key: cord-0775812-v1ehhpzj
authors: Mountantonakis, Stavros E.; Saleh, Moussa; Fishbein, Joanna; Gandomi, Amir; Lesser, Martin; Chelico, John; Gabriels, James; Qiu, Michael; Epstein, Laurence M.
title: Atrial Fibrillation is an Independent Predictor for In-hospital Mortality in Patients Admitted with SARS-CoV-2 Infection
date: 2021-01-22
journal: Heart Rhythm
DOI: 10.1016/j.hrthm.2021.01.018
sha: f92adaa5ef1c2891d1a858f7dbcfcae26b637817
doc_id: 775812
cord_uid: v1ehhpzj

BACKGROUND: Atrial fibrillation (AF) is the most encountered arrhythmia and has been associated with worse in-hospital outcomes. OBJECTIVE: To determine the incidence of AF among patients hospitalized with COVID-19 as well as its impact on in-hospital mortality. METHODS: Patients hospitalized with a positive COVID-19 polymerase chain reaction test between March 1st and April 27(th), 2020, were identified from the common medical record system of 13 Northwell Health hospitals. Natural language processing search algorithms were utilized to identify and classify AF. Patients were classified as having AF or not. AF was further classified as new-onset vs past history of AF. RESULTS: AF occurred in 1,687 of 9,564 patients (17.6%). Of those, 1,109 patients (65.7%) had new-onset AF. Propensity score matching (PSM) 1,238 pairs of patients with AF and without AF showed higher in-hospital mortality in the AF group (54.3% vs 37.2%, p < 0.0001). Within the AF group, PSM of 500 pairs showed higher in-hospital mortality in patients with new onset AF as compared to a past history of AF (55.2% vs 46.8%, p=0.009). The risk ratio of in-hospital mortality for new onset AF over sinus rhythm patients was 1.56 (95% CI: 1.42 to 1.71, p<0.0001). The presence of cardiac disease was not associated with a higher risk of in-hospital mortality among patients with AF (p=0.1). CONCLUSIONS: Among patients hospitalized with COVID-19, 17.6% experienced AF. AF, particularly new-onset, was an independent predictor of in-hospital mortality.

Atrial fibrillation (AF) is the most common arrhythmia and has been associated with worse outcomes in hospitalized patients (1, 2) . Its prevalence and clinical impact are even more prominent in patients with pulmonary disease, critical illness, and systemic inflammatory response syndrome (3) (4) (5) .

COVID-19 in hospitalized patients is often characterized by respiratory involvement with hypoxia, systemic inflammatory response, and cardiac involvement, all of which are known predisposing factors for development of AF. In addition, the onset of AF has deleterious hemodynamic effects (6) that can further deteriorate the clinical presentation of a tenuous patient. Finally, the hypercoagulable state during the SARS-CoV-2 infection is likely to influence thromboembolic events related with AF (7) . The purpose of this study was to examine the incidence of AF among patients hospitalized with COVID-19 and evaluate its impact on inhospital mortality.

The study was conducted at 13 hospitals in the Northwell Health system. Patients admitted between March 1 st and April 27 th , 2020, with a positive COVID-19 polymerase chain reaction test were included in the study. Patients were followed through May 31 st , 2020. Patients with multiple re-admissions during the study period were evaluated as a single presentation.

Patients with a negative or absent nasopharyngeal swab, patients without progress notes or documented basic laboratory values within the first 96 hours of initial registration, patients who J o u r n a l P r e -p r o o f were still in the hospital as of May 31 st , and those whose outcome could not be determined as of May 31 st were excluded.

The Northwell Health Institutional Review Board approved this study. The research reported in this paper adhered to Helsinki Declaration as revised in 2013.

For data extraction, a detailed EMR search was performed, which included scanning of all medical notes, diagnoses, medications, orders, electrocardiogram and telemetry interpretations. First available laboratory results within 96 hours from initial presentation were used for the analyses. Natural language processing (NLP) techniques were used to detect AF diagnosis for each patient from a total of over 24 million progress notes in the dataset. To this end, a rule-based text classification algorithm was developed using the word "atrial fibrillation" and all possible variants (e.g., "atrial flutter", "rapid ventricular rate", "paroxysmal atrial fibrillation") and abbreviations (e.g., "AF", "afib", "RVR", "PAF"). The algorithm was designed to include misspellings and hyphenations in acronyms (e.g., "a-fib"). It excluded irrelevant acronym matches and false partial matches (e.g.,"has a fibroid"). Furthermore, a set of explicit language rules were developed to handle negation (e.g., "no evidence of AF"), uncertainty (e.g., "at risk of AF"), a reference to family history and other false positive cases such as "afib[ ]"

(where missing an "X" in square brackets indicates the absence). Python's NLTK library was used to tokenize the text into sentences and regular expressions module was used to develop and implement inclusion/exclusion rules at the sentence level. To evaluate the accuracy of the algorithm in detecting the presence of an AF diagnosis in patients' charts, a random set of 250 patients (~2.6% of the population) were independently and blindly reviewed by two physicians.

This revealed an overall accuracy of 99.6% (sensitivity = 100%, specificity = 98%) for AF detection. The rate of agreement between the two clinicians was 100%.

Similarly, a rule-based NLP algorithm was developed to determine whether a patient has new onset AF or a past history of AF. SAS Rule Builder procedure was first used to automatically extract an initial rule set based on 500 sentences labeled by two clinicians. Rules were then fine-tuned by iteratively implementing and evaluating the results at the patient level.

Satisfactory results were achieved after 11 iterations with a total of 46 inclusion/exclusion rules.

The final validation of the algorithm in differentiating between new vs known AF was performed by two clinicians who blindly and independently evaluated a random sample of 400 patients (~4.1% of the population). They had a disagreement rates of 4% on "past history of AF" and 9% on "New onset of AF". Disagreements were settled with deliberation. The final labels showed an overall accuracy of 95.0% for "new onset of AF" (sensitivity = 98.6%, specificity = 91.0%) and overall accuracy of 98.8% for "past history of AF" (sensitivity = 98.3%, specificity = 99.1%).

The AF group was defined as those patients who had AF during hospitalization irrespective of a history of AF. Those with AF were further classified into two sub-groups: patients with new-onset AF and patients with a past history of AF. Moreover, final diagnosis (ICD-9 and ICD-10) codes were also reviewed among the subset of patients who had any prior hospitalization recorded in the EMR to identify pre-existing conditions, including AF, not recorded during the patients' COVID admission.

J o u r n a l P r e -p r o o f

The primary outcome of this study was in-hospital mortality. In-hospital mortality was compared between patients with AF vs those without AF, between the two AF subgroups, and new onset AF and no AF patients. Finally, to examine the effect of underlying cardiac disease (i.e., coronary artery disease, congestive heart failure, valvular heart disease) on hospital mortality in patients with AF, patients with a history of cardiac disease were compared to those without such history.

Propensity score matching (PSM) was utilized to control for potential confounding variables. The efficiency of the PSM procedure in producing comparability of the two groups after matching was evaluated using standardized mean differences (SMD). An absolute SMD of <0.1 was deemed as adequate comparability (8) . Descriptive statistics on the factors used to develop the propensity score in the entire cohort before and after matching are presented in Table   1 . Continuous factors are reported as mean, standard deviation or median (first quartile, third quartile) in their original units, and categorical factors are presented using frequencies and proportions.

Exact binomial methods were used to compute the proportion of patients with AF along with the corresponding 95% confidence intervals.

McNemar's test for matched pairs was utilized to assess if the risk of in-hospital mortality differed according to status of AF and relevant subgroups during the patients' hospital stay (9) .

Risk ratios (RR) for the matched pairs analyses were computed and corresponding 95% confidence intervals (95% CIs) were estimated using 1,000 bootstrap resamples with replacement.

All analyses were conducted using SAS version 9.4 (SAS Institute Inc., Cary, NC).

A total of 10,207 COVID-19 patients were admitted between March 1 and April 27, 2020. Of those patients, 9,564 patients met inclusion criteria. Eight thousand fifty-eight patients (84.3%) had one hospital admission and 1,506 (15.7%) had more than one hospital admission. 63.4% to 68.0%) had new onset AF and 578 (34.3%, 95% CI: 32.0% to 36.6%) had a past medical history of AF, of which 518 were identified using NLP only and 60 patients were additionally identified using ICD-9/ICD-10 codes from prior hospitalization records in the EMR.

There were 109 patients with a history of AF but no documented AF during hospitalization (1.4%).

Before matching, the average age of the cohort was 64.8 ± 16.4 years and 5,632 (58.9%) were male. Comorbidities that were more frequent in patients with AF included history a of hypertension, coronary artery disease, congestive heart failure, peripheral vascular disease, renal disease, chronic obstructive pulmonary disease, cancer, and diabetes ( Table 1 ). As measured by the SMD, BUN and creatinine values were higher in the AF patients compared to sinus rhythm patients. The lymphocyte count was lower in the AF group. There were no differences in the J o u r n a l P r e -p r o o f inflammatory biomarkers namely C-reactive protein (CRP), ferritin and d-dimer. Patients with in-hospital AF were more likely to receive mechanical ventilation than patients who never experienced AF during their hospitalization (37.5% vs 15.9%, p<0.0001).

After controlling for possible confounding factors using PSM, 1,238 patients with AF were successfully matched with patients with no AF, for a total of 2,476 patients. In-hospital mortality was observed in 672 patients with AF compared to 460 no AF patients (54.3% vs 37.2%, p < 0.0001). The relative risk (RR) of in-hospital mortality among those with AF compared to those without was 1.46 (95% CI: 1.34-1.59).

To examine the incremental effect of new onset AF within the AF group, 500 patients with new onset AF were again matched with 500patients with a past history of AF. Among those patients, 276 with new onset AF died in-hospital compared to 234 with a past history of AF (55.2% vs 46.8%, p=0.009), for a RR of in-hospital mortality of 1.18 (95% CI: 1.04-1.33).

In-hospital mortality of the new onset AF group was also compared to the no AFgroup.

Among 1,109 patients with new onset AF, 1,107 patients were matched and compared to patients without AF. In-hospital mortality risk was significantly higher in the new onset AF group (56.1% vs 36.0%) yielding a risk ratio of 1.56 (95% CI: 1.42 to 1.71, p<0.0001).

The relationship between AF and in-hospital mortality was also examined in patients with cardiac disease. PSM analysis of 568 matched pairs did not suggest a difference in risk of in-J o u r n a l P r e -p r o o f hospital mortality between AF patients with cardiac disease vs patients without cardiac disease.

Finally, 48.1% of those with history of cardiac disease and AF died in-hospital compared to 52.6% without cardiac disease history and AF (RR: 0.91, 95% CI: 0.81 to 1.03, p=0.12). Relative risk of in-hospital mortality for each analysis are shown in figure 2.

The main findings of this study in a sick cohort of hospitalized COVID 19 patients were:

(1) 17.6% experienced AF and 12.5% of patients without a history of AF were diagnosed with new onset AF, (2) the risk of in-hospital mortality was significantly higher in patients that experienced AF (54% vs 37%), and (3) AF, particularly new onset AF, was independently associated with in-hospital mortality.

It is estimated that 33.5 million people carry a diagnosis of AF worldwide (1-2), with AF more prevalent in the elderly, males, patients with hypertension, heart failure, coronary artery disease, valvular disease, obesity, diabetes, and renal disease (3) (4) (5) . Patients with the above comorbidities are also more likely to develop a more aggressive course of COVID-19 and be hospitalized (10) . Baseline characteristics of the entire unmatched cohort showed that patients who developed AF had a higher prevalence of the above comorbidities, were more likely to receive mechanical ventilation, and had significantly higher in-hospital mortality.

The prevalence and incidence of AF during hospitalization for COVID-19is unclear; however, one should expect similarities with other systemic inflammatory response syndromes and sepsis. In the largest series of 60,209 Medicare beneficiaries with sepsis and a mean age of 80.2 years, AF was present in 25.5% of patients. In the present study, 17.6 % of patients J o u r n a l P r e -p r o o f experienced AF, which is particularly high, considering that this cohort was much younger and included patients treated in a non-ICU setting (11) . In a different series of 49,082 patients (mean age 69 years, 48% females) with severe sepsis, AF was newly diagnosed in 5.9% of patients (12) . Bhatla et al reported a 3.5% incidence of AF in 700 patients (mean age of 50 ± 18 years) hospitalized with COVID-19 of which 79 (11.2%) were in an ICU setting. (13) . This is in contrast to the much higher 11.6% of new onset AF in this study. In addition to the much greater size of the current cohort, differences are most likely due to the stricter admission criteria 

AF is previously described as a risk factor of increased all-cause mortality (14) (15) (16) (17) . In 3,100 hospitalized patients with sepsis, AF was associated with a 1.45 RR increase in mortality (18) . A retrospective study of 3,240,083 patients showed that patients with severe sepsis that experienced AF were more likely to die (OR 1.19; 95% CI 1.14-1.24) (19) . In particular, new onset AF is an independent risk factor of mortality in patients admitted to the ICU with severe sepsis or septic shock (20) . For cardiac patients, AF is an independent mortality predictor for patients presenting with myocardial infarction (21) or heart failure exacerbation (17) . The present cohort supports the notion that AF is a marker of severe systemic illness similarly to the sepsis literature. Nevertheless, the results of the strict propensity score matching that controlled for all previously described demographic, clinical and laboratory confounders support an J o u r n a l P r e -p r o o f independent association between COVID-19 and AF that has not been described previously. The strong association between AF and COVID-19 was most likely driven by the new AF cases, with more than 50% of patients with newly diagnosed AF dying; again, suggesting the strong association between the development of AF and in-hospital mortality.

This study highlights the importance of utilizing AF as a clinical and non-invasive marker of in-hospital mortality in hospitalized COVID-19 patients. To date, laboratory markers on admission such as CRP and D-Dimer have been utilized clinically to identify COVID-19 patients with a poorer prognosis. In a recent report, higher levels of CRP and a D-dimer level > 2.0 mg/L have a relative risk for mortality of 11.97 and 10.17 respectively (22) (23) (24) . In the present study, both CRP and D-dimer did not differ between the AF and no AF groups. In addition, both biomarkers were controlled for in all PSM analyses. The RR of 1.56 of new onset AF provides an incremental risk to the two inflammatory biomarkers that are currently used as mortality predictors, suggesting that the association between COVID-19 and AF might be due to mechanisms other than systemic inflammatory response.

Recently, it has been suggested that the SARs-CoV-2 virus may directly contribute to the pathogenesis of AF through attaching to pericytes, cells responsible for microvascular integrity of cardiac tissue. This results in the release of a number of growth factors, causing cardiac tissue inflammation and altering atrial cellular electrophysiology (25) (26) (27) . Similarly, dysregulation of cellular ACE-2 receptors by the SARs-CoV-2 virus results in the release of angiotensin II further contributing to AF (38-30).

Understanding how AF contributes to the increased risk of in-hospital mortality is important. The present cohort is reflective of a sick COVID-19 population admitted to the J o u r n a l P r e -p r o o f hospital at the height of the NY pandemic. The health system was flooded with the sickest COVID-19 patients, overwhelming care and rapidly depleting resources. Most patients were treated with a rate-controlled strategy. Trans-esophageal echocardiograms and cardioversions were discouraged due to exposure risk. The use of anticoagulation across the health system evolved rapidly during the study period. Though most patients admitted during the study period were likely to have received some form of anticoagulation for COVID-19, the true prevalence and type of anticoagulation in this cohort, especially related to the timing of the development of AF was not in the scope of this study.

The study is subject to the same limitations as other retrospective studies. Due to the nature of collecting the data from the EMR, and the decreased rigor in documentation during the pandemic, the true incidence of AF may be higher due to reporting bias. However, the NLP algorithm was created after rigorous review of a random sample of medical records by electrophysiologists and enabled a review of more than 24 million notes. The results are not generalizable to those with that may develop AF in the outpatient setting.

Atrial fibrillation occurred in 17.6% of patients hospitalized with COVID-19. In 12.5% of patients there was no prior history of AF (new onset AF). The occurrence of any AF and particularly new onset AF, was independently associated with a significantly higher in-hospital mortality.

J o u r n a l P r e -p r o o f The above plot provides risk ratios along with corresponding 95% confidence intervals for four separate analyses. Each risk ratio should be interpreted independently and not

Worldwide epidemiology of atrial fibrillation: a Global Burden of Disease 2010 Study

Estimates of current and future incidence and prevalence of atrial fibrillation in the U.S. adult population

Epidemiology of atrial fibrillation: European perspective

An update on the prognosis of patients with atrial fibrillation

Distribution and risk profile of paroxysmal, persistent, and permanent atrial fibrillation in routine clinical practice: insight from the real-life global survey evaluating patients with atrial fibrillation international registry

Haemodynamic effects of an irregular sequence of ventricular cycle lengths during atrial fibrillation JACC

Thrombosis in Hospitalized Patients With COVID-19 in a New York City Health System

Using the Standardized Difference to Compare the Prevalence of a Binary Variable Between Two Groups in Observational Research

Comparing paired vs non-paired statistical methods of analyses when making inferences about absolute risk reductions in propensity-score matched samples

Comorbidity and its Impact on Patients with COVID-19

Atrial fibrillation among Medicare beneficiaries hospitalized with sepsis: incidence and risk factors

Incident stroke and mortality associated with new-onset atrial fibrillation in patients hospitalized with severe sepsis

COVID-19 and cardiac arrhythmias. Heart Rhythm

Impact of atrial fibrillation on the risk of death: the Framingham Heart Study

A population-based study of the long-term risks associated with atrial fibrillation: 20-year follow-up of the Renfrew/ Paisley study

All-cause mortality in 272,186 patients hospitalized with incident atrial fibrillation 1995-2008: a Swedish nationwide long-term casecontrol study

Presence of atrial fibrillation is independently associated with adverse outcomes in patients J o u r n a l P r e -p r o o f hospitalized with heart failure: an analysis of get with the guidelines-heart failure. Circ Heart Fail

New-onset atrial fibrillation in sepsis is associated with increased morbidity and mortality

Atrial fibrillation is more frequent in patients with severe sepsis and septic shock predicts a worse outcome

Incidence and Prognosis of Atrial Fibrillation in Patients With Sepsis

Prognostic impact of types of atrial fibrillation in acute coronary syndromes

C-reactive protein: A promising biomarker for poor prognosis in COVID-19 infection

Value of leukocytosis and elevated C-reactive protein in predicting severe coronavirus 2019 (COVID-19): A systematic review and metaanalysis

D-dimer as a biomarker for disease severity and mortality in COVID-19 patients: a case control study. j intensive care

It takes two: endothelial-perivascular cell cross-talk in vascular development and disease

Capillary and arteriolar pericytes attract innate leukocytes exiting through venules and 'instruct' them with pattern-recognition and motility programs

Immune functions and properties of resident cells in the heart and cardiovascular system: pericytes

Microvascular responses to cardiovascular risk factors

Angiotensin-(1-7) and angiotensin-(1-9): function in cardiac and vascular remodelling

Distinct patterns of atrial electrical and structural remodeling in angiotensin II mediated atrial fibrillation