Figures
Abstract
Introduction
As atrial fibrillation (AF) is often asymptomatic, it may remain undiagnosed until or even after development of complications, such as stroke. Consequently the observed prevalence of AF may underestimate total disease burden.
Methods
To estimate the prevalence of undiagnosed AF in the United States, we performed a retrospective cohort modeling study in working age (18–64) and elderly (≥65) people using commercial and Medicare administrative claims databases. We identified patients in years 2004–2010 with incident AF following an ischemic stroke. Using a back-calculation methodology, we estimated the prevalence of undiagnosed AF as the ratio of the number of post-stroke AF patients and the CHADS2-specific stroke probability for each patient, adjusting for age and gender composition based on United States census data.
Results
The estimated prevalence of AF (diagnosed and undiagnosed) was 3,873,900 (95%CI: 3,675,200–4,702,600) elderly and 1,457,100 (95%CI: 1,218,500–1,695,800) working age adults, representing 10.0% and 0.92% of the respective populations. Of these, 698,900 were undiagnosed: 535,400 (95%CI: 331,900–804,400) elderly and 163,500 (95%CI: 17,700–400,000) working age adults, representing 1.3% and 0.09% of the respective populations. Among all undiagnosed cases, 77% had a CHADS2 score ≥1, and 56% had CHADS2 score ≥2.
Citation: Turakhia MP, Shafrin J, Bognar K, Trocio J, Abdulsattar Y, Wiederkehr D, et al. (2018) Estimated prevalence of undiagnosed atrial fibrillation in the United States. PLoS ONE 13(4): e0195088. https://doi.org/10.1371/journal.pone.0195088
Editor: Giuseppina Novo, University of Palermo, ITALY
Received: October 10, 2017; Accepted: March 18, 2018; Published: April 12, 2018
Copyright: © 2018 Turakhia et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The authors have posted all underlying program code at the Open Science Framework website located: https://osf.io/ufgdc/. The underlying data used in the analysis come from two sources: Optum and the Centers for Medicare & Medicaid Services. Our data use agreements with Optum and CMS have legally enforceable usability restrictions that preclude our making the data available publicly. However, researchers can access the Medicare Limited Data Sets (the data used in our analysis) from CMS. As stated on the ResDAC website (https://www.resdac.org/cms-data/request/limited-data-sets), “LDS requests do not require a ResDAC review and can be submitted directly to CMS by the researcher." The DUA-Limited Data Sets (LDS) page on the CMS website (https://www.cms.gov/Research-Statistics-Data-and-Systems/Files-for-Order/Data-Disclosures-Data-Agreements/DUA_-_NewLDS.html) describes the process for requesting LDS files. Researchers can contact Optum about accessing the Touchstone data by visiting their website: https://www.optum.com/solutions/prod-nav/claims-data.html.
Funding: The research reported in this manuscript was funded by Bristol-Myers Squibb and Pfizer based on a proposal that co-author Dana P. Goldman and Mintu P. Turakhia presented to the sponsors. Jason Shafrin and Katalin Bognar are employees of; Dana P. Goldman serves as a consultant to Precision Health Economics, who were paid consultants to Pfizer in connection with the development of this manuscript. Jeffrey Trocio, Younos Abdulsattar, and Daniel Wiederkehr are employees and shareholders of Pfizer, Inc. Bristol-Myers Squibb provided funding but was not involved in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; or decision to submit the manuscript for publication. Through Pfizer-affiliated coauthors, Pfizer was involved in the design of the study, the interpretation of data, and the preparation, review, and approval of the manuscript for publication.
Competing interests: I have read the journal's policy and the authors of this manuscript have the following competing interests: Dr. Shafrin had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. In accordance with ICMJE guidelines, all authors have read and approved the final manuscript being submitted. The study was funded by Bristol-Myers Squibb and Pfizer based on a proposal that co-author Dana P. Goldman and Mintu P. Turakhia presented to the sponsors. Jason Shafrin and Katalin Bognar are employees of; Dana P. Goldman serves as a consultant to Precision Health Economics, who were paid consultants to Pfizer in connection with the development of this manuscript. Jeffrey Trocio, Younos Abdulsattar, and Daniel Wiederkehr are employees and shareholders of Pfizer, Inc. Bristol-Myers Squibb provided funding but was not involved in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; or decision to submit the manuscript for publication. Through Pfizer-affiliated coauthors, Pfizer was involved in the design of the study, the interpretation of data, and the preparation, review, and approval of the manuscript for publication. This does not alter our adherence to PLOS ONE policies on sharing data and materials.
Introduction
Atrial fibrillation (AF) is a major cause of ischemic stroke, but often has minimal or no symptoms and therefore can be difficult to diagnose [1]. Current prevalence estimates are based on cohort studies and analyses of health care claims, but these approaches are unable to account for the prevalence of undiagnosed AF and therefore unable to measure true disease prevalence. Estimates of the prevalence of undiagnosed AF have been based on patient screening, and have varied widely from 1–2% of the general population [2–4] to over 15% among patients with a previous stroke [5].
In this paper, we propose an indirect back-calculation method to estimate the prevalence of undiagnosed AF. Back-calculation is a process whereby generally unobservable features of an event (i.e., undiagnosed AF leading to ischemic stroke) can be inferred. The methodology was first developed in the late 1980s for obtaining short-term projections of Acquired Immunodeficiency Syndrome [6]. For the present paper, we applied a non-parametric back-calculation methodology to estimate the prevalence of undiagnosed AF based on measuring the incidence of downstream complications (stroke) of the disease (AF), and then back-calculated total AF prevalence based on the attributable risk of the complication (stroke) to the disease (AF). We used retrospective health insurance claims data to derive AF prevalence estimates in both the working age adult and elderly (Medicare) populations.
Methods
We performed a retrospective cohort study using 2004–2010 health insurance claims data from a commercial claims dataset representing a number of large, self-insured companies and administrative claims data from Medicare. The commercial claims data were used to study AF prevalence in the pre-Medicare, working adult population (aged 18–64 years); Medicare Limited Data Set was used to study the prevalence of AF in older adults (age ≥65 years). Data from commercial and Medicare administrative claims databases were fully anonymized before the authors accessed them. Our sample was restricted to U.S. residents aged ≥18 years with continuous enrollment for at least 12 month before and 15 months after the fourth quarter of 2009.
We estimated the prevalence of undiagnosed AF using a four-step back-calculation methodology (Fig 1) [7]. Step 1 identified patients with diagnosed AF at any point between 2004 and 2009. Patients had an AF diagnosis if they had an International Classification of Diseases, Ninth Revision (ICD-9-CM) code for atrial fibrillation (427.31) on ≥1 hospital inpatient, or on ≥2 hospital outpatient or physician visit claims during this time [8]. To minimize rule-out diagnoses, we did not count medical claims submitted by durable medical equipment providers, home health agencies, laboratories, or non-physician providers such as dentists or physical therapists [9, 10]. Furthermore as our back-calculation uses stroke event rates among patients with nonvalvular AF; we did not consider patients with valvular or transient AF. Patients with valvular AF were identified by having a diagnosis for a heart valve replacement or mitral valve stenosis (ICD-9-CM codes 42.2, 394.x, 396.1, 396.2, 396.8, or 746.5) or a procedure for a valve replacement (CPT codes 33405, 33420, 33422, 33425–33427, 33430, or 33496; ICD-9 procedure codes 35.0x, 35.1x, or 35.2x). Patients with transient AF were identified by an AF diagnosis appearing within 12 months after hyperthyroidism (ICD-9-CM codes 242.x) or within 30 days after coronary artery bypass surgery (ICD-9-CM codes 36.10 or 36.19), pericardial surgery (ICD-9-CM codes 37.10–37.12, 37.24, 37.25, 37.31–33, 37.35, or 37.40) or structural cardiac repair surgery (ICD-9-CM codes 35.31–35.39, 35.41–35.42, 35.50–35.54, 35.60–35.63, or 35.70–36.73) [11].
Step 2 identified patients who: (i) had no AF diagnosis between 2004–2009 while enrolled, (ii) had a stroke in 2010, and (iii) were subsequently diagnosed with non-valvular AF within 3 months after the stroke (“observable” cases of undiagnosed AF). Patients were classified as having an ischemic stroke if there was an inpatient or emergency room claim with a primary ICD-9-CM code of 433.01, 433.11, 433.21, 433.31, 433.81, 433.91; 434 or 436. This approach to identify patients with ischemic stroke has been previously validated with a sensitivity of 74%, specificity of 95%, and a positive predictive value of 88% [12]. As administrative coding of ischemic stroke has been shown to have limited sensitivity, we performed a sensitivity analysis in which we defined stroke to include ischemic stroke, hemorrhagic stroke (ICD-9-CM codes 430.x, 431.x, 432.0–432.9) and transient ischemic attack (TIA) (ICD-9-CM codes 435.x). Patients with post stroke non-valvular AF were identified as above using the appropriate timeframe. Because AF may cause stroke, but stroke is generally not considered to cause AF, we assumed that these individuals with AF coded soon after stroke are “observed” cases of undiagnosed AF. To account for the fact that some AF cases may appear spontaneously and not be the direct cause of a stroke, we adjusted the number of cases with stroke followed by AF by the rate of quarterly AF incidence among patients with no prior AF diagnosis.
Step 3 assigned stroke risk to patients with AF based on CHADS2 scores. We measured each patient’s CHADS2 score based on whether the patient had at least one claim for congestive heart failure (ICD-9 codes 428.x), hypertension (ICD-9-CM codes 401.x, 402.x, 403.x, 404.x, 405.x), diabetes (ICD-9-CM codes 250.x), or a previous stroke (ICD-9-CM codes 433.x1, 434.x1, 435.x). Furthermore, we used CHADS2 rather than CHA2DS2-VASc scores as there were very few people in the working-age population with low CHA2DS2-VASc scores who experienced AF after a stroke, leading to highly unstable estimates. Then we approximated stroke risks for patient with stroke and followed by an AF diagnosis using all-cause stroke risk by CHADS2 score among patients with diagnosed AF but not on anticoagulation from the literature [13]. As a sensitivity analysis, we used stroke incidence rates derived from our data (both commercial and Medicare) of patients with diagnosed AF, which include patients who are treated, untreated, and variably treated with anticoagulation.
Step 4 divided the number of observable cases (by CHADS2 score) of undiagnosed AF from Step 2 by the all-cause CHADS2-specific stroke probabilities determined in Step 3 to estimate the total number of undiagnosed AF cases.
To estimate the total prevalence of undiagnosed AF in the U.S. population, we adjusted our results according to the age/gender composition of the population [14]. As our measure of the prevalence of undiagnosed AF is a ratio of “observable” undiagnosed AF cases after stroke and the probability of stroke conditional on an AF diagnosis, confidence intervals were constructed using a bootstrapping method, a method well-suited for constructing confidence intervals for ratios [15].
We conducted three sensitivity analyses. First, we measured stroke incidence rates among AF patients from our claims data rather than the literature. Second, we replicated the analysis but defined a stroke more broadly to include ischemic stroke, transient ischemic attack, and hemorrhagic stroke. Third, we measured stroke incidence stratified by CHA2DS2-VASc score rather than CHADS2 score, based on stroke risk as measured in previous studies [16].
Results
Of the initial sample of 2.4 million Medicare beneficiaries and 2.2 million commercially insured individuals in our database as of 2009, we identified 1.1 million Medicare and 400,000 privately insured beneficiaries who met the inclusion criteria (Fig 2).
Patients with observed (diagnosed) AF were older, more likely to be male, and had higher CHADS2 scores compared to patients without AF (Table 1). Among elderly patients who had a stroke, patients diagnosed with AF prior to the stroke were 1.8 years older (p<0.001) with a CHADS2 score that was 0.7 points (p<0.001) higher than patients who were first diagnosed with AF within 3 months after the stroke (Table 2).
For working adults aged 18–64, the prevalence of diagnosed AF in our data was 0.83% (95% CI: 0.79%-0.86%), whereas 8.70% (95% CI: 8.62%-8.78%) of elderly adults had diagnosed AF (Table 3). The latter figure was comparable to a 2007 estimate of AF prevalence among Medicare beneficiaries (8.58%) calculated in Piccini et al [17]. Extrapolating the general U.S. population, we estimated that 4.63 million individuals had diagnosed AF, of which 1,293,600 (95% CI: 1,238,300–1,347,100) were working age adults and 3,338,500 (95% CI: 3,305,000–3,372,200) were elderly.
Applying the back-calculation methodology, we estimated that 163,500 (95% CI: 17,700–400,000) working age adults had undiagnosed AF. Although the prevalence of undiagnosed AF among working age adults was rare overall (0.09% prevalence; 95% CI: 0.01%-0.22%), 11.2% of all persons with AF in the working age population were undiagnosed. We estimated that 1.32% (95% CI, 0.86%-1.94%) or 535,400 (95% CI: 331,900–804,400) elderly patients had undiagnosed AF. Among the elderly, 13.8% of persons with AF were undiagnosed.
The total prevalence (both diagnosed and undiagnosed) of AF in the U.S. was estimated to be 5.33 million (2.4% of adults) comprised of 1,457,100 (95% CI: 1,218,500–1,695,800) working age adults and 3,873,900 (95% CI: 3,675,200–4,702,600) patients aged ≥65 years representing 0.92% and 10.0% of these populations, respectively. Of the total 5.33 million patients with AF, 698,900 (13.1%) were undiagnosed. Although males were more likely to have AF, a larger proportion of female patients with AF were undiagnosed. Similar to patients with diagnosed AF, undiagnosed AF increased with age. Overall, 56% of undiagnosed AF cases had a CHADS2 score ≥2, and 77% had CHADS2 ≥1 (Fig 3).
When we applied CHADS2-specific stroke incidence rates derived from our datasets of patients with diagnosed AF, our estimated prevalence of undiagnosed AF increased. Under this alternative assumption, the estimated prevalence of undiagnosed AF was 0.85 million (0.47%; 95% CI: 0.04%-1.15%) among working age individuals and 2.0 million (5.06%; 95% CI: 4.08%-6.10%) among the elderly, a total of 2.85 million individuals. When we defined stroke in the claims data using ischemic stroke, hemorrhagic stroke, or TIA, the prevalence of undiagnosed AF was 0.20% (95% CI: 0.02%-0.48%) for working age adults the 2.88% (95% CI: 2.55%-3.32%) for elderly adults. In our third sensitivity analysis using CHA2DS2-VASc scores to predict stroke risk rather than CHADS2 scores, we found that the results for our commercial population were unstable due to the small sample size of patients with stroke and the increased number of potential scores (i.e., 0 to 9 for CHA2DS2-VASc rather than 0 to 6 for CHADS2). Among the elderly individuals, however, we estimated that 1.22 million (3.08%; 95% CI: 2.11% to 4.15%) of people aged 65 years and older had undiagnosed atrial fibrillation.
Discussion
The estimated prevalence of undiagnosed AF in the U.S. in 2009 was 700,000 while the total prevalence of AF was 5.3 million. One in eight patients with AF was undiagnosed. Furthermore, over half of the population with undiagnosed AF was at moderate to high risk of stroke.
Our total prevalence estimate of 5.3 million (2.4% of all adults) for 2009 is markedly higher than prior projections of 2.2–2.7 million (1.0–1.2%) in 2009–2010 from the American Heart Association (AHA) Statistical Update [18] and estimates of 3.0 million (1.6%) by Naccarelli and colleagues [19]. The AHA projections are derived from cross-sectional studies of 1996–1997 Kaiser Permanente Northern California enrollees [11] while Naccarelli and colleagues used a claims-based method, neither accounting for undiagnosed AF. However, our baseline projections are aligned with a model-based projection (2.5%) derived from a 1990–2000 community sample of Olmsted County, Minnesota residents [20].
Prior estimates of undiagnosed AF prevalence come primarily from AF screening studies, reporting 1.7% in a general European population [3] and 2.3–15.1% among patients with diabetes, hypertension, or previous stroke or TIA [5, 21].
These data show appreciable rates of undiagnosed AF such that targeted screening strategies could prove beneficial, particularly in subgroups of older (age ≥65) patients or in patients with multiple CHADS2 risk factors, in whom prevalence is substantially higher. Our results indicated that the prevalence of undiagnosed AF increases from 0.11% of the population of patients with a CHADS2 score of zero to 0.73% of the population of patients with a CHADS2 score of 2 and up to 0.95%-1.39% of the population of patients with a CHADS2 score between 3 and 6. At the same time, our data indicated that routine full population screening may be of low yield, as the younger working age population had an undiagnosed AF prevalence of 0.09%.
There are several limitations to our analysis. First, our back-calculation methodology assumed that all AF detected in the three months after ischemic stroke was present prior to stroke. We believe this to be reasonable and biologically plausible, and our estimates showed good calibration to known AF prevalence estimates. To address the case where AF appears spontaneously after stroke, we did adjust our prevalence estimates based on the baseline AF incidence in our data. Second, our model may have underestimated the prevalence of undiagnosed AF since a number of people with AF have silent strokes that would not be captured by claims data, have short episodes of AF that may not meaningfully increase stroke risk, or may have cryptogenic strokes that were in fact due to undiagnosed AF. Third, estimates were modeled from data sources that may not be truly representative of the U.S. population (e.g. Medicaid, uninsured/unemployed, Medicare Advantage patients, broader employee base are not well represented). Fourth, our model did not account for differences in stroke event rates by race or ethnicity and we assumed that stroke incidence rates among patients with undiagnosed AF were similar to those of patients with diagnosed AF. Fifth, data and funding available to the authors at the time of research was only through 2010. Finally, the estimates presented rely on the accuracy of the back-calculation model; population AF screening studies will be required to better estimate point prevalence of undiagnosed AF.
Conclusion
Based on a model of back-calculating disease prevalence from ischemic stroke events, we estimate that the total AF prevalence in 2009 was approximately 5.3 million persons, which is substantially higher than prior estimates based on cohort and population studies. Of these patients with AF, about 700,000 persons were undiagnosed—about one in eight patients—corresponding to 0.31% of the total US adult population.
Acknowledgments
Trieu Lai, BS and Steven Sison, MS of Precision Health Economics provided programming support, and Laura Henkhaus, BS and Avrita Campinha-Bacote, MPH of Precision Health Economics provided research support at the time the study was conducted.
References
- 1. Samol A, Masin M, Gellner R, Otte B, Pavenstadt HJ, Ringelstein EB, et al. Prevalence of unknown atrial fibrillation in patients with risk factors. Europace: European pacing, arrhythmias, and cardiac electrophysiology: journal of the working groups on cardiac pacing, arrhythmias, and cardiac cellular electrophysiology of the European Society of Cardiology. 2013;15(5):657–62. Epub 2012/12/22. pmid:23258819.
- 2. Clua-Espuny JL, Lechuga-Duran I, Bosch-Princep R, Roso-Llorach A, Panisello-Tafalla A, Lucas-Noll J, et al. Prevalence of Undiagnosed Atrial Fibrillation and of That Not Being Treated With Anticoagulant Drugs: the AFABE Study. Revista Española de Cardiología (English Edition). 2013;66(7):545–52.
- 3. Marfella R, Sasso FC, Siniscalchi M, Cirillo M, Paolisso P, Sardu C, et al. Brief episodes of silent atrial fibrillation predict clinical vascular brain disease in type 2 diabetic patients. Journal of the American College of Cardiology. 2013;62(6):525–30. pmid:23684685
- 4. Claes N, Van Laethem C, Goethals M, Goethals P, Mairesse G, Schwagten B, et al. Prevalence of atrial fibrillation in adults participating in a large-scale voluntary screening programme in Belgium. Acta cardiologica. 2012;67(3):273–8. Epub 2012/08/09. pmid:22870733.
- 5. Suissa L, Lachaud S, Mahagne MH. Optimal timing and duration of continuous electrocardiographic monitoring for detecting atrial fibrillation in stroke patients. Journal of stroke and cerebrovascular diseases: the official journal of National Stroke Association. 2013;22(7):991–5. Epub 2012/02/22. pmid:22349706.
- 6. Egan JR, Hall IM. A review of back-calculation techniques and their potential to inform mitigation strategies with application to non-transmissible acute infectious diseases. Journal of The Royal Society Interface. 2015;12(106):20150096.
- 7. Brookmeyer R, Gail MH. A method for obtaining short-term projections and lower bounds on the size of the AIDS epidemic. Journal of the American Statistical Association 1988;83(402):301–8.
- 8. Pope GC, Kautter J, Ingber MJ, Freeman S, Sekar R, Newhart C. Evaluation of the CMS-HCC Risk Adjustment Model. RTI International and the Centers for Medicare & Medicaid Services. 2011.
- 9. Piccini JP, Hammill BG, Sinner MF, Jensen PN, Hernandez AF, Heckbert SR, et al. Incidence and prevalence of atrial fibrillation and associated mortality among Medicare beneficiaries, 1993–2007. Circulation Cardiovascular quality and outcomes. 2012;5(1):85–93. Epub 2012/01/12. pmid:22235070; PubMed Central PMCID: PMCPMC3332107.
- 10. Turakhia MP, Solomon MD, Jhaveri M, Davis P, Eber MR, Conrad R, et al. Burden, timing, and relationship of cardiovascular hospitalization to mortality among Medicare beneficiaries with newly diagnosed atrial fibrillation. Am Heart J. 2013;166(3):573–80. pmid:24016509
- 11. Go AS, Hylek EM, Phillips KA, Chang Y, Henault LE, Selby JV, et al. Prevalence of diagnosed atrial fibrillation in adults: national implications for rhythm management and stroke prevention: the AnTicoagulation and Risk Factors in Atrial Fibrillation (ATRIA) Study. Journal of the American Medical Association 2001;285(18):2370–5. pmid:11343485
- 12. Tirschwell DL, Longstreth W. Validating administrative data in stroke research. Stroke. 2002;33(10):2465–70. pmid:12364739
- 13. Gage BF, Waterman AD, Shannon W, Boechler M, Rich MW, Radford MJ. Validation of clinical classification schemes for predicting stroke: results from the National Registry of Atrial Fibrillation. Journal of the American Medical Association 2001;285(22):2864–70. pmid:11401607.
- 14.
U.S. Census. Age and sex composition in the United States: 2012 [Webpage]. U.S. Census; 2012 [cited 2014 March 13]. Available from: http://www.census.gov/population/age/data/2012comp.html.
- 15. DiCiccio TJ, Efron B. Bootstrap confidence intervals. Statistical science. 1996:189–212.
- 16. Lip GY, Frison L, Halperin JL, Lane DA. Identifying patients at high risk for stroke despite anticoagulation. Stroke. 2010;41(12):2731–8. pmid:20966417
- 17. Piccini JP, Hammill BG, Sinner MF, Jensen PN, Hernandez AF, Heckbert SR, et al. Incidence and prevalence of atrial fibrillation and associated mortality among Medicare beneficiaries: 1993–2007. Circulation: Cardiovascular Quality and Outcomes. 2012;5(1):85–93.
- 18. Lloyd-Jones D, Adams RJ, Brown TM, Carnethon M, Dai S, De Simone G, et al. Heart disease and stroke statistics—2010 update A report from the American Heart Association. Circulation. 2010;121(7):e46–e215. pmid:20019324
- 19. Naccarelli GV, Varker H, Lin J, Schulman KL. Increasing prevalence of atrial fibrillation and flutter in the United States. American Journal of Cardiology. 2009;104(11):1534–9. Epub 2009/11/26. pmid:19932788.
- 20. Miyasaka Y, Barnes ME, Gersh BJ, Cha SS, Bailey KR, Abhayaratna WP, et al. Secular trends in incidence of atrial fibrillation in Olmsted County, Minnesota, 1980 to 2000, and implications on the projections for future prevalence. Circulation. 2006;114(2):119–25. Epub 2006/07/05. pmid:16818816.
- 21. Weber-Krüger M, Gröschel K, Mende M, Seegers J, Lahno R, Haase B, et al. Excessive supraventricular ectopic activity is indicative of paroxysmal atrial fibrillation in patients with cerebral ischemia. PloS one. 2013;8(6):e67602. pmid:23840747