1. Statistical Challenges Related to Population Screening for AFChristine M. Albert, MD, MPHDirector, Center for Arrhythmia Prevention, Brigham and Women’s HospitalProfessor of Medicine, Harvard Medical School

2. Screening: AF Prevalence • Effectiveness of Screening • Sensitivity and Specificity of the Test • Prevalence of AF in the Population. . Relationship between Disease Prevalence and Positive and Negative Predictive Value • False Positives • Expense of monitoring • Exposure to risks of AOC without benefit Sensitivity 95% and Specificity 85%

3. SAFE Trial: Cluster Randomized Trial of Office -Based ECG Screening for Atrial Fibrillation Overall Yield: 0.6 % new AF Focused on “Men over 85” 4.2% AF Fitzmaurice et al. BMJ 2007l

4. Mass Screening for Atrial Fibrillation in 75 Year OldsThe STROKESTOP Study • Screened twice daily with a handheld device for 2 weeks • AF Defined as: Irregular rhythm lasting 30 secs, or 2 episodes of at least 10 seconds 1% of Unknown AF Diagnosed on Baseline ECG 12% • Add > 1 stroke risk factor to 75+: 7.4% have undetected AF. Svennberg E et al. Circulation 2015;131:2176-84 Engdahl J et al. Circulation 2013: 127:930-93

5. Characteristics of Patients with SCAF Detected Female sex, lower weight, and absence of vascular disease were significantly associated with no detection of AF. CHA2DS2-VASc was not associated with AF detection Svennberg E et al. Circulation 2015;131:2176-84

6. Remote Heart Rhythm Sampling Using the AliveCor Heart Monitor to Screen for Atrial Fibrillation Twice weekly iECGs over 12 months • The REHEARSE-AF Study • 1001 Patients • Age>65 • CHADS-VASc >2.0 19 versus 5 diagnosed with AF Log-rank P=0.004 (Mantel-Cox). 1.2% Cost per AF diagnosis of $10,780 Kaplan-Meier plot showing the estimated detection probabilities for atrial fibrillation (AF) in each study arm over the 52 weeks of the trial. Shaded areas represent 95% confidence regions. Halcox JPJ et al. Circulation 2017; 136:1784-1794

7. The mHealth Screening to Prevent Strokes (mSToPS)Trial Impact of Immediate Monitoring with a 2 week ECG-Patch Intention to Treat Results: Incidence of new AF cases: • AF Incidence 3.9% (53/1366) in the immediate monitoring group vs 0.9% (12/1293) in the delayed monitoring group • Absolute difference, 3.0% [95% CI,1.8%-4.1%]). Population: 2659 individuals Randomized: • Age 75 years or older • Male > 55 years or Female > 65 years with 1 or more comorbidities Intervention: • Randomized to Immediate (within 2 weeks) versus delayed monitoring (within 4 months) • 3476 Matched Control Group with no monitoring • 34.5% did not wear the patch Endpoint: • >30 seconds of AF or new AF diagnosis in Claims Data Steinhubl SR et al. JAMA 2018; 320:146-155

8. US Taskforce: Analytic Framework: Screening for Atrial Fibrillation With Electrocardiography Jonas DE et al. JAMA 2018; 320:485-498

9. Screening for Atrial Fibrillation with Electrocardiography Systematic Review for the US Preventive Services Task Force Inadequate Evidence to Support ECG Screening for AF Jonas DE et al. JAMA 2018; 320:485-498

10. ECG Screening for AF • Sample Calculation of NNS for AF With ECG to Prevent 1 Stroke • New AF found on initial ECG in 0.5% of screened population • NNS to diagnose 1 new case of AF, 1/0.005 = 200 people • Estimated absolute risk reduction from anticoagulation, 2% • Number needed to treat to prevent 1 stroke, 1/0.02 = 50 people • NNS to prevent 1 stroke, 200 x 50 = 10,000 people Mandrola J et al. JAMA Internal Medicine 2018; 178:1296-1298 Kinsinger LS et al. JAMA Internal Medicine 2017; 177:399-406

11. Stroke Stop: TwoWeek Monitor Screening for AF NNS to Prevent 1 Stroke in Patients over Age 75 Sample Calculation: • New AF found in 3.0% of screened population • NNS to diagnose 1 new case of AF, 1/0.03 = 33 people • Estimated absolute risk reduction from anticoagulation, 2% • Number needed to treat to prevent 1 stroke, 1/0.02 = 50 people • NNS to prevent 1 stroke, 33 x 50 = 1650 people • Svennberg E et al. Circulation 2015;131:2176-84

12. Stroke Stop:Two Week Monitor Screening for AF NNSto Prevent 1 Stroke in Patients over Age 75 with 1 Stroke Risk factor Sample Calculation: • New AF found in 7.4% of screened population • NNS to diagnose 1 new case of AF, 1/0.074 = 13 people • Estimated absolute risk reduction from anticoagulation, 2% • Number needed to treat to prevent 1 stroke, 1/0.02 = 50 people • NNS to prevent 1 stroke, 13 x 50 = 650 people Svennberg E et al. Circulation 2015;131:2176-84

13. Sensitivity and Specificity of Different Methods of Screening for Atrial Fibrillation Freedman B et al. Circulation 2017; 135:1851-1867 Paper #6: Table 4

14. Passive Detection of Atrial Fibrillation using Neural Network Analysis of a Commercially Available Smartwatch 51 patients in Health eHeart Study ECG Diagnosis 1617 patients in Health eHeart Study Self Report of Persistent AF C-Statistic 0.97 C-Statistic 0.72 Sensitivity 98.0% Specificity 90.2% Sensitivity 67.7% Specificity 67.6% Tue MP et al. JAMA Cardiol 2018; 3:409-416

15. Screening for Atrial Fibrillation: False Positives: Importance of Prevalence Sample Calculation of Misdiagnosis at ECG Specificity of 95% • Screening Specificity, 95% • Screened population, 1,000,000 people • AF prevalence, 3% • Number with AF, 0.03 x 1,000,000 = 30,000 people • Number without AF, 1,000,000 – 30,000 = 970,000 people • True negative, 0.95 x 970,000 = 884,450 people reassured they have no AF • False positive, 0.05 x 970,000 = 48,500 people falsely diagnosed with AF Mandrola J et al. JAMA Internal Medicine 2018; 178:1296-1298 Paper #2: Box 2

16. Screening for Atrial Fibrillation: False Positives: Importance of Prevalence Sample Calculation of Misdiagnosis at ECG Specificity of 98% • Screening Specificity, 98% • Screened population, 1,000,000 people • AF prevalence, 10% • Number with AF, 0.10 x 1,000,000 = 100,000 people • Number without AF, 1,000,000 – 100,000 = 900,000 people • True negative, 0.98 x 900,000 = 882,000 people reassured they have no AF • False positive, 0.02 x 900,000 = 18,000 people falsely diagnosed with AF Modified from Mandrola J et al. JAMA Internal Medicine 2018; 178:1296-1298 Paper #2: Box 2

17. Subclinical Atrial Fibrillation detected on Continuous Monitoring (PPM or ILR) REVEAL-AF ASSERT-I ASSERT-II Van Gelder IC et al. Eur Heart J 2017; 38:1339-1344 Healey JS et al. Circulation 2017; 136:1276-1283 Reiffel JA et al. JAMA Cardiol 2017; 2:1120-1127

18. What are the implications of detecting short durations of AF?

19. ASSERT: Ischemic Stroke/Systemic Embolism According to Time-Dependent Durations of AF Van Gelder IC et al. European Heart Journal 2017; 38:1339-1344

20. Uncertainty of AbsoluteStroke Rate in Population (events per 100 person-years) Resultant Variability in NNS Quinn GR et al. Circulation 2017; 135:208-219

21. Reported Stroke Rates Stratified by CHA2DS2-VASc Scores of 0, 1, and 2 According to Different Cohorts *<1% annual stroke rate, low or no expected net clinical benefit from anticoagulation. †1% to 2% annual stroke rate, indeterminate expected net clinical benefit from anticoagulation. ‡>2% annual stroke rate, high expected net clinical benefit from anticoagulation. ACTIVE indicates Atrial fibrillation Clopidogrel Trial with Irbesartan for prevention of Vascular Events; AF, atrial fibrillation; ATRIA, AnTicoagulation and Risk Factors In Atrial Fibrillation; AVERROES, Apixaban Versus Acetylsalicylic Acid [ASA] to Prevent Stroke in Atrial Fibrillation Patients Who Have Failed or Are Unsuitable for Vitamin K Antagonist Treatment; J-Rhythm, Japanese Rhythm Management Trial for Atrial Fibrillation; and NHIRD, National Health Insurance Research Database. Quinn GR et al. Circulation 2017; 135:208-219

22. Statistical Challenges for Population Screening for AF Summary • Positive Predictive Value (PPV) of AF Screening in the population is dependent on the prevalence of undetected AF in the population. • Screening higher risk individuals will improve the PPV and the number needed to screen (NNS) to prevent one stroke. • The NNS to prevent one stroke also depends on the absolute incidence of stroke in the population and the benefit of anticoagulants in that population. • Even in high risk populations using high specificity measures, thepositive predictive value will still be limited resulting in potentially large numbers of false positives. • Continuous monitoring suggests the prevalence of brief episodes of AF may be as high as 30-40% in high risk populations, but the stroke risk and response to anticoagulation is unknown. • Randomized trials are required before adopting systematic AF screening.