2.5.1. Dependent variables.

• Cost-effectiveness of hypertension treatment based on 2020 ISH guidelines.

2.5.2. Independent variables.

• Patient-related (socio-demographic characteristics, heart disease knowledge, healthy lifestyle and heart disease risk perception, presence of comorbidity, type of medications, treatment adherence, shared decision making, health-related quality of Life)
• Health professional related (sociodemographic characteristics, qualification, experience, knowledge of evidence-based guidelines)

2.5.3. Cost related variables.

1. ○ Medical costs (inpatient hospital stay/hospitalization cost, outpatient clinic visit, drug acquisition costs, drug administration cost, laboratory test, and imaging study costs)
2. ○ Non-medical costs (transportation, meal, patient time cost due to treatment, cost due informal care by family or friends)
3. ○ Indirect costs (absenteeism, presenteeism, unemployment, early retirement, disability, premature death)

2.6.1. Sample size determination.

The sample size was determined by using the single population proportion formula by taking the prevalence of patients who controlled their BP as 14% from the WHO 2016 BP control rate report [21–23] and Z value of 1.96 at a 95% confidence interval. We added 10% for non-response rate and two for design effect due to multi-stage sampling technique involvement. Finally, a formula giving a larger sample size was used. A total of 407 hypertensive adult patients who are on follow-up care will be included.

= 185 + (185 * 10%) = 203.5

= 203.5 *2 = 407

Where: n = is the sample size

• Z²= standard normal deviation, set at 1.96, correspond to the 95% confidence interval
• d = is the desired level of precision/margin of error (0.05)
• p= prevalence of patients taking anti-hypertensive (p=28.4%), and q is 1-p.

2.6.2. Sampling techniques.

A multi-stage simple random sampling technique was used. We randomly selected three zones from a total of 12 zones found in the Southern region. Three general hospitals with experience in providing CVD care for at least five years from selected three zones were included in this study. The total sample size was allocated to these hospitals based on an estimated number of adult hypertensive patients attending respective hospitals (i.e., we included 212 patients from Arba Minch General hospital, 107 patients from Jinka general hospital, and 88 patients from Sawula general hospital). Finally, a consecutive sampling technique was applied in each facility until the desired sample size was achieved.

2.7.1. Disease states and transition probabilities.

The 2020 world population prospect estimate was used for the baseline population [29]. Coronary heart disease and stroke deaths in 2020 were extracted from the WHO-STEPS survey and systematic reviews. Coronary heart disease deaths and stroke deaths were CVD deaths and all other deaths were considered non-CVD deaths. The annual probability of CHD and stroke, data that are not available can be taken from well-accepted international studies like Framingham Heart Study due to lack of country-specific [30] and the Framingham Offspring Study [31], by contextualizing to Ethiopian scenario. The Framingham offspring studies 1–8 presented the 10-year CHD and stroke events, and we used annual CHD and stroke risk. We tested our annual risk assumption, by developing a scenario analysis with a linearly increasing risk of CVD events over the 10 years [32].

The mortality rate in 2015 stratified by sex and 5-year age groups in selected zones was used. We combined disease-related and disease-unrelated mortality in a multiplicative fashion. The demographic profile of the cohort was derived from the estimated resident population of Ethiopia in 2020. The rate of BP control was drawn from the National STEPs 2015 survey [24] and our effectiveness study. Identified rates were applied to projections from the 2019 United Nations World Population Prospects [33]. Transition probabilities (TP) and relative risk of mortality were taken from the natural history of hypertension studies with good quality [34–39] (S1 and S2 Tables).

The prevalence of cardiovascular risk factors (MI, angina, heart failure, stroke, TIA) was estimated from systematic reviews and meta-analyses. Incorporating the risk factor prevalence data in the relevant Framingham risk equation, the age and sex-specific probability of CHD and cerebrovascular disease (i.e., stroke and TIA) events were estimated. The probability of each health state was calculated using the age- and sex-specific CHD and cerebrovascular disease event distributions [24, 40]. To estimate the corresponding probabilities, separate relative risk estimates were used for CHD events (Angina, HF, and MI) and cerebrovascular diseases (stroke and transient ischemic attack). We assumed antihypertensive treatment affects the probability of every disease state similarly across all age and sex groups. Relative risk reductions attributable to antihypertensive treatment were extracted from the peer-reviewed literature [39, 41, 42]. We estimated the probability of death separately for (1) all-cause mortality in absence of hypertension and related complications for the productive age population and (2) mortality attributable to the included disease states. The first component was estimated using Ethiopian Life Table 2019, and the second component was calculated based on standardized mortality ratios extracted from literature [34] (S2 and S3 Tables).

Interventional trials suggested that it could be possible to achieve effective BP targets in about 70% of patients by improving adherence and/or intensifying therapy [35]. Incident coronary heart disease events were allocated to angina pectoris, MI, or cardiac arrest. Prevalence and joint distributions of Ethiopia’s risk factor values were estimated from the national STEPS survey [24]. Betas for risk function for non-blood pressure risk factors were estimated separately for the risk of incident coronary heart disease events, incident strokes, and non-CVD deaths from the Framingham offspring cohort [31]. Risk factors are assumed to affect the incidence of MI, arrest, and angina in proportion to the overall incidence of CHD, except tobacco smokers are assumed to have a higher relative risk for infarction and arrest [43]; and a proportionately lower coefficient for angina [24, 44]. Case-fatality rates, rates of MI, prehospital arrest deaths, and out-of-hospital cardiac arrests surviving to hospital discharge were estimated from evidence-based literature (S2 and S3 Tables).

Survival after a CHD event was estimated from international data sources (California data on the ratio of in-hospital survival to 30-day survival) [45] and calibrated based on the findings of Huffman et al. [46]. Stroke incidence was assumed to be independent of the risk of new-onset coronary heart disease in the same year. The number of hospitalized stroke cases was obtained from national and regional studies. The annual probabilities of stroke after MI [47, 48] and the probability of CHD in stroke patients were based on natural history studies and systematic reviews of BP control trials [49–54]. A 30-day heart failure mortality and re-hospitalization data were from the THESEUS-HF registry [55] and Korean Acute Heart Failure Registry (KorAHF) [56, 57] (S2 and S3 Tables). The background prevalence of CVD by age, sex, and CVD disease state (stroke, coronary heart disease, or both stroke and coronary heart disease) in 2020 was estimated from GBD 2017 [58] (S2 Table).

2.7.2. Cost estimation.

Both direct and indirect costs were included in this study. The direct costs were divided into two subcategories: direct medical costs and direct non-medical costs. Direct medical costs include; inpatient stays, outpatient clinic visits, medical services, drug acquisition, dispensing, administration, monitoring, laboratory test, and imaging study costs. The costs associated with outpatient/inpatient visits were estimated by multiplying the numbers of outpatient visits related to hypertension by the outpatient costs per year (i.e., twelve times WHO cost per outpatient visit for secondary hospitals inflated to 2021) [59].

Data concerning medications prescribed for the management of hypertension, and associated comorbidities, and laboratory tests and imaging studies were collected by patient chart abstraction in index year (2020). The cost of medications used for management of hypertension and associated comorbidities was taken from Ethiopian Pharmaceutical supply agency Arba Minch regional hub selling price and retail price of Arba Minch General Hospital in 2020. The retail price of Arba Minch General Hospital was used because of the minimum distance from the Pharmaceutical supply agency hub, which could minimize the markup added to the retail price due to transportation costs. Costs of laboratory tests and imaging studies were also taken from Arba Minch Hospital Laboratory’s service price list. The salary scale of the health workforce was based on the federal ministry of health (FMOH) of Ethiopia (S4 Table).

Ongoing program costs for hypertension care were estimated from WHO tool outputs for CVD and diabetes care and the National strategic action plan (NSAP) for prevention & control of non-communicable diseases in Ethiopia from 2014-to 2016 and adjusted for the 2021 inflation target population [60]. Adjustment for the study population was done by multiplying the national cost by the proportion of the study population (i.e., 3%). National and regional cost estimates were based on the proportion of patients studied (i.e. 3% and 20%). We considered this strategy since the age and sex distribution of hypertension among different regions in the country did not vary significantly. The collected cost data were added up and averaged by using a bottom-up approach. Facility-based or reference costs were used during computing costs. The total medical cost of hypertension treatment was calculated as the sum of the product of medical costs with their respective unit prices. Costs were discounted at an annual rate of 3% and reported in 2021 USD [13, 61].

Direct non-medical costs include transportation costs and patient time costs due to care. The cost of patient time due to care was estimated by using the average daily wage of the patient (97.00 WTB) which was calculated from the average monthly income (2910.00 ETB) from our treatment effectiveness survey. Transportation cost was determined by using the cost of average traveling distance and local transportation tariff (42.00 ETB) in January 2021. According to EDHS 2016 survey showed that 33% of women and 88% of men are currently employed [34]. This proportion was used to determine the patient time cost due to care for employed groups. For the unemployed proportion, the average daily wage of daily laborers workers working 8 hours per day for 6 days per week was used (26.53 ETB) from the monthly wage of 796.00 ETB (420–1172 ETB) [62].

Indirect costs include productivity loss due to illness and the cost of death. Cost-of hypertension-related hospitalization was taken from WHO Choice [59], costs per inpatient stay and cost per inpatient bed day times duration of hospitalization inflated for 2021, and professional time (physician, nurse laboratory professional, and pharmacist time). If a patient had multiple admissions during the year, the costs for each admission were aggregated as the total costs [63]. Age and sex-specific mortality rates among the adult general population in Ethiopia were taken from EDHS 2016 survey and extrapolated to selected populations [34]. According to EDHS 2016, the probability of dying before age 50 years among adults ≥ 15 years was 10% and 12%, in women and men respectively [34]. Mortality risk in the general population attributed to those with and without hypertension using sex-specific estimates of the relative risk (RR) of all-cause mortality associated with hypertension by treatment and control status was derived from a study conducted in India [64]. A cohort study conducted in India among adults 20 years and above to determine the rate and risk of all-cause mortality among people with HTN showed that the incidence of deaths in the study was 4.28%. The relative risk of mortality was 3.13 (CI: 2.91–3.37) and 1.2 in the high BP group and at age of 60 years. The age-adjusted hazard ratio of all-cause mortality for the high BP group was 2.96 (2.56–3.42) [64] (S6–S9 Tables).

In 2020 crude death rate of the Ethiopian population-based on global estimates was 6.29 deaths per 1000 population (i.e. 680,032 deaths per 108,113,150) [65]. The estimated prevalence of hypertension among adults was calculated from National STEPS Survey 2016, systematic review and meta-analysis, World health organization report, and local studies (19.6%, for 15–30 years, 23% for 30–40 years, 25.9% for 40–49 years and 41.9% for 50 years and above [24, 34, 64, 66–69]). (S3 Table). The mean estimated prevalence of hypertension is 21.39%. The mean relative risk (hazard ratio) of all-cause mortality among the hypertensive population when compared to those without hypertension was 1.39 (0.95 to 1.95) [70]. Only 28.4% of patients with the diagnosis of hypertension were taking antihypertensive medication [24].

Years of life lost due to hypertension were determined by first calculating disability weights for specific ages based on BP control status (X). Then subtract this value (X) from the life expectancy of the Ethiopian population (i.e., 66.7 years for men, and 70.4 years for women) (Y). The productivity loss cost due to hypertension was calculated by multiplying Y with the sex-specific employment rate based on a monthly average income of 2059.078 ETB from the National STEPS survey 2015 adjusted for 2021 inflation (1.372) STEPS Survey, 2015 [24] and EDHS 2016 survey showed that 33% of women and 88% of men are currently employed [34] and for unemployed, 2019 minimum average monthly earnings (ETB) of daily laborers reported by the Ethiopian Ministry of Labor and Social Affairs (MOLSA) 796 ETB (420–1172 ETB) [62]. Concerning, the cost of productivity lost due to premature mortality: first we calculated potential years of life lost (YLL) by subtracting life expectancy from the sex-specific age of death at which the death is recorded (Z). Then Z is multiplied by the number of deaths in each age group (Xi). Finally, we multiplied Xi with sex-specific employment rates like productivity loss due to hypertension-related morbidity above [71]. Excess mortality and morbidity due to hypertension to hypertension were determined by subtracting age and sex-specific morbidity and mortality among the general population from the hypertensive cohort. Both were determined by using age, sex, and blood pressure treatment status mortality rate per 1000 person-years (S9 Table).

2.7.3. Calculating Disability Adjusted Life Years (DALYs).

Disability-adjusted life years (DALYs) are composed of years lived with disability (YLDs) and years of life lost due to premature mortality (YLLs) [72]. The YLDs arising from a disease or injury are the sum of the YLDs for each of the sequelae associated with that disease. The GBD 2015 study estimated YLDs by country, age, and sex for 2619 sequelae of 310 diseases and injuries [72]. The YLDs (morbidity component) of the DALYs, are calculated as follows [4]:

Disability weights of hypertensive heart disease, complications, and comorbidities were taken from WHO estimates of global health state weights, systematic reviews, and clinical or observational studies [73]. According to the Global burden of disease 2016 report, the average disability weight of hypertensive heart disease was 0.246 (0.201–0.300). Untreated 0.323 and treated and controlled 0.171 [74, 75] (S2 Table).

Years lost to live (YLLs), the mortality component of the DALYs, is calculated as follows:

DALYs for a specific cause are calculated as the sum of the YLLs from that cause and the YLDs for people living in states of less than good health resulting from the specific cause:

• For a given cause c, age a, sex s, and year t

According to the GBD, 2010 study simplified calculation of DALYs was recommended. Therefore, we used a new normative standard life table for the loss function used to compute YLLs; calculate of YLDs simply as the prevalence of each sequela multiplied by the relevant disability weight; adjust for comorbidity in the calculation of YLDs, and no discounting for time or unequal age weights [76].

2.7.4. Adjusting for comorbidity.

Patients with hypertension may have more than one disease, the addition of YLDs across causes may result in an overestimation of the total loss of health [77]. Therefore, it is recommended to estimate comorbidities using the assumption of independence within age-sex groups [4]:

Where P₁₊₂ is the prevalence of the two comorbid diseases 1 and 2,

P₁ is the prevalence of disease 1 and P₂ is the prevalence of disease 2.

The combined disability weight for individuals with multiple conditions is estimated assuming a multiplicative model as follows:

Since prevalence YLDs are calculated for each cause as:

Two preceding equations can be combined into a single calculation resulting in:

2.7.5. Cost-effectiveness of hypertension treatment.

Generalized CEA was conducted from a societal perspective to evaluate the cost-effectiveness of hypertension treatment based on the ISH 2020 guideline [78]. Generalized cost-effectiveness analysis (GCEA) was used to derive average cost-effectiveness ratios (ACERs) for the full implementation of ISH guidelines. The Cost-effectiveness of interventions is judged by their incremental cost-effectiveness ratio (ICER) which is given as the ratio of the incremental cost of the intervention to its incremental health gain relative to a comparator. The ratio was reported as cost expressed in monetary units per health gain (e.g., the cost in USD per disability-adjusted life year (DALY) averted. ICER informs us of how much additional cost the intervention under consideration requires for a unit increase in health benefits over its comparator. Therefore, the lower the ratio, the more cost-effective the intervention is [13]. ICERs equal to one to three times the per capita gross domestic product of Ethiopia (936 USD in 2020) [79] were considered cost-effective. An ICERs below USD 936.30 per DALY averted was considered very cost-effective and above 2,808.90 USD (122,187.15 ETB) were not cost-effective. In addition to this, we used half of the GDP per capita threshold (468.15 USD or 20,364.53 ETB) as it is more in accord with countries’ realities than the 1–3 GDP per capita, and could be used as an interim rule of thumb rather than the 1–3 GDP rule [80]. Results were presented based on Consolidated Health Economic Evaluation Reporting Standards (CHEERS) [81].

2.7.6. Model assumptions and transition probabilities.

When the hypertensive population enters the model they can be in one of two health states: a treated and untreated hypertension. People in a treated/untreated state move to a treated controlled/uncontrolled state, then people in a controlled or uncontrolled state can move to a non-fatal CHD and stroke event state or the dead state. Once people are diagnosed as having hypertension, they receive antihypertensive treatment, and they will get a reduction in their risk of having a cardiovascular event [82]. We assumed that truly normotensive people can become hypertensive over time because there is an additional possibility in a cycle that they will have a BP check-up annually and move back to a hypertension state. The probability will be estimated from the annual probability of developing hypertension based on age and sex distribution. Once a person has moved to a non-fatal event state (MI, UA, SA, stroke, or TIA) they stay there unless they die. This was considered a reasonable simplification for modeling purposes and also reflected in 12 trials where an individual is censured at their first event. The non-fatal CHD and stroke events were assumed to have two states; event and post-event. This is so that a different cost can be applied in the first cycle reflecting acute management and/or diagnostic costs. The event state is a tunnel state where patients move automatically to the post-event state in the next cycle (unless they die). The probability of death due to the CVD states will vary by type of event. Once people have moved to the dead state in the model they cannot of course move elsewhere; this is known as an absorbing state. To avoid modeling errors, we used the TECH-VER verification checklist containing five domains [input calculations; event-state calculations; result calculations; uncertainty analysis calculations; and other overall checks (e.g. validity or interface)] to reduce errors in models and improve their Credibility [83].

2.7.7. Blood pressure treatment assumptions.

According to guidelines, after a brief trial period of lifestyle interventions, patients with hypertension selected for pharmacologic treatment were grouped into those with or without diabetes or CKD. For the initial choice of antihypertensive agent, blacks were recommended to start with any thiazide diuretic, calcium channel blocker (CCB) not necessarily ACEIs/ARBs. Patients with diabetes and CKD were recommended to start with an ACE or an ARB. The BP response to different agents is similar in most patients [84] and we chose not to model differential use of antihypertensive medication classes in order not to bias cost-of-treatment inputs. Antihypertensive dose intensification and frequency of BP monitoring were based on the heart’s technical package guideline for blood pressure control. We did not simulate the effects of any particular medication; instead, we simulated “standard dose” effects and assumed average drug prices across classes [85] (S10 and S11 Tables).

The amount of BP change was assumed to be a function of the baseline BP and the effect of a standard-dose antihypertensive agent at that pre-treatment level [86] (S9 and S10 Tables). It is important to note that for patients with very high BP (mean SBP of ≥185 mmHg) it is assumed that even with taking four standard dose medications, these patients would on average achieve a BP of about 143 mmHg, but not a target of <140 mmHg. Another important consideration is the effect of true treatment-resistant hypertension (TRH) [87]. We did not include the effect of TRH on costs and outcomes due to a lack of national data. Other important assumptions include, that hypertensive deaths reported or recorded on hospital registries were considered hypertension-related unless specified as due to other causes. All Disabilities in hypertensive patients are considered as disabilities due to hypertension except for comorbidities and accidents. Both costs and outcomes for cost-effectiveness analysis were discounted at 3 and 6% discount rates.