Cost-effectiveness of Screening for Atrial Fibrillation Using Wearable Devices

Key Points Question Is population-based atrial fibrillation (AF) screening using wearable devices cost-effective? Findings In this economic evaluation of 30 million simulated individuals with an age, sex, and comorbidity profile matching the US population aged 65 years or older, AF screening using wearable devices was cost-effective, with the overall preferred strategy identified as wearable photoplethysmography, followed conditionally by wearable electrocardiography with patch monitor confirmation (incremental cost-effectiveness ratio, $57 894 per quality-adjusted life-year). The cost-effectiveness of screening was consistent across multiple scenarios, including strata of sex, screening at earlier ages, and with variation in the association of anticoagulation with risk of stroke associated with screening-detected AF. Meaning This study suggests that contemporary AF screening using wearable devices may be cost-effective.


Event-related costs
For clinical events modeled (e.g., ischemic stroke, intracranial hemorrhage, and major bleeding), upfront costs were stratified by severity and obtained from the Agency for Healthcare Research and Quality (https://hcupnet.ahrq.gov/#setup) as follows: First, we extracted separate cost statistics for all International Classification of Diseases, 10 th revision (ICD-10) diagnosis codes corresponding to the event of interest. Then, we sorted the costs in ascending order and divided them into quantiles equal in number to the categories of severity (e.g., tertiles for mild/moderate/severe groupings). Within each quantile, we utilized the mean hospital cost as the base case cost for the event at the corresponding severity level. The lower and upper bounds were set as the minimum and maximum cost values observed within the quantile.
In cases where one has multiple competing event-related costs, either the most relevant cost is incurred, or the maximum of the costs is incurred. For example, a history of stroke is associated with a maintenance cost associated with chronic poststroke care. If a recurrent acute stroke occurs, only the upfront cost corresponding to the new stroke is invoked (since it is greater than the maintenance cost associated with chronic post-stroke care), with no additional maintenance cost.

Drug/visit costs
In cases where anticoagulation was stopped due to a history of major bleeding, or in accordance with modeled discontinuation rates, we assumed that the monthly drug cost would stop accumulating until the treatment regimen was resumed. We also assumed that physician visits for acute events (e.g., major bleeding) would also fulfill potential maintenance visit requirements. For example, if an individual on anticoagulation has a physician visit secondary to an acute bleed, that individual's next annual physician visit for anticoagulation maintenance would be no less than one year after the acute bleed.

Screening costs
For discrete screening modalities, namely single-lead ECG, 12-lead ECG, pulse palpation, and patch monitor, a one-time screening cost was incurred if and only if the test was performed.
For costs associated wrist-worn wearable screening, a one-time upfront cost was incurred upon the start of screening (corresponding to initial purchase of the device) and an additional cost of replacing the device every five years was applied as long as the given strategy called for continued wearable screening.
For all screening strategies, a one-time nurse visit cost was incurred upon screening. Also, for strategies involving a wrist-worn wearable followed by a confirmatory patch monitor, an additional nursing visit cost was incurred after an abnormal wearable signal for prescription and application of the patch monitor.
Lastly, a physician visit cost was incurred for all instances where an ultimate diagnosis of AF was made (either true or false positive), corresponding to diagnosis counseling and prescription of anticoagulation if appropriate (i.e., no history of major bleeding).

Modeling of paroxysmal AF
Given lack of reliable data regarding the test characteristics of wearable devices for detecting paroxysmal AF over longer durations of monitoring (i.e., months to years), we modeled the temporal effect of screening via a wearable device as follows: We applied literature-based values for the estimated prevalence of paroxysmal AF among individuals with screen-detected AF (59%). [1][2][3][4] We then utilized estimates of the average AF burden among individuals with paroxysmal AF (4.5%). [4][5][6] We assumed that the average AF burden follows a uniform distribution on the order of days (i.e., an individual with an AF burden of 4.5% would be expected, on average, to spend 4.5% of each day in AF).
Then, the probability that an individual will not experience a single AF episode over t days is (1-0.045) t . The probability that an individual will experience at least one AF episode over t days is the complement, or 1-(1-0.045) t . We then applied the known static test characteristics of the wearable device to the probability of observing AF with each cycle of simulation (i.e., one month or 30 days).
For example, an individual with AF wearing a watch for 3 months would have a probability of the device being exposed to an AF episode after one cycle of 1-(1-0.045) 30 , or 0.749. If this individual is wearing a W-PPG (sensitivity 95.3, specificity 99.7), they will be diagnosed with AF with probability 0.749 * 0.953, or 0.714 after one cycle. As with other screening modalities, if a diagnosis of AF is not made, and the screening strategy under evaluation includes continued screening, then the screening process will repeat as dictated by the length of the screening interval being evaluated. In this case of 3-month screening, screening would continue for three cycles, with a probability of being diagnosed with AF of 0.714 after each cycle, and the overall probability of being diagnosed with AF of 1-(1-0.714) 3 or 0.977.
Although the data provided by a recent study by Diedrichsen et al. are insufficient to primarily inform test characteristics over the necessary durations required to model wearable screening approaches, we were able to validate that our approach described above resulted in comparable estimates of sensitivity for paroxysmal AF at 30 days, after allowance for the uncertainty in AF burden, which we modeled in probabilistic sensitivity analyses (

Sensitivity analysis assumptions
In cases where uncertainty in model parameters could not be estimated based on the available published literature, we varied point estimates by +/-20% when performing both one-way and probabilistic sensitivity analyses.

Simulation size determination
To determine sufficient cohort size for base case simulation taking into account firstorder uncertainty (i.e., Monte Carlo error), we followed the guidelines provided by the ISPOR-SMDM Modeling Good Research Practices Task Force Working Group-6. 8 Specifically, we tested results at increasing sample size from 10 million to 50 million and noted the comparative clinical effectiveness of all 8 screening strategies with respect to no screening, i.e., d(QALY), as well as the cost effectiveness results for all 5 cases. We report these values in the tables below. At a precision of 0.001 (i.e., 100 QALYs per 100,000 persons), one can see that d(QALY) is well-stabilized at simulation sizes at or above 30 million (Table B). Further, the cost-effectiveness strategy remained the same for all simulation sizes and the ICER stabilizes at a precision of $100,000 at or above 30 million ( Table C). As a result, we utilized a simulation size of 30 million for the base case analysis.
eMethods Defined using NHANES 2013-2016 health interviews. Coronary heart disease was considered present if a person reported "yes" to being told by a healthcare professional that he or she had coronary heart disease, angina or angina pectoris, heart attack, or myocardial infarction. Those who answered "no" but were diagnosed with angina based on the Rose questionnaire were also included. * denotes baseline condition ICER = incremental cost-effectiveness ratio; Freq = frequency; PP = pulse palpation; 12L = 12-lead electrocardiogram; PPG = wearable photoplethysmography; 1L = wearable single-lead electrocardiogram; PM = patch monitor; AF = atrial fibrillation; RR = relative risk; OAC = oral anticoagulant; NOAC = novel oral anticoagulant