League (MCLeague) software to assess the effect of uncertainty surrounding the costs and effectiveness estimates [58]. A truncated normal distribution was used to execute 1000 simulation runs with 15 and 25 % coefficient of variation for costs and effectiveness estimates, respectively. We assessed interventions that were not dominated by respective comparators in each intervention category. In addition, one-way sensitivity analysis was undertaken, applying the lower boundary of the effectiveness range; doubling the price of drugs, procedures and laboratory tests; a zero discounting rate to health benefits; and applying 50 % of the effectiveness point estimates (Tables 1, 2).