Model design
Cost-effectiveness analysis and sensitivity analysis were performed using Markov model, with TreeAge Software (TreeAge Software, Inc., Williamstown, United States of America). The base-case model of this study was a Markov simulation of a hypothetical cohort of 100000 55/65/75-year-old residents underwent annual AF screening by 12-lead ECG compared to routine healthcare. In this study, the model length was 35/25/15 years (1 year circles) for 55/65/75-year-old cohort, respectively, until reaching 90 years old. The half-cycle correction was conducted for quality of life and cost every step in this model.
The base case scenario of this model was as following. A 55/65/75-year-old cohort of 100 000 was assigned to no screen or screen cohort. People at no screening cohort would have a certain group diagnosed with AF when they had symptoms and seek doctor’s help or went to hospital for other disease and diagnosed accidentally. People at screening cohort would be diagnosed with AF through screening method, and people who did not participant screening in the screening cohort would have the same chance to be diagnosed AF as people at no screening cohort. Furthermore, biennial and triennial 12-lead ECG screening in 55/65/75-year-old cohort were simulated to see the effect of screen intervals. The parameters used in these models were listed in table 1.
There were 5 independent health states in this model (Fig 1): no event, AF (with treatment & without treatment), ischemic stroke (IS) (with different severity rank), intracerebral hemorrhage (ICH) (disabled and fatal) and death. In one stage, there would be none or one and only clinical event happens. The clinical states could only progress forward and would not improve. AF, IS and ICH were assumed to be chronic diseases and could persist for a long time.
Some assumptions were made in this model. The anticoagulation therapy was warfarin prescription. Studies had proved that warfarin did a great job in IS risk reduction, minimizing the risk of bleeding with better persistence and can be cost effective than new oral anticoagulants (NOACs) [19, 20]. In this model, AF patients who was on warfarin was assumed to be adherence to it. And the AF patients who rejected anticoagulation therapy at diagnosis would never take the medication. Anticoagulation treatment with warfarin need relative frequently international normalized ratio (INR) monitoring in the first month and then INR monitoring monthly in the following long-term treatment. Patients developed with IS would be simulated to be at different disabled states in the acute phase: mild (modified Rankin Scale [mRS] score, 0–1), moderate (mRS score, 2–3), severe (mRS score, 4–5) and fatal (mRS score, 6), and would die of other causes and stroke after acute phase. And ICH patients were simulated the same way but without severity ranks.
The benefits in this model were risk reduction of ischemic stroke and all-cause mortality, increasing life years and life quality in patients of AF due to early detection and anticoagulation treatment. The costs in this model were expenses on treatment of AF, IS, ICH and screening. According to the Commission on Macroeconomics and Health of the World Health Organization, if the incremental cost-effectiveness ratio (ICER) was less than Chinese Yuan Renminbi (CNY) 70892 (1 gross domestic product [GDP] per capita of China in 2019, US $10635) per QALY gained, then the screening was considered highly cost-effective, and if the ICER was less then CNY 212676 (3 GDP per capita of China in 2019, US $31905) per QALY gained, then the screening was cost-effective.
Probabilities and rate
The resources of disease morbidity in this model was from the results of Global Burden of Diseases (GBD) 2019 study, including the incidence of AF, IS, ICH, the prevalence and mortality of AF and all-cause mortality in China. The GBD 2019 is the world’s most comprehensive catalog of surveys, vital statistics and other health-related data by the institute for health metrics and evaluation[21, 22]. The age-related IS risk ratio (RR) of AF patients were derived from the Framingham study[23]. The IS incidence of people with and without AF was calculated using attributable risk formula:
In the equations above, I1, I0 and I were the IS incidence of AF patients, people without AF, and general population, respectively. f was the AF prevalence of general population. RR was the relative risk of IS in AF patients.
The case-fatality of AF was estimated as quotient of AF mortality divided by AF prevalence. The distribution of stroke severity under different conditions was derived from the studies of Hart et al, Sorensen et al and Gabet et al. The case fatality of stroke after the acute phase were assumed based on a two-year retrospective analysis of Chinese population[24]. The risk reduction of IS and risk increase of ICH due to warfarin anticoagulation therapy were form the RELY study[25], and the all-cause mortality decrease data was obtained from a meta-analysis[20]. The anticoagulation rates and their ranges of no screen and screen cohort were assumed based on published literatures[18, 26].
Utility
The utility values used to calculate the quality-adjusted life year (QALY) were based on the EQ-5D scale. The baseline health state utility values of no event people were from pooled health surveys for England, to avoid overestimating the life qualities of no event population by simply defining the utility as 1[27]. The utility values of AF, IS with different severity status, disabled ICH were derived from the studies of Sullivan et al and Gage et al [28, 29].
Costs
This study was conducted from the societal perspective, and only direct healthcare cost was included, consisting of the expenses on screening program and anticoagulation medication, the economic burden of IS, ICH. The cost of 12-lead ECG screening was calculated as the average outpatient ECG examination fee in China, which was supposed to cover the depreciation charge of electrocardiogram machines and the salary of nurses or general practitioners for reading results. And the screening program management expenses, including questionnaires inquiry, staff training, expert literature, consulting, material design, printing, screening invitation, on site epidemiological research and so on, were from the handbook of stroke screening program of China pressed by government. The anticoagulation treatment fee of warfarin included medication cost and INR test expense[25, 30, 31]. The expenses due to IS and ICH were divided into first year acute hospital treatment and annual long-term stroke rehabilitation and healthcare. The data used in this model was from Chinese published literatures[5, 32].
Discount rate
All costs parameters used in this model were expressed in 2019 CNY with the discount rate of 5%. The values of utility and costs in this model were discounted 5% per year, with the sensitivity analysis range of 3% to 8%.
Uncertainty and sensitivity analysis
To analyze the uncertainty of some important parameters, deterministic 1-way sensitivity analysis was performed with parameters of plausible ranges. A probabilistic sensitivity analysis was also adopted in this model to study the variability of all variables. The proportion and utility variable were assumed to be beta distributions. The ratio risk (RR) variables were assumed to be lognormal distributions. Stroke severity ranks were assumed to be dirichlet distributions. Cost parameters were assumed to be normal distributions. All the variables above were varied simultaneously and 10000 estimates of costs and effects were obtained by sampling from the distributions.