Risk Factors for Symptomatic Atrial Fibrillation-Analysis of an Outpatient Database

Annabelle Santos Volgman 1, Patrick Dunn 2, Allison Sundberg 2, Scott Conard3, Pavitra Chakravarty3, Zin Htway4, Albert Waldo5, Christine Albert6, Mintu P. Turakhia7, Gerald V. Naccarelli 8

1 Rush University Medical Center .2 American Heart Association .3 Converging Health .4 Walden University .5 Case Western Reserve University, University Hospitals Cleveland Medical Center .6 Brigham and Women’s .7 VA Palo Alto Health Care System; Stanford University .8 Penn State University College of Medicine .

Abstract

Atrial fibrillation (AF) is the most common sustained arrhythmia encountered in practice and is the leading cause of debilitating strokes with significant economic burden. It is currently not known whether asymptomatic undiagnosed AF should be treated if detected by various screening methods. Currently, United States guidelines have no recommendations for identifying patients with asymptomatic undiagnosed AF due to lack of evidence. The American Heart Association Center for Health Technology & Innovation undertook a plan to identify tools in 3 phases that may be useful in improving outcomes in patients with undiagnosed AF. In phase I we sought to identify AF risk factors that can be used to develop a risk score to identify high-risk patients using a large commercial insurance dataset. The principal findings of this study show that individuals at high risk for AF are those with advanced age, the presence of heart failure, coronary artery disease, hypertension, metabolic disorders, and hyperlipidemia. Our analysis also found that chronic respiratory failure was a significant risk factor for those over 65 years of age and chronic kidney disease for those less than 65 years of age.

Key Words : Atrial fibrillation, Left atrial appendage, Stroke.

Correspondence to: Annabelle Santos Volgman, MD FACC FAHA McMullan-Eybel Endowed Chair for Excellence in Clinical Cardiology Professor of Medicine, Rush College of Medicine Medical Director, Rush Heart Center for Women Rush University Medical Center

Introduction

Atrial fibrillation (AF) is the most common sustained arrhythmia encountered in practice and is the leading cause of debilitating strokes[1], leading to significant economic burden[2]. While AF presently affects 2.7-6.1 million, it is expected to increase to 12.1 million by 2030.[1] AF may be asymptomatic and underdiagnosed with the first presentation being a stroke.[3] Studies estimate that 13% to 40% of patients with AF are undiagnosed.[4,5] Strokes associated with AF have worse outcomes resulting in larger cerebral infarct size, more hemorrhagic transformation, subsequent disabilities, and death.[6] Increasing awareness of AF by clinicians and patients may lead to an earlier diagnosis and treatment, resulting in fewer adverse health outcomes. However, the United States Preventive Task Force has stated there is insufficient evidence to endorse electrocardiographic (ECG) screening for AF. [7]

The American Heart Association’s Center for Health Technology & Innovation undertook a plan to identify tools that may be useful in improving outcomes in patients with undiagnosed AF. The work plan consists of 3 phases. Phase I is to develop a predictive screening tool, using multivariate logistic regression to calculate the risk of developing AF. Phase II is to create and evaluate the use of the screening tool developed in Phase I to prospectively identify individuals at high risk for the onset of AF, as compared to usual care. Phase III will ask patients with newly diagnosed with AF to enroll in a study that would compare compliance with AF treatment in patients using a digital tracking device to usual care. The results of our analysis reported here are from phase I of the study. The aim of this analysis was to address the problem of undiagnosed AF in the US population by gaining a better understanding of the factors associated with AF.

We hypothesized that using a large population database could potentially identify clinically important risk factors associated with AF. The primary objective of this phase of the study was to identify patients at high risk for undiagnosed AF.

Methods

We performed a retrospective cohort study using a commercial dataset to identify risk factors that are associated with AF ICD diagnosis codes of 427.31.

Data source

A commercial dataset representing over 50 health plans and self-insured employers, representing all 50 states and containing 535,499 records, including 4862 cases of AF from 2010-17 were used in this analysis. The dataset was cross-sectional, and all records were de-identified. The dataset included demographic data, including age and gender, frequency of 40 chronic conditions as identified by ICD codes, biometric measures, including height, weight, and blood pressure, and cost data, including pharmacy and total paid claims.

Calculations

To assess the risk of stroke in this patient population the CHADS2 and CHA2DS2-VASc scores were calculated,[8] as well as the number of chronic conditions.

Statistical analysis

Statistical analysis was performed with SPSS version 25. Chi Square, binary logistic regression and hierarchical logistic regression were conducted to test the hypothesis that AF could be predicted from demographic, biometric, and claims data. Chronic conditions, including AF were coded as binary variables, Yes=1 and No=0. Frequencies, prevalence of AF and odds ratios were calculated from the binary variables. Frequencies of AF were calculated by CHADS2 and CHA2DS2-VASc scores. Independent sample t-tests were calculated for height, weight, body mass index, systolic blood pressure, diastolic blood pressure, number of chronic conditions, age, total pharmacy cost, and lifetime paid claims, comparing cases with AF and cases without AF. Binary and hierarchical logistic regression was performed with AF as the dependent variable.

Results

Descriptive statistics

[Table 1] provides a breakdown of cases with and without AF by gender and age. The total number and rate of AF was higher in males than in females. The rate of AF increased with age. [Figure 1] shows the age distribution of the patient population in the database. [Table 2] shows the distribution of the presence of AF with respect to sex and age.

There was a slight majority of male participants (n = 286,710; 53.4%) to female participants (n = 248,101; 46.2%). For the male participants, (n = 3,255; 1.14%) were coded YES for AF and (n = 283,455; 98.86%) were coded NO for AF. For the female participants, (n = 1,603; 0.65%) were coded YES for AF and (n = 246,496; 99.35%) were coded NO for AF. Overall for both genders, (n = 4,858; 0.91%) were coded YES for AF and (n = 529,963; 99.09%) were coded NO for AF.

Table 1. Frequency of AF by age and gender
Group + AF -AF Rate /1000
Male 3255 283455 .011
Female 1603 246498 .003
Age: 18-64 2488 466468 .005
Age: 65-74 1302 31724 .039
Age: ≥75 1403 11092 .112



Figure 1. Age distribution of the patient population in the database



Table 2. Distribution of age and sex in patients with AF.
<65 years ≥65 years
Males 0.66% 7.03%
Females 0.28% 4.75%



Most participants were under the age of 65 years (n = 490,566; 91.6%). A clear minority of participants were age 65 years or older (n = 44,883; 8.4%). For the under age 65-year participants, (n = 2,353; 0.48%) were coded YES for AF and (n = 488,213; 99.5%) were coded NO for AF. For the age 65 years or older, (n = 2,509; 5.59%) were coded YES for AF and (n = 42,374; 94.41%) were coded NO for AF. Overall for both age groups, (n = 4,862; 0.91%) were coded YES for AF and (n = 530,587; 99.09%) were coded NO for AF.

Associated chronic conditions

The dataset included a total of 40 of the most common chronic conditions to look for possible trends. Metabolic disorders included unspecified metabolic conditions not including metabolic syndrome or diabetes. [Table 3] is a breakdown of each condition, including a 2X2 table of the presence or absence of AF and each chronic condition. From the table the rate of AF and each factor can be calculated, as well as the odds ratio. A Chi Square analysis was performed for each chronic condition to determine of there was a statistically significant relationship between AF and each chronic condition. The top 10 chronic conditions were used in the logistic regression model. Since age of a robust predictor of AF, and since this commercial dataset is heavily weighted to younger individuals, the rates, odds ratio and Chi Square were conducted for cases under age 65 and over age 65.

Table 3. Calculation of odds ratio and Chi Square for common chronic conditions for the total group.
Factor +AF, + Factor +AF, -Factor -AF, + Factor -AF, -Factor Odds Ratio Chi Square P value
ADHD 26 4837 7148 524546 .394 .394 <.05
Affective psychosis 33 4830 3073 528621 1.2 .848 NS
Alzheimer’s 62 4801 236 531458 29.1 1314 <.05
Asthma 343 4520 14726 516968 2.6 323 <.05
Autism 0 4863 228 531466 .99 2.086 NS
Blood disorders 1428 3435 21495 510199 9.8 7555 <.05
Bronchopulm. dysplasia 0 4863 3 531691 .99 .027 NS
CAD 1652 3211 9675 522019 27.7 24105 <.05
CKD 638 4225 4064 537630 19.6 8468 <.05
COPD 690 4173 4547 527147 19.1 8864 <.05
Cancer 997 3866 16434 515260 8.0 4647 <.05
Cerebral palsy 1 4862 139 531555 .787 .058 NS
Chromosomal abnorm. 4 4859 161 531533 2.7 4.3 <.05
Chronic pain 317 4546 7089 524605 5.1 951 <.05
Chronic resp fail 227 4636 638 531056 40.7 6192 <.05
Heart failure 1354 3509 2902 528792 70.3 45630 <.05
Demyelinating diseases 16 4847 953 530741 1.8 5.9 <.05
Depression 257 4606 14358 517336 2.0 121 <.05
Developmental disorders 5 4858 308 531386 1.7 1.6 NS
Diabetes 1370 3493 28755 502939 6.8 4712 <.05
ESRD 154 4709 682 531012 25.4 2859 <.05
Eating disorders 39 4824 745 530949 5.7 144 <.05
HIV/AIDS 8 4855 341 531353 2.5 7.4 <.05
Hyperlipidemia 2717 2146 61692 470002 9.6 8939 <.05
Hypertension 3628 1235 70102 461592 19.3 15336 <.05
Immune disorders 48 4815 1006 530668 5.2 156 <.05
Inflammatory bowel dis 54 4809 2171 529523 2.7 57 <.05
Intellectual disabilities 2 4861 77 531617 2.8 2.3 NS
Liver diseases 289 4574 6625 525069 5.0 835 <.05
Lower back pain 825 4038 37016 494678 2.7 735 <.05
Metabolic disorders 3305 1558 76023 455671 12.7 11015 <.05
Metabolic syndrome 49 4814 1586 530108 3.4 79.8 <.05
Morbid obesity 420 4443 8780 522914 5.6 1395 <.05
Osteoarthritis 1072 3791 18686 513008 7.7 4664 <.05
Paralysis 82 4781 690 531004 13.1 812 <.05
Peripheral vascular dis 341 4522 2183 529511 18.2 4485 <.05
Rheumatoid Arthritis 98 4765 2480 529214 4.3 241.7 <.05
Sickle cell disease 1 4862 88 531606 1.2 .047 NS
Sleep apnea 1389 3474 23540 508154 8.6 6336 <.05
Stroke 1303 3850 9511 499791 17.8 13595 <.05

+AF indicates number of positive cases of atrial fibrillation, -AF indicates number of negative cases of atrial fibrillation, +Factor means the number of positive cases of condition listed on the left, - Factor means the number of negative cases of the condition listed on the left. ADHD = attention deficit and hyperactive disorder, CAD=coronary artery disease, CKD=chronic kidney disease, COPD=chronic obstructive pulmonary disease, ESRD=end stage renal disease.

CHADS2 and CHA2DS2-VASc and AF. The CHADS2 and CHA2DS2-VASc scores were calculated from the data. [Table 4] is a summary of the rate of AF by CHADS2 scores, and [Table 5] is the summary of the rate of AF by CHA2DS2-VASc scores. In both scores the rate of AF increases with each level.

Table 4. CHADS2 score distribution of the patient population
CHADS2 Total % AF % Rate/1000
0 415,583 80.8 874 0% 2.10
1 66,589 12.9 1,362 2% 20.45
2 22,089 4.3 1,078 5% 48.80
3 5,671 1.1 751 13% 132.43
4 3,136 .6 544 17% 173.47
5 1,110 .2 398 36% 358.56
6 268 .1 146 54% 544.78

CHADS2=CHADS2 score, Total: Number of cases in CHADS2 score, %: percentage of cases in CHADS2 score, AF=number of cases of atrial fibrillation, %=percentage of atrial fibrillation cases

Table 5. CHA2DS2-VASc score distribution of the patient population
CHA2DS2-VASc Score Total % AF % Rate/1000
0 193,811 37.7 490 0% 2.53
1 241,617 47.0 964 0% 3.99
2 48,182 9.4 965 2% 20.03
3 19,012 3.7 838 4% 44.08
4 6,461 1.3 613 9% 94.88
5 2,902 .6 524 18% 180.57
6 1,511 .3 370 24% 244.87
7 663 .1 259 39% 390.65
8 237 .0 101 43% 426.16
9 39 .0 27 69% 692.31

Total: Number of cases in CHA2DS2VASc score, %: percentage of cases in CHA2DS2VASc score, AF=number of cases of atrial fibrillation, %=percentage of atrial fibrillation cases

Comparison of means

The dataset included a subset of continuous variables including body composition, blood pressure, number of chronic conditions and cost variables. [Table 6] shows the results of independent samples t-test comparing cases with AF to those without AF. While there was no difference in height or blood pressure, there were statistically significant differences in weight and body mass index. Individuals with AF were significantly older, had more chronic conditions and had higher medical and pharmacy costs than those without AF.

Table 6. Comparison of means for continuous variables
Factor AF N Mean SD P value
Height Yes 526 65.53 15.8 .105
No 47285 64.48 14.9
Weight Yes 526 211.07 75.1 .000
No 47285 182.37 62.0
Body mass index Yes 526 25.4 15.3 .000
No 47285 22.0 13.8
Systolic blood pressure Yes 526 124.6 17.8 .896
No 47285 122.1 34.1
Diastolic blood pressure Yes 526 77.4 11.3 .051
No 47285 76.2 13.1
# of chronic conditions Yes 4863 5.64 2.9 .000
No 531694 .8 1.5
Age Yes 4862 66.1 14.6 .000
No 530587 42.0 15.4
Total Rx cost Yes 4852 9800.9 32100 .000
No 417586 2578.9 15861
Lifetime paid claims Yes 4852 163592 70031 .000
No 417586 46933 12007



Logistic Regression

Since the chronic conditions are binary factors (yes or no) simple binary logistic regression was used to determine the relationship between AF and each chronic condition [Table 7]. The highest Nagelkerke r squared is for the number of chronic conditions. Age, hypertension (HTN), coronary artery disease (CAD), congestive heart failure (CHF) and metabolic disorders had the highest Nagelkerke scores. Binary logistic regression was also performed on the under age 65 and over age 65 cohorts.

Table 7. Binary Logistic Regression for total group and common chronic conditions
Factor Wald Nagelkerke OR LL UL P value
ADHD .22.298 .001 .394 .268 .580 .000
Affective psychosis .846 .000 1.175 .833 1.658 .846
Alzheimer’s 551.969 .005 29.082 21.954 38.527 .000
Asthma 299.409 .004 2.644 2.384 2.977 .000
Blood disorders 5036.616 .066 9.867 9.263 10.511 .000
CAD 10807.959 .130 27.759 26.073 29.544 .000
CKD 4315.374 .046 19.605 17.940 21.425 .000
COPD 4564.768 .049 19.169 17.595 20.884 .000
Cancer 3298.310 .042 8.086 7.529 8.683 .000
Chromosomal abnorm. 3.898 .000 2.718 1.007 7.332 .048
Chronic pain 765.563 .010 .010 4.594 5.796 .000
Chronic respiratory failure 2220.662 .022 40.757 34.933 47.552 .000
Heart failure 13202.329 .148 70.311 65.391 75.601 .000
Demyelinating diseases 5.815 .000 1.838 1.121 3.016 .016
Depression 116.379 .002 2.010 1.771 20282 .000
Developmental disorders 1.621 .000 1.176 .734 4.298 .203
Diabetes 3521.769 .048 6.860 6.437 7.310 .000
ESRD 1282.070 .013 25.463 21.328 30.400 .000
Eating disorders 112.783 .001 5.762 4.171 7.960 .000
HIV/AIDS 6.939 .000 2.568 1.273 5.179 .008
Hyperlipidemia 6026.756 .103 9.646 9.109 10.214 .000
Hypertension 7964.790 .177 19.343 18.125 20.643 .000
Immune disorders 125.030 .002 5.259 3.931 7.035 .000
Inflammatory bowel disorder 52.901 .001 2.739 1.572 3.593 .000
Intellectual disabilities 2.124 .000 2.841 .698 11.566 .145
Liver diseases 677.300 .009 5.008 4.435 5.654 .000
Lower back pain 677.651 .010 .010 2.531 2.945 .000
Metabolic disorders 6736.697 .135 12.715 11.966 13.511 .000
Metabolic syndrome 70.554 .001 3.402 2.557 4.527 .000
Morbid obesity 1097.189 .014 5.630 5.083 6.236 .000
Osteoarthritis 3354.483 .043 7.763 7.243 8.321 .000
Paralysis 480.473 .005 13.199 10.480 16.624 .000
Peripheral vascular disease 2337.602 .024 18.291 16.258 20.579 .000
Rheumatoid Arthritis 202.197 .003 4.389 3.579 5.381 .000
Sleep apnea 4414.928 .059 8.0631 8.099 9.198 .000
Gender 352.507 .007 .565 .532 .600 .000
Age 9395.060 .192 1.091 1.089 1.093 .000
Age group 7409.449 .118 12.285 11.603 13.007 .000
Chronic conditions 16330.588 .319 1.856 1.839 1.874 .000
CHADS2 15633.265 .234 3.301 2.959 3.063 .000
CHA2DS2-VASc 13933.352 .211 2.358 2.325 2.392 .000

OR=Odds ratio, LL=lower limit of the odds ratio, UL=upper limit of the odds ratio, ADHD = attention deficit and hyperactive disorder, CAD=coronary artery disease, CKD=chronic kidney disease, COPD=chronic obstructive pulmonary disease, ESRD=end stage renal disease.

A hierarchical logistic regression model was performed for the whole group [Table 8] and separately for the under 65 and over 65 cohorts. Variables selected for the logistic regression model were based on the variables with the top 10 odds ratios. Since age resulted in a higher Nagelkerke score than any of the chronic conditions it was added to the model.

Table 8. Hierarchical logistic regression total group
Factor Wald Nagelkerke Odds Ratio 95% CI ROC
Heart failure 13202.329 0.148 70.31 65.40-75.60 0.636
CAD 10807.959 0.13 27.76 26.07-29.54 0.661
Hypertension 7964.79 0.177 19.34 18.13-20.64 0.807
Metabolic disorders 6736.697 0.135 12.72 11.97-13.51 0.768
Hyperlipidemia 6026.756 0.103 9.65 9.109-10.21 0.721
Age 9395.06 0.192 1.10 1.089-1.093 0.798

95%CI = 95% confidence interval for the odds ratio; ROC=receiver operating charasteristic

CAD, HTN, CHF, chronic respiratory failure and age were common to all 3 models. Only CKD in the under 65 cohort and COPD in the over 65 cohort were added. Since age is in the model and the highest ROC and Nagelkerke r squared was achieved in the total group there appears to be no reason to have a separate predictive model for each age group.

Discussion

Our principal findings show that individuals at high risk for AF are those with advance age, the presence of CHF, CAD, HTN, metabolic disorders, and hyperlipidemia. Our analysis also found that chronic respiratory failure was a significant risk factor for those over 65 years of age and chronic kidney disease for those less than 65 years of age.

Risk scores for predicting AF have been developed by the Framingham Heart Study and Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE)-AF consortium. The risk score was validated in different ethnic groups including whites, Hispanics and African Americans.[9] In a community-based cohort, the CHARGE-AF risk score was compared to the CHA2DS2-VASc risk score and found to perform better at predicting AF.[10] In another AF risk score study COPD was found to be a significant risk factor in a Chinese population study. [11]

Awareness of this high-risk group can be a signal to primary care physicians to pay more attention to the possibility that these patients are at greater risk for AF and, in turn, for stroke. This predictive tool has the potential, following further validation, to assess large amounts of patient claims and electronic medical record data to identify patients in need for AF detection devices. An additional possible use of this data would be in risk stratifying corporate employees for proactive encouragement to establish care with a primary care physician and have frequent follow-up.

Several studies are underway to determine if opportunistic AF detection leads to decreased strokes, heart failure and mortality.[12] Screening tools such as pulse palpation followed by ECG, sphygmomanometer with rhythm determinations and rhythm monitoring devices are being studied to identify patients with asymptomatic AF.[12] However, it has yet to be determined whether asymptomatic AF detected through opportunistic means such as implanted devices or screening studies should prompt the same treatment for symptomatic AF.[12] The AF Screen International Collaboration acknowledged that health resources vary widely between countries and health systems and thus AF screening should be both country and health system specific. Large randomized outcomes studies are needed to strengthen the evidence base of the value of detecting asymptomatic AF. Guidelines vary in their recommendations for opportunistic screening for AF. The European Society of Cardiology AF guidelines has a level IB recommendation for opportunistic screening for AF by pulse taking or ECG rhythm strip in patients >65 years of age as well as routine detections of atrial high rate episodes in patients with implanted devices. Further evaluation for treatment of AF is then recommended.[13] The American guidelines make no recommendations for opportunistic AF screening but consider it a priority for stroke prevention. Efforts are underway to provide randomized controlled trials to determine the value of opportunistic AF screening.

For the next phase of our study, we plan to develop and use the predictive screening tool using the high-risk conditions to see if this would predict the presence of asymptomatic AF in patients in different populations and other databases. In addition, we will need to evaluate the risk of stroke and thromboembolism in patients with asymptomatic subclinical AF since this is not yet known. The threshold amount of AF that should be treated with anticoagulation in these patients is not yet universally established.

Limitations

There are several limitations in our study. The commercial dataset is derived from specific codes of the diseases entered by healthcare providers and errors from incorrect coding from misclassification of diseases may be possible. Another limitation was that the accuracy of the diagnosis of AF was not verified by cardiologists. We assume the patients in the dataset were symptomatic, as asymptomatic diagnosis is unlikely with the exception of occasional incidental diagnosis in a routine visit with an observation such as an irregular pulse. Our study strength is the large cohort of a diverse population in terms of sex and age. Because the database consists only of patients known to have been diagnosed with AF, the findings of AF markers in this report may or may not hold for undiagnosed AF. Other considerations such as treatment for hypertension or slower ventricular rate that may reduce symptoms must be taken into account, and thus further validation is needed.

Disclosures

Dr. Waldo has received consulting fees/honoraria from Biosense Webster, AtriCure, Milestone Pharmaceuticals, Cardiac Insight, Correvio Pharms, Pfizer, Bristol-Myers Squibb; Dr. Naccarelli has received consulting fees/honoraria from Acesion, Glaxo-Smith-Kline, Janssen, Milestone, Omecos, and Sanofi; Dr. Albert has received funding from the NIH (R01 HL116690; Dr. Turkhia has received grants from Janssen, AstraZeneca, Veterans Health Administration, Boehringer Ingelheim, Cardiva Medical, Bristol Myers-Squibb, and the American Heart Association, and consulting fees/honoraria from Medtronic, AliveCor, Abbott, Precision Health Economics, Zipline Medical, IBeat, and iRrythm. The remaining authors have no disclosures.

Conclusions

This analysis demonstrates that individuals at risk for AF can be identified from the general population with the use of a predictive algorithm. Increasing age and the presence of heart failure, coronary artery disease, hypertension, metabolic disorders, and hyperlipidemia represent this high-risk group. Respiratory failure and chronic kidney disease may also identify certain age groups at risk for AF. Awareness of this high-risk group can be a signal to primary care physicians to pay more attention to the possibility that these patients are at greater risk for AF, and, in turn, for stroke. With further validation, this predictive tool can be used to determine the need for AF detection devices, clinical decision-support tools and appropriate treatment plans.

References

  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]
  • [ PubMed ]