You are here

Article Text

Article menu
Arrhythmias and sudden death
Original research
Characterising symptom clusters in patients with atrial fibrillation undergoing catheter ablation
  1. Mollie Hobensack1,
  2. Yihong Zhao1,
  3. Danielle Scharp1,
  4. Alexander Volodarskiy2,
  5. David Slotwiner2,3 and
  6. Meghan Reading Turchioe1
  1. 1Columbia University School of Nursing, New York City, New York, USA
  2. 2Cardiology, NewYork-Presbyterian Queens Hospital, Flushing, New York, USA
  3. 3Department of Population Health Sciences, Weill Cornell Medicine, New York City, New York, USA
  1. Correspondence to Dr Meghan Reading Turchioe; mr3554{at}cumc.columbia.edu

Abstract

Objective This study aims to leverage natural language processing (NLP) and machine learning clustering analyses to (1) identify co-occurring symptoms of patients undergoing catheter ablation for atrial fibrillation (AF) and (2) describe clinical and sociodemographic correlates of symptom clusters.

Methods We conducted a cross-sectional retrospective analysis using electronic health records data. Adults who underwent AF ablation between 2010 and 2020 were included. Demographic, comorbidity and medication information was extracted using structured queries. Ten AF symptoms were extracted from unstructured clinical notes (n=13 416) using a validated NLP pipeline (F-score=0.81). We used the unsupervised machine learning approach known as Ward’s hierarchical agglomerative clustering to characterise and identify subgroups of patients representing different clusters. Fisher’s exact tests were used to investigate subgroup differences based on age, gender, race and heart failure (HF) status.

Results A total of 1293 patients were included in our analysis (mean age 65.5 years, 35.2% female, 58% white). The most frequently documented symptoms were dyspnoea (64%), oedema (62%) and palpitations (57%). We identified six symptom clusters: generally symptomatic, dyspnoea and oedema, chest pain, anxiety, fatigue and palpitations, and asymptomatic (reference). The asymptomatic cluster had a significantly higher prevalence of male, white and comorbid HF patients.

Conclusions We applied NLP and machine learning to a large dataset to identify symptom clusters, which may signify latent biological underpinnings of symptom experiences and generate implications for clinical care. AF patients’ symptom experiences vary widely. Given prior work showing that AF symptoms predict adverse outcomes, future work should investigate associations between symptom clusters and postablation outcomes.

  • Atrial Fibrillation
  • Electronic Health Records
  • Catheter Ablation

Data availability statement

Data are available upon reasonable request. De-identified data used in this study will be made available upon reasonable request to the corresponding authors, following completion of an institutional data sharing agreement.

http://creativecommons.org/licenses/by-nc/4.0/

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

https://doi.org/10.1136/openhrt-2023-002385

Statistics from Altmetric.com

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

    What is already known on this topic

    • Symptom burden is an important consideration in the clinical decision-making process to determine appropriate treatment strategies for patients with atrial fibrillation (AF). Symptom clusters have been identified in patients with AF, but not specifically at the time of catheter ablation, making comparisons challenging.

    What this study adds

    • This study highlights the utility of natural language processing and machine learning clustering analyses to reveal the multidimensional nature of co-occurring symptoms among patients with AF.

    How this study might affect research, practice or policy

    • The six symptom clusters identified provide a foundation for future work examining whether and how specific symptom clusters predict future adverse events postablation such as recurrence, hospitalisations and stroke.

    Introduction

    Atrial fibrillation (AF) is the most common cardiac arrhythmia and its prevalence continues to grow; an estimated 6–12 million people will be diagnosed with AF in the USA by 2050 and 17.9 million people in Europe by 2060.1 2 Individuals living with AF experience burdensome symptoms including dyspnoea, chest pain, fatigue and palpitations that impact quality of life.3 Symptom burden is an important consideration in the clinical decision-making process to determine appropriate treatment strategies, such as catheter ablation.

    For many clinicians, AF symptoms remain an enigma. Symptoms may relate to cardiac rhythm; asymptomatic or ‘silent AF’ is common as is patient reports of symptoms during normal sinus rhythm.4 Conceptual models of AF symptoms propose numerous correlates of symptoms, including cardiac function (autonomic factors, heart rate and rhythm, decreased atrial contraction, decreased cardiac output, decreased left ventricular filling), affect (positive or negative), socioeconomic status and comorbidities.5 Symptoms are important to detangle because of their associations with future adverse outcomes, particular hospitalisations.6 7 However, there is a dearth of research describing the presentation of AF symptoms in practice and confirming potential biological and sociodemographic correlates of symptoms.8

    Electronic health records (EHRs) contain valuable information about patient symptoms. Informatics methods including data mining and natural language processing (NLP) can be used to study symptoms at scale within and across health systems. NLP can be applied to clinical notes to process, extract and capture instances of symptom reports from free text contained in clinical notes.9 Machine learning methods including clustering analyses applied to data extracted from EHRs can reveal which symptoms co-occur (ie, symptom clusters of two or more symptoms) to characterise the presentation and correlates of symptoms.10

    Despite the applications of NLP to uncover symptom clusters in other domains,9 to the best of our knowledge no previous study has analysed AF symptom clusters among patients undergoing ablation.11 Examining AF symptoms among all patients regardless of treatment stage makes comparisons challenging, given the profound impact treatments may have on symptoms. Examining patients who are about to undergo catheter ablation standardises the comparison timepoint and captures patients at a time that they are likely to be symptomatic (as symptoms are a major indication for ablation).

    Therefore, this study aims to leverage these novel datasets and methods to (1) identify co-occurring symptoms of patients prior to undergoing catheter ablation and (2) describe clinical and sociodemographic correlates of symptom clusters.

    Methods

    Study design and participants

    This study uses cross-sectional, retrospective data from one large, urban academic medical centre in New York City. Patients included had a primary diagnosis of paroxysmal AF which was identified using ICD-9 and ICD-10 codes. Patients were treated at our institution between 1 January 2010 and 31 December 2020. Eligibility criteria to be included in the patient cohort were as follows: (1) underwent de novo catheter ablation for the treatment of AF, determined using CPT codes, (2) age 18 or older and (3) had at least one encounter note available within 30 days of the date of the ablation. We intentionally examined patients undergoing ablation because it is typical for preablation notes to be rich with symptom descriptions, given that symptom relief is a primary indication for ablation.12

    Data sources

    Structured data

    We extracted structured EHR data, including information such as demographics, comorbid diagnoses and medications, from our institution’s clinical data warehouse using SQL. Demographic variables included age, gender, race, ethnicity and comorbid heart failure (HF) at the time ablation. We also extracted relevant preablation medications: (1) antiarrhythmic agents (amiodarone, dofetilide, dronedarone, flecainide and propafenone), (2) rate control medications (atenolol, diltiazem, metoprolol, atenolol and verapamil) and (3) anticoagulant medications (apixaban, dabigatran, edoxaban, rivaroxaban and warfarin). Finally, as a measure of socioeconomic vulnerability, we calculated the social deprivation index corresponding with each patient’s home address using zip codes.13

    Unstructured data

    Unstructured data included inpatient and outpatient notes (eg, nursing notes, encounter notes, history and physicals, discharge summaries) written by cardiac electrophysiology clinicians within 30 days of the index catheter ablation. We used the NLP software Nimbleminer, an R shiny application that allows the user to create a customised symptom vocabulary using word embeddings and extracts concepts of interest (such as symptoms) from notes using rule-based methods.14 Our preliminary list of 10 AF symptoms was created through literature searches15 and in consultation with cardiologists and registered nurses: anxiety, chest pain, dizziness, fatigue, malaise, palpitations, dyspnoea, oedema, lightheadedness, and weakness.

    To evaluate the accuracy of the NLP tool in this context, we first created a gold standard corpus of a subset of 400 randomly selected narrative notes manually annotated by at least two clinical experts. We compared model performance to the gold standard corpus using precision (the percentage of results that were relevant among all results), recall (the percentage of results that were obtained by the model among all results that should have been obtained) and F-score (the harmonic mean of precision and recall). The model achieved acceptable performance (precision=0.73, recall=0.93 and F-score=0.81). We then applied our algorithm to the remaining notes (total n=13 416) on our entire cohort of patients (n=1293). Further description of this process can be found here which extensively describes the NLP involved in this project.16

    Statistical methods

    Descriptive statistics (eg, mean and percentages) were used to summarise the sample characteristics. One patient was excluded from the analyses due to missing data on the baseline symptom measures. Hierarchical cluster analyses were performed to (1) characterise symptom co-occurrence patterns and (2) identify subgroups of patients with similar symptom co-occurring patterns, respectively. Specifically, Ward’s hierarchical agglomerative clustering method is a type of unsupervised machine learning that illuminates natural groups using unlabelled data.17 18 Since the symptom measures were coded as binary indicators in this study, the Jaccard distance19 was chosen as the dissimilarity measure in the clustering analyses.

    To determine the optimal number of clusters in the data, a wide variety of clustering evaluation criteria have been proposed in literature, which include but are not limited to Davies-Bouldin index,20 Calinski-Harabasz index,21 Silhouette index22 and GAP.23 Each criterion measures different statistical properties of the data. In this study, the NbClust function implemented in the R NbCluster package24 was used to determine the optimal number of clusters. Our analyses discovered two symptom co-occurrence patterns. In addition, based on patients’ symptom profiles, clustering analyses identified six distinct subgroups. To better understand the characteristics of these subgroups, we calculated the prevalence of each symptom within each subgroup. We noticed that one subgroup did not report any symptoms and thus was treated as our reference group (ie, asymptomatic group) in subsequent analyses.

    Fisher’s exact tests were used to investigate subgroup differences in terms of key demographic and clinical characteristics such as age, gender, race and HF status. Here, age was defined as 65 versus under 65, and race was coded as white, missing and others. All analyses were conducted in R V.4.1.2. All tests were two-sided. To adjust for multiple comparison problems, we controlled the false discovery rate at 0.05 level.

    Patient and public involvement

    Patients were first involved in the conceptualisation of this research question. We previously conducted qualitative interviews with patients to determine their symptom experiences before and after catheter ablation.25 This served as the foundation for our research question and study design. This was a retrospective observational study with no patient recruitment or intervention delivery for patients to become involved in. In future work we plan to create a patient advisory board to present the findings of this study and other ongoing projects, and guide dissemination based on their feedback.

    Results

    Sample characteristics

    There was a total of 1293 patients included in our analysis with a primary diagnosis of paroxysmal AF treated at our institution between 1 January 2010 and 31 December 2020. In our sample, the mean age of the participants was 65.5 (SD 12.6) years old. In addition, approximately one-third of our sample identified as female (35.2%) and two-thirds identified as white (58.0%). Further patient characteristics are described in table 1.

    View this table:
    Table 1

    Patient characteristics (mean (SD) or n (%)), n=1293

    Symptom characteristics

    AF symptoms from preablation notes were extracted from 76% of patients undergoing ablation, indicating high symptom enrichment in the notes and suggesting routine interrogation of the symptoms in clinical practice. The most frequently documented symptom was dyspnoea (64%), oedema (62%) and palpitations (57%). The prevalence of each symptom annotated in the clinical notes is contained in figure 1.

    Figure 1
    Figure 1

    Prevalence of each symptom at baseline among 1293 patients undergoing catheter ablation.

    Symptom clusters

    We identified six clusters as optimal which revealed symptoms that co-occurred together. In figure 2, the y-axis indicates the prevalence of each symptom in the cluster and the x-axis shows one of the 10 symptoms. Cluster 5, the reference group, includes all asymptomatic patients. Cluster one is characterised as broadly symptomatic, as the prevalence of nearly all symptoms exceeds 50% in this cluster. Cluster two is characterised by shortness of breath and swelling. Cluster three is characterised by chest pain. Cluster four is characterised by anxiety. Cluster six is characterised by fatigue and palpitations. For the purposes of more clearly articulating the unique attributes of each cluster, we will now refer to each cluster by its predominant symptom(s).

    Figure 2
    Figure 2

    Symptom clusters identified at baseline (n=1293).

    Associations between patient characteristics and symptom clusters

    Table 2 highlights the statistically significant variables; all variables included in the statistical analysis are presented in online supplemental table 1. We found statistically significant differences in gender, race and HF between multiple clusters compared with the reference group (asymptomatic group). There was a significantly higher proportion of females in the broadly symptomatic (39%) and anxiety (42%) clusters compared with the asymptomatic cluster (19%). There was also a significantly higher proportion of non-white patients in the broadly symptomatic (44%), dyspnoea/oedema (43%), chest pain (43%), anxiety (41%) and fatigue/palpitations (62%) clusters compared with the asymptomatic cluster (19%). There was a significantly lower proportion of patients with a comorbid diagnosis of HF in the chest pain (41%) and fatigue/palpitations (44%) cluster compared with the asymptomatic cluster (69%). The clusters with the most prevalent proportions of HF patients were asymptomatic (69%) and dyspnoea/oedema (66%).

    View this table:
    Table 2

    Associations between symptom clusters and age, gender, race, ethnicity and heart failure status, n (%)

    Discussion

    Symptom clusters help to identify latent biological underpinnings of symptom experiences and generate important implications for clinical care. In this study we processed symptom concepts from over 13 000 clinical notes from AF patients undergoing ablation and identified six distinct symptom clusters, each with unique defining symptoms. The six distinct symptom clusters highlight the broad range of potential symptom experiences that patients with AF may experience.

    Prior work has described prevalence rates for typical AF symptoms, including palpitations, dyspnoea and fatigue.15 This analysis builds on prior work by highlighting the multidimensional nature of co-occurring symptoms, some unexpected (eg, anxiety); while many clusters could be characterised by one or two defining symptoms, a proportion of patients in each cluster also experienced a number of other symptoms. Streur and colleagues also described AF symptom clusters but used self-reported symptoms from a registry of any AF patient (not specifically preablation).10 They identified two main clusters, versus our six, but many subclusters visible on their dendrogram mirrored our cluster results. Additionally, they identified associations between symptom clusters and gender, race and HF, but not age, which also aligns with our findings. The concordance between their findings, derived from patient self-report, and ours, derived from clinician documentation in EHRs, adds validity to the secondary use EHR data to study AF symptoms and other outcomes. Whereas EHRs have certain known data quality and missingness issues they provide information on large samples of patients when patient self-reported data may be infeasible to collect at scale and inhibited by high attrition.10

    While prior work suggests that patient-reported and EHR-documented symptoms may not always be concordant,6 26 among patients with AF concordance is higher7 potentially due to systematic interrogation and symptoms being a primary indication for ablation. One way to improve concordance both for rhythm and symptom reports is through the use of mobile apps to integrate patient generated-data into the EHR.27–29 Future research should investigate correlations between symptom clusters and AF burden using robust detection methods, such as wearable monitors. Because symptom-rhythm correlation is notoriously difficult to disentangle, symptom clusters may be helpful for clinicians trying to create a clinical picture of a patient’s symptoms in correlation with their AF. Knowledge of existing clusters may guide the clinical assessment. Furthermore, in future work, we plan to evaluate whether clusters predict adverse outcomes—which may further enhance their clinical utility.

    Identification of symptom clusters lays the foundation for future work examining whether and how specific symptom clusters predict future adverse events postablation, including recurrence, hospitalisations and stroke. Prior work has related non-specific and intermittent AF symptoms to higher stroke and mortality rates, attributed partially to delayed care-seeking, compared with cardiac-specific symptoms (chest pain, palpitations).30 Large cohort studies including Outcomes Registry for Better Informed Treatment of AF,31 and Rate Control Efficacy in Permanent AF,32 have associated AF symptom severity with higher hospitalisation rates. Similarly, preablation symptom clusters may be predictive of future ablation success and other AF outcomes, and potentially serve as one prognostic indicator as patients and clinicians consider ablation therapies in clinical care. Future researchers are encouraged to explore cluster phenotype overtime to predict adverse events that are individualised to each patient.33 34

    Important differences between patients in the asymptomatic cluster and those in other symptom clusters may suggest underlying differences in symptom perception. A significantly higher proportion of patients in the broadly symptomatic and anxiety clusters were female compared with male, and 81% of the asymptomatic cluster were males. Concordantly, prior work has suggested greater symptom perception among female patients.15 A higher prevalence of patients taking antiarrhythmic medications were in the five symptom clusters versus the asymptomatic cluster, although this did not rise to the level of statistical significance (p<0.05). Patients taking antiarrhythmic medications may have a greater symptom burden.3 Consistent with prior reports, a higher prevalence of non-white patients were in the five symptom clusters versus the asymptomatic cluster, but reasons for racial and ethnic differences in AF symptom experiences are largely underexplored.15

    The asymptomatic cluster may also reflect blunted symptom experiences from both the pathophysiological changes of specific diseases (such as vascular changes) as well as mild cognitive impairment associated with chronic conditions and the ageing process.35 Although no differences rose to the level of statistical significance, there were a higher prevalence of patients with comorbid diabetes, hypertension and vascular disease in the asymptomatic cluster compared with other symptom clusters. The possibility of AF symptom blunting as a result of other pathophysiological changes of these conditions has been explored in HF but not AF.35

    The majority of asymptomatic patients also had HF (69%). Catheter ablations of HF patients in the setting of asymptomatic AF have become more common since the CASTLE-AF trial demonstrated a significant mortality benefit for these patients.36 In fact, within our dataset we saw a rise in the proportion of asymptomatic patients undergoing ablation in 2020, after CASTLE-AF trial findings were published (online supplemental file). Associations between HF and other symptom clusters highlight how HF modifies the AF symptom experience. Specifically, more HF patients were in the dyspnoea/oedema cluster and fewer in the chest pain or fatigue/palpitations clusters. Therefore, HF patients may experience less burden of classic AF symptoms, or potentially a blunted AF symptom experience due to pathophysiological changes of the disease.35 This raises important questions for future work regarding the effectiveness of various rhythm control strategies in treating symptoms experienced by patients with AF and HF.

    Finally, while a majority of the defining symptoms in each cluster are commonly acknowledged in the literature,15 anxiety, a less frequently discussed symptom related to AF, surfaced as a main driver of one of the clusters. Previous research has examined the importance of anxiety on AF symptom severity15 and others have noted that anxiety may actually increase after catheter ablations to treat AF and lead to higher odds of recurrence.37 Therefore, anxiety as a predominant symptom of AF represents an important symptom cluster warranting closer examination.

    Limitations of this study include high missingness of race and ethnicity, which are typically under-reported and sometimes inaccurate in EHRs. This limited our ability to critically examine race and ethnicity as biological variables in relation to symptom clusters. Additionally, our study reports on a single medical centre in New York City where patients are predominantly White and many are in middle to high socioeconomic groups, which limits generalisability of findings to other populations.

    Conclusion

    In this study of over 13 000 clinical notes from 1293 AF patients undergoing ablation, we identified six distinct symptom clusters with unique defining symptoms, which we associated with gender, race and HF status. AF patients undergoing ablation may experience a broad range of symptoms which may differ by demographic and clinical characteristics. Grounded in prior work demonstrating that AF symptoms are predictive of adverse outcomes, future research relating symptom clusters to postablation outcomes may generate informative prognostic tools for clinicians.

    Data availability statement

    Data are available upon reasonable request. De-identified data used in this study will be made available upon reasonable request to the corresponding authors, following completion of an institutional data sharing agreement.

    Ethics statements

    Patient consent for publication

    Ethics approval

    This study was approved by the Weill Cornell Medicine (protocol: 19-11021059) and Columbia University Irving Medical Centre Institutional Review Boards (protocol: AAAU2028).

    References

    1. ↵
      1. Patel NJ,
      2. Deshmukh A,
      3. Pant S, et al
      . Contemporary trends of hospitalization for atrial fibrillation in the United States, 2000 through 2010. Circulation 2014;129:2371–9. doi:10.1161/CIRCULATIONAHA.114.008201
      OpenUrlAbstract/FREE Full Text
    2. ↵
      1. Krijthe BP,
      2. Kunst A,
      3. Benjamin EJ, et al
      . Projections on the number of individuals with atrial fibrillation in the European Union, from 2000 to 2060. Eur Heart J 2013;34:2746–51. doi:10.1093/eurheartj/eht280
      OpenUrlCrossRefPubMedWeb of Science
    3. ↵
      1. Freeman JV,
      2. Simon DN,
      3. Go AS, et al
      . Association between atrial fibrillation symptoms, quality of life, and patient outcomes: results from the outcomes Registry for better informed treatment of atrial fibrillation (ORBIT-AF). Circ Cardiovasc Qual Outcomes 2015;8:393–402. doi:10.1161/CIRCOUTCOMES.114.001303
      OpenUrlAbstract/FREE Full Text
    4. ↵
      1. Heidt ST,
      2. Kratz A,
      3. Najarian K, et al
      . Symptoms in atrial fibrillation: a contemporary review and future directions. J Atr Fibrillation 2016;9:1422. doi:10.4022/jafib.1422
      OpenUrl
    5. ↵
      1. Rienstra M,
      2. Lubitz SA,
      3. Mahida S, et al
      . Symptoms and functional status of patients with atrial fibrillation: state of the art and future research opportunities. Circulation 2012;125:2933–43. doi:10.1161/CIRCULATIONAHA.111.069450
      OpenUrlFREE Full Text
    6. ↵
      1. Sikorskii A,
      2. Wyatt G,
      3. Tamkus D, et al
      . Concordance between patient reports of cancer-related symptoms and medical records documentation. J Pain Symptom Manage 2012;44:362–72. doi:10.1016/j.jpainsymman.2011.09.017
      OpenUrlPubMed
    7. ↵
      1. Katsumata Y,
      2. Kohsaka S,
      3. Ikemura N, et al
      . Symptom under-recognition of atrial fibrillation patients in consideration for catheter ablation: a report from the Kics-AF Registry. JACC Clin Electrophysiol 2021;7:565–74. doi:10.1016/j.jacep.2020.10.016
      OpenUrl
    8. ↵
      1. Asad ZUA,
      2. Yousif A,
      3. Khan MS, et al
      . Catheter ablation versus medical therapy for atrial fibrillation: a systematic review and meta-analysis of randomized controlled trials. Circ Arrhythm Electrophysiol 2019;12:e007414. doi:10.1161/CIRCEP.119.007414
    9. ↵
      1. Koleck TA,
      2. Dreisbach C,
      3. Bourne PE, et al
      . Natural language processing of symptoms documented in free-text narratives of electronic health records: a systematic review. J Am Med Inform Assoc 2019;26:364–79. doi:10.1093/jamia/ocy173
      OpenUrlCrossRefPubMed
    10. ↵
      1. Streur M,
      2. Ratcliffe SJ,
      3. Callans D, et al
      . Atrial fibrillation symptom clusters and associated clinical characteristics and outcomes: a cross-sectional secondary data analysis. Eur J Cardiovasc Nurs 2018;17:707–16. doi:10.1177/1474515118778445
      OpenUrl
    11. ↵
      1. Reading Turchioe M,
      2. Volodarskiy A,
      3. Pathak J, et al
      . Systematic review of current natural language processing methods and applications in cardiology. Heart 2022;108:909–16. doi:10.1136/heartjnl-2021-319769
      OpenUrlAbstract/FREE Full Text
    12. ↵
      1. Writing Group Members,
      2. January CT,
      3. Wann LS, et al
      . 2019 AHA/ACC/HRS focused update of the 2014 AHA/ACC/HRS guideline for the management of patients with atrial fibrillation: a report of the American College of Cardiology/American Heart Association task force on clinical practice guidelines and the Heart Rhythm Society. Heart Rhythm 2019;16:e66–93. doi:10.1016/j.hrthm.2019.01.024
      OpenUrlCrossRefPubMed
    13. ↵
      1. Social Deprivation Index (SDI)
      . Robert Graham center. 2023. Available: https://www.graham-center.org/maps-data-tools/social-deprivation-index.html
    14. ↵
      1. Topaz M,
      2. Murga L,
      3. Bar-Bachar O, et al
      . Nimbleminer: an open-source nursing-sensitive natural language processing system based on word embedding. Comput Inform Nurs 2019;37:583–90. doi:10.1097/CIN.0000000000000557
      OpenUrl
    15. ↵
      1. Jurgens CY,
      2. Lee CS,
      3. Aycock DM, et al
      . State of the science: the relevance of symptoms in cardiovascular disease and research: a scientific statement from the American Heart Association. Circulation 2022;146:e173–84. doi:10.1161/CIR.0000000000001089
      OpenUrl
    16. ↵
      Abstract 13012: atrial fibrillation symptom improvement following de novo catheter ablation: Trajectories and predictors of symptom resolution [Circulation]. Available: https://www.ahajournals.org/doi/abs/10.1161/circ.146.suppl_1.13012
    17. ↵
      1. Murtagh F,
      2. Legendre P
      . Ward’s hierarchical agglomerative clustering method: which algorithms implement ward’s criterion? J Classif 2014;31:274–95. doi:10.1007/s00357-014-9161-z
      OpenUrlCrossRef
    18. ↵
      1. Greene D,
      2. Cunningham P,
      3. Mayer R
      . Unsupervised learning and clustering. Cogn Technol 2008;51. doi:10.1007/978-3-540-75171-7
    19. ↵
      1. Real R,
      2. Vargas JM,
      3. Olmstead R
      . The probabilistic basis of Jaccard’s index of similarity. Syst Biol 1996;45:380–5. doi:10.1093/sysbio/45.3.380
      OpenUrlCrossRefWeb of Science
    20. ↵
      1. Davies DL,
      2. Bouldin DW
      . A cluster separation measure. IEEE Trans Pattern Anal Mach Intell 1979;1:224–7.
      OpenUrlCrossRefPubMedWeb of Science
    21. ↵
      1. Calinski T,
      2. Harabasz J
      . A Dendrite method foe cluster analysis. Comm in Stats - Theory & Methods 1974;3:1–27. doi:10.1080/03610927408827101
      OpenUrl
    22. ↵
      1. Rousseeuw PJ
      . Silhouettes: a graphical aid to the interpretation and validation of cluster analysis. J Comput Appl Math 1987;20:53–65. doi:10.1016/0377-0427(87)90125-7
      OpenUrlCrossRefPubMedWeb of Science
    23. ↵
      1. Tibshirani R,
      2. Walther G,
      3. Hastie T
      . Estimating the number of clusters in a data set via the gap statistic. J R Stat Soc B Stat 2001;63:411–23. doi:10.1111/1467-9868.00293
      OpenUrl
    24. ↵
      1. Charrad M,
      2. Ghazzali N,
      3. Boiteau V, et al
      . Nbclust: an R package for determining the relevant number of clusters in a data set. J Stat Softw 2014;61:1–36. doi:10.18637/jss.v061.i06
      OpenUrlCrossRefPubMed
    25. ↵
      1. Reading Turchioe M,
      2. Mangal S,
      3. Ancker JS, et al
      . "Replace uncertainty with information": shared decision-making and decision quality surrounding catheter ablation for atrial fibrillation. Eur J Cardiovasc Nurs 2023;22:430–40. doi:10.1093/eurjcn/zvac078
      OpenUrl
    26. ↵
      1. Pakhomov SV,
      2. Jacobsen SJ,
      3. Chute CG, et al
      . Agreement between patient-reported symptoms and their documentation in the medical record. Am J Manag Care 2008;14:530–9.
      OpenUrlPubMedWeb of Science
    27. ↵
      1. Hermans ANL,
      2. Gawalko M,
      3. Slegers DPJ, et al
      . Mobile app-based symptom-rhythm correlation assessment in patients with persistent atrial fibrillation. Int J Cardiol 2022;367:29–37. doi:10.1016/j.ijcard.2022.08.021
      OpenUrl
    28. ↵
      1. Shurrab M,
      2. Langer A,
      3. Crystal E, et al
      . Symptom/rhythm correlation with patient owned device: insights into practice and challenges. J Atr Fibrillation 2016;9:1390. doi:10.4022/jafib.1390
      OpenUrl
    29. ↵
      1. Newham WG,
      2. Tayebjee MH
      . Excellent symptom rhythm correlation in patients with palpitations using a novel smartphone based event recorder. J Atr Fibrillation 2017;10:1514. doi:10.4022/jafib.1514
      OpenUrl
    30. ↵
      1. McCabe PJ,
      2. Schumacher K,
      3. Barnason SA
      . Living with atrial fibrillation: a qualitative study. J Cardiovasc Nurs 2011;26:336–44. doi:10.1097/JCN.0b013e31820019b9
      OpenUrlCrossRefPubMed
    31. ↵
      1. Freeman JV,
      2. Simon DN,
      3. Go AS, et al
      . Association between atrial fibrillation symptoms, quality of life, and patient outcomes: results from the outcomes registry for better informed treatment of atrial fibrillation (ORBIT-AF). Circ Cardiovasc Qual Outcomes 2015;8:393–402. doi:10.1161/CIRCOUTCOMES.114.001303
      OpenUrlAbstract/FREE Full Text
    32. ↵
      1. Vermond RA,
      2. Crijns H,
      3. Tijssen JGP, et al
      . Symptom severity is associated with cardiovascular outcome in patients with permanent atrial fibrillation in the RACE II study. Europace 2014;16:1417–25. doi:10.1093/europace/euu151
      OpenUrlCrossRefPubMed
    33. ↵
      1. Xu H,
      2. Mohamed M,
      3. Flannery M, et al
      . An unsupervised machine learning approach to evaluating the association of symptom clusters with adverse outcomes among older adults with advanced cancer: a secondary analysis of a randomized clinical trial. JAMA Netw Open 2023;6:e234198. doi:10.1001/jamanetworkopen.2023.4198
    34. ↵
      1. Song J,
      2. Chae S,
      3. Bowles KH, et al
      . The identification of clusters of risk factors and their association with hospitalizations or emergency department visits in home health care. J Adv Nurs 2023;79:593–604. doi:10.1111/jan.15498
      OpenUrl
    35. ↵
      1. Lee CS,
      2. Gelow JM,
      3. Bidwell JT, et al
      . Blunted responses to heart failure symptoms in adults with mild cognitive dysfunction. J Cardiovasc Nurs 2013;28:534–40. doi:10.1097/JCN.0b013e31826620fa
      OpenUrlCrossRefPubMed
    36. ↵
      1. Marrouche NF,
      2. Kheirkhahan M,
      3. Brachmann J
      . Catheter ablation for atrial fibrillation with heart failure. N Engl J Med 2018;379:492. doi:10.1056/NEJMc1806519
    37. ↵
      1. Efremidis M,
      2. Letsas KP,
      3. Lioni L, et al
      . Association of quality of life, anxiety, and depression with left atrial ablation outcomes. Pacing Clin Electrophysiol 2014;37:703–11. doi:10.1111/pace.12420
      OpenUrlCrossRefPubMed
      1. Lip GYH,
      2. Nieuwlaat R,
      3. Pisters R, et al
      . Refining clinical risk stratification for predicting stroke and thromboembolism in atrial fibrillation using a novel risk factor-based approach. Chest 2010;137:263–72. doi:10.1378/chest.09-1584
      OpenUrlCrossRefPubMedWeb of Science

    Supplementary materials

    • Supplementary Data

      This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Footnotes

    • Contributors MRT obtained funding for this project. MRT, YZ and MH conceived of the idea and study design. YZ conducted all analyses. All authors provided critical feedback on the data analysis procedures. MH drafted the manuscript and revised it based on critical feedback from all remaining authors. MRT is responsible for the overall content as guarantor.

    • Funding This work was supported by the National Institute of Nursing Research (NINR) (T32NR007969 [MH, DS], R00NR019124 [MRT]) and the Jonas Scholarship (MH).

    • Competing interests MRT: Iris OB Health Inc., New York (co-founder/equity); Boston Scientific Corporation (consulting).

    • Provenance and peer review Not commissioned; externally peer reviewed.

    • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.