Research-In-Progress

Anish Sanjay Shah, MD/MS
Cardiology Fellow
University of Illinois Chicago
Task Deadline
Epidemiology (NA)
AFEQT and SDOH Manuscript revisions from co-authors 2024-06-06
Arrhythmia (NA)
AFGen Fellowship ECG/genetics validation cohort proposal 2024-06-06
AFGen Fellowship Bimonthly meetings with AFGen Consortium 2024-06-30
ECG and Genetics Training to classify structural and ion-channel based variants 2024-05-30
ECG and Genetics Training to classify PGS from ECG directly 2024-05-30
Phenotype AF Generation of clinical features of patients with paroxysmal AF 2024-06-01
Clinical (NA)
Clinic JBVA FIT Clinic 2024-06-30
EP Lab Working with EP faculty (Avitall/Johnson/Ringwala) 2024-06-30
Table 1: Progress updates

June 24, 2024

Updates

  1. AFEQT/SDOH draft manuscript has been revised, pending tentative feedback from EJB
  2. Cluster access
    1. Time this Thursday/Friday for cluster introduction (pending everyone getting access)
  3. UK Biobank proposal (to be shared later this week) overlaps with Michael Reese Foundation award for Aim #2, will share proposal if in agreement about aims
    1. Aim 1 = WES expansion
    2. Aim 2 = ECG-based genetic phenotyping
    3. Aim 3 = ECG ← EP correlation
  4. AF-CTRS updates (later this week)

June 10, 2024

Updates

  1. AFEQT/SDOH draft manuscript revisions
    • Potential journals: Annals of Internal Medicine, JAMA IM, Circulation: Quality/Outcomes, JACC EP
  2. Cluster access requested!
  3. AHA abstract + Early Career Award application (both extended to June 10 and June 13 respectively)
  4. AF-CTRS
    • Aim to have generic weights in ~2 weeks
  5. Future collaborations/projects

Research products

  • 4 manuscripts (1 published, 1 in preparation, 2 under review)
  • 4 co-authored manuscripts (published)
  • 3 software packages published, peer-reviewed
  • Completion of an NIH-F32 award
  • K23 aims with 50-60% preliminary data collected
  • AFGen fellowship
  • ECG/genetics with UIC AF Registry + UK Biobank project/proposal

June 3, 2024

Updates

  1. AFEQT/SDOH draft pending feedback from Alvaro/Emelia
  2. Cluster access has been set-up for the lab
  3. AHA application for Early Career Award
  4. AF-CTRS weights on Crossover Study dataset + AFEQT dataset
  5. UK Biobank proposal/access tentatively approved (still waiting for feedback from AFGen Consortium)
    • Structural/sarcomeric genes
    • Ion channel genes
    • Metabolism genes
    • Immune/inflammation-related genes
  6. Future meetings will likely be virtual

LMNA/ECG

Premises:

  • All LMNA patients will develop AF, but contribution of genetics to how AF develops is unknown (prior to or after cardiomyopathy)

Cases/controls:

  • sinus beat in LMNA patient prior to AF development
  • sinus beat in LMNA patient after AF development
  • sinus beats in AF patients wild-type, age-matched
  • sinus beats in wild-type patients without AF, age-matched

Power:

  • Likely between 30-60 patients, with another 10-15 for validation although can likely get adequate power with lower number

May 20, 2024

Updates

May 15, 2024

Updates

AFEQT/SDOH:

SDOH research:

ECG/Genetics:

Logistics

  1. Meet with ACER initially to set-up shared workspaces for the lab, and to provide cost estimates for compute time.
  2. Discuss need for PHI-safe data storage on cluster/remote server for combination of all clinical data and ECG data
  3. UIC/UIH access…
    1. Complete ECG-based phenotype/genotype preliminary data for K23/CDA
    2. Manage ECG-dataset (n = 800,000)
    3. Update both CCTS/MUSE datasets
    4. Pipeline for data pull requests (MRN, code, diagnosis, etc)
  4. Timeline for research updates for SS and DS

May 6, 2024

Updates

  1. AFEQT/SDOH: manuscript, pending initial co-author review. Meeting with Alvaro Alonso on 5/17 to discuss informally. Then likely send to additional senior authors.
  2. ECG/genetics: validation cohort proposal completed
  3. AF-CTRS: Have drafted initial scoring algorithm. Need to trial basic weights on a dataset.
  4. CV research data: Meeting with ACER to help host a safe cloud environment for future data pulls.

April 29, 2024

Updates

  1. AFEQT paper to be disseminated to co-authors by Wednesday 5/1
  2. ECG/genetics: UK Biobank Proposal to be shared to mentors by 5/3
  3. K23 aims: draft to share with key mentors on 5/6
  4. AF-CTRS: writing out different weighting options currently
  5. LLM: upcoming meetings on 5/3
  6. AF phenotyping/trajectories: pending application/integration of DS and SS coding projects into our CCTS dataset
  7. Cluster access for data - need to discuss lab-wide access / data maintenance (low priority for now)

April 26, 2024

Updates

AFEQT paper:

ECG/genetics:

April 22, 2024

Tasks

AF-CTRS

flowchart LR
    score[AF Composite Treatment Response]
    burden[Frequency/duration of AF]
    sx[Symptoms]
    ecg[ECG]
    mct[Continuous telemetry]
    afeqt[AFEQT score]
    fc[Functional class]
    
    sx --> score
    burden --> score
    
    fc --> sx
    afeqt --> sx
    
    ecg --> burden
    mct --> burden

AF-CTRS draft

  • \(ECG = ECG_{sinus} \times n_{sequential}\)
  • \(ECG = \frac{ECG_{sinus} - ECG_{AF}}{ECG_{total}}\)
  • \(MCT = Burden_{sinus} \times days\)
  • \(AFEQT = AFEQT_{total} / 10\)
  • \(FC = 5 - NYHA\)

A “true” model would need to impute missing data, but this would require multiple datasets for training/testing/validation

April 8, 2024

Updates

  • ECG/Genetic Variants
    • Re-analyze/label sarcomeric/ion channel/“wild-type” AF WES dataset
    • AHA submission for abstract
    • Pending validation cohort
  • AFEQT/SDOH
    • Table 1: Baseline characteristics of full cohort (longitudinal \(\rightarrow\) supplement)
    • Table 2: AFEQT change by SDOH (NDI and ethnicity)
    • Table 3: AFEQT change predictors (SDOH)
    • Table 4: Rhythm/rate control strategy predictors
    • Figure 1: Study flow-chart
    • Figure 2: AFEQT predictive means over time (by cluster)
    • Pending data collection
  • AF-CTRS Score
    • Meeting with MB and JS

April 1, 2024

Updates

  • AFEQT/SDOH
    • Additional analysis on cross-sectional cohort
    • Conditional/clustered analyses for predictive means
  • Atrial Fibrillation Composite Treatment Response Score (AF-CTRS)
    • AF-CTRS to be used to predict efficacy of treatment (regardless of burden % or AFEQT score)
    • Output is algorithm that ties “response parameters” to a dose-responsive score \(\rightarrow\) validate in AF Registry
    • Algorithm: ECG findings, AFEQT score, clinical parameters, MCT/ILR (require complex MI)

ECG Prediction of Structural and Ion Channel Mutants in AF

This is an expansion of the ECG/TTN project, which found ECG could feasibly provide above-chance accuracy in identify TTN LOF variants. Now, expand to additional wild-types and mutants…

  • Sarcomeric/structural variants (e.g. TTN, PITX2, LMNA, NUP155, GJA1/5)
  • Ionic channel variants (e.g. KCNE1-5, KCNQ1, SCN5A)
  • No known genetic contributors (excluding any patient with non-benign VUS in any above genes \(\pm\) SNPs associated with polygenic risk for AF)

Models, trained on ECG in sinus rhythm in those with known AF, will need validation. However, should publication happen before or after validation (e.g. with TOPMed/UK Biobank)

March 29, 2024

Update

  • AFL/FH:
    • Revisions complete, pending table placement considerations
  • ECG/TTN:
    • Results presented to AFGen fellows
    • Revised project proposal in conjunction with AFGen (for UK Biobank)
    • Potential K23 aim
  • AFEQT/SDOH:
    • Additional analysis on cross-sectional cohort
    • Conditional/clustered analyses for predictive means

ECG/TTN

Can ECG discriminate between wild-type AF and mutant-type with known, causal/associated genes?

  • Increase scope to include additional monogenic genes
  • Train to target individual cases based on genetic variant
  • Train to classify those with no genetic mutations
  • Validate in UK Biobank

March 25, 2024

Updates

  • ECG/TTN
    • Draft analysis complete, presenting to AFGen fellows on 3/27
    • Brainstorm scope of paper
  • AFEQT/SDOH
    • Positive feedback from co-author on results draft
    • Potential expansion of “baseline” cohort (AFEQT)
    • Paper writing (new draft in progress), expect 2 weeks for first draft \(\pm\) new analyses
      AFL/FMHX
    • Revisions in progress, responses likely Thursday/Friday

February 26, 2024

Updates

  • AFEQT/SDOH
    • data cleaning started on JBVA data, largest hurdle is getting NDI data from addresses
    • have rough draft of analysis plan, with plans to get early feedback from Alvaro (standing meeting next week)
  • ECG/TTN
    • Filtering pipeline effective for TTN variant, will review with Choi (have 8-10 P/LP variants that are potentially LOF)
    • Presentation on 3/27 with AFGen group
  • K23
    • Feedback from Benjamin: after refining genetics aim, can consider implementation in UK Biobank
    • Feedback from Ranjan: need to consider the sponsorship/mentorship committee at Utah (e.g. Martin Tristani-Firouzi, Rob McLeod, Jared Bunch, etc)

AFEQT/SDOH paper proposal

  1. Table = cohort description by baseline AFEQT scores, with sections on clinical covariates, social determinants, etc
  2. Table = Cohort description with statistical testing by AFEQT score changes
  3. Figure = AFEQT score breakdown in the population, with subscales (bar graph), showing score changes
  4. Table = regression model for SDOH factors and AFEQT change
  5. Table = mediation analysis for SDOH factors, treatment strategies, and AFEQT scores
  6. Figure = forest plot of interaction of the highest yield SDOH marker and AFEQT score changes

February 12, 2024

Whole Exome Sequencing

In light of project-focus on deleterious TTN variants and a/w ECG features, needed a statistically appropriate method for variant selection.

  • Using variant effect prediction software, including loss-of-function predction, Polyphen, SIFT, and phenotypic data, created software pipeline to analysis WES
  • Each WES VCF, roughly 50-100 Mb in size, can be annotated to 500-800 Mb of annotations based on dozens of classifier models (called consequences)
  • These consequence additions are part of Choi’s recommendation for having agreement in >80% of predictions to serve as deleterious variant “ground truth”
  • Need to manually adjudicate the TTN variants, thus far is 1) reproducible with different prediction cut-offs, 2) identifies some miRNA sequenes that may be consequential, 3) has high degree of confidence in the LOF and INDEL variants

Tip

Variant pipeline requires combination of cluster + local software, and is stable for running new variants PRN.

AFEQT/SDOH

  1. VA data cleaning, expected 10-12 hours of cleaning
  2. Results and tables to be created throughout this week, pending VA data
  3. Need to decide on key figures and tables for the paper
  4. Expect manuscript draft (dependent on results) week of 2/23

January 29, 2024

Updates

  • AFEQT/SDOH: CDW data pull pending, may be able to get all ICD codes \(\pm\) medicaions
  • AFL/FH: Completing revisions to paper, sharing in next 2-3 days
  • ECG/TTN: focusing on variant pipeline currently

January 21, 2024

ECG/TTN Project

  • Variant annotation pipeline drafted, with classification based on series of annotation tools used in parallel
  • AFGen fellowship: suggested minimum threshold of >80% of the ~30 annotation system to be in agreement for tentative “ground truth” for the “severe” or LOF variants
  • Algorithm for mis-sense variants being developed

ECG/TTN Project

  • Have developed an alpha version of end-to-end pipeline for sinus beat analysis
  • Utilizing time varying machine learning layers to add power for repeat samples and time-to samples
  • UK Biobank may be a fair option for validation if race/ethnicity factors isn’t a major concern

January 19, 2024

AFL/FH Revisions

  • Reviewer #1: No comments as had no initial comments or suggestions on initial submission
  • Reviewer #2: Complimentary of revisions and thought the concerns had been satisfactorily addressed
  • Reviewer #3: Thankful for thoughtful and detailed responses to questions
  • EDI Reviewer: Thought comments had been appropriately addressed

Important

A statistical reviewer has entered the fray

Statistical Review

  • “The study is described in the abstract as a prospective, observational cohort but it seems to be a cross-sectional analysis. Can the authors please clarify the design.”
  • “The abstract also notes that the primary independent variable is family history of atrial fibrillation in first degree relatives, and that secondary independent variables include self-reported race-ethnicity and sex. The statistical analysis section indicates that the secondary exposure includes race-ethnicity, sex, and potential modifiable risk factors. The title suggests that interest lies in the association between family history and early onset atrial flutter and that race/ethnicity might be a potential confounder as the title suggests exploring the association in racial/ethnic subgroups. Nothing is mentioned about sex. The focus of the paper needs to be made very clear throughout.”
  • “The statistical analysis section is confusing.”
  • “The section indicates that multivariate (should be multivariable) logistic regression analysis was used with sequential adjustment for sex, race, smoking, etc. Sex and race should not be listed as potential confounders (per point above – please clarify the role of these variables in the analysis).”
  • “The analysis in Tables 2 are unadjusted but very problematic given the 10-year age difference in individuals with early- versus late-onset atrial flutter. These should be, at a minimum, adjusted for age.”
  • “The statistical analysis section indicates the use of multivariable logistic regression, mixed effects models and interaction models – but the analysis presented are almost all unadjusted.”
  • “Table 3 stratifies by family history. Why stratify by the primary independent variable? The sample size is also too small in the family history group.”
  • “The statistical analysis approach needs to be better organized to address the key questions.”
  • “In the abstract, please clarify that the odds ratios that are reported are all unadjusted, if this is the case.”
  • “Throughout, the authors refer to effect size when they mean effect – it might even be clearer to refer to odds ratios rather than effects here.”
  • “The language in the clinical implications section needs to be toned down as it infers causality – e.g., individuals with a family history and individuals who self-identify as Black had higher odds.”
  • “Lastly, I am not sure the Figure is needed.”

Statistical Response

January 8, 2024

AFEQT/SDOH background

  • Currently there is a push for social determinant research to be integrated into atrial fibrillation research (Benjamin et al. 2023)
  • Social determinants in AF research has been predominantly oriented at diagnosis and clinical outcomes (Essien et al. 2021)
  • Neither race, ethnicity, or socioeconomic differences have significant research progress in AF diagnosis or management [Norby2021]

We have a rare opportunity to evaluate how symptom burden is affected by SDOH factors in a multi-ethnic population.

December 18, 2023

AFEQT/SDOH updates

December 11, 2023

ECG & TTN Variant Project

Background: VUS/LP/P TTN variants are associated with increased incidence of EOAF. The mechanism is not fully understood.

Aims: Identify if VUS/LP/P TTN variants may cause structural atrial changes (an atrial myopathy) that is identifiable or ECG, or other related conduction changes. We aim to identify if VUS/LP/P TTN variants can be classified on ECG using signal processing techniques

Approach: Utilize sinus rhythm ECGs from patients with WES/CGS with paroxysmal AF. Develop beat-by-beat algorithm to identify features that may be associated with VUS/LP/P TTN variants.

ECG data

  • n = 298 patients were available and had ECG in sinus rhythm at some point from 2010 to 2023
  • This amounted to n = 7953 unique ECG that were thought to be in sinus rhythm
  • Using a wavelet-decomposition approach, identified morphology of P wave, QRS complex, and T wave and created single-beat structures
  • Using these n = 51493 beats, there were n = 6202 cases (which represents n = 36 individuals with potential pathogenic TTN variants (based on SIFT/PolyPhen scores)

Model

  • Sequential convoluted neural network approach
  • Masking layers to control for 0 values (representing non-depolarized intervals and padded sequences)
  • Use combination of relu (rectified linear activation) and increasing dense layers (with signal drop out)
  • Categorize using a softmax approach for classification between case/control
  • Initial training/testing to first-beat-only approach to avoid sample bias
  • Trained for 10 epochs based on model accuracy and minimizing loss on each repetition

Results

Training

  • Training accuracy was 88% with only 2% loss after total of 10 steps

Testing

  • Testing accuracy was 85% with 13% loss
  • Example: single beat (control) may have a confidence of 75% in being a control

First 10 epochs of training

Next steps

  1. Confirm the TTN variant cases, trialing different definitions of pathological beats
  2. Extend analysis to full data-set
  3. Modify approaches in terms of machine learning “layers”

December 4, 2023

HRS abstract proposals

  • TTN variant prediction: Would like to submit abstract, currently training an ECG-based model to classify variants (pathogenic vs. controls). Training data of approximately 50-60k sinus beats, with 500-700 “pathogenic TTN cases”. Data augmented by breaking ECGs into individual beats
  • AFL & Family History: Abstract to be submitted, with equivalent co-authors as manuscript submission
  • TTN variants and EP outcomes: Gursuk/James and others wanted to pursue EP-based outcomes. Have ~9 cases of PVI performed on TTN variants that are non-benign (e.g. VUS or LP/P).

November 20, 2023

AFL/FH updates

Why do we believe that Black v. Non-Black individuals had higher rates of AFL despite balanced risk factors?

AF Registry and CCTS Data

  1. Have all EMR data from 2010 to 2023 now, which increased file sizes by 30% (now over 70 Gb), and increased n = 140k to n = 210k
  2. Have all ECG for same patient population (approximately 40 Gb in XML format)
  3. Created parquet format to utilize Apache Arrow for I/O

AFEQT/SDOH

Which key variables should be included for evaluation? Some of these have been studied, but not in context of AFEQT/symptom burden

  • ECG factors at baseline compared to at time of repeat evaluation (heart rate, rhythm)
  • Maintenance of sinus rhythm (percent burden of AF, frequency of ECG in sinus rhythm)
  • Medication changes or intensification (addition of anti-arrhythmic)
  • Referral to cardioversion or ablation
  • Administration of anticoagulation therapy

ECG/TTN

Timeline for abstract submission to HRS (12/15):

  • Complete initial model training on pathogenic TTN cases
  • Exclude other TTN variants from analysis to have cleaner controls
  • Identify appropriate metrics (e.g. AUC, net classification, etc)
  • Identify if segments of ECG are more highly weighted (variance importance)

November 13, 2023

Updates

October 26, 2023

AFL/FH Simple Revisions

  • Conform with AHA guidelines on reporting health differences by races
  • Including Clinical Perspective section
  • More detailed methods on patient selection and recruitment
  • Directionality when discussing associations (e.g. weight versus increased weight)
  • Add definitions for social drinking
  • Clarify anti-hypertensive agents and selection
  • Clarify results and statistical methods to match discussion
  • Requesting how and why we defined EOAFL as age < 65

AFL/FH Complex Revisions

  1. Concern is the definition of family history to include AF, HF, and CVA, as is inconsistent with the idea of EOAFL as a separate entity. Unsure how to respond
  2. Can occult AF be analyzed as a comparative group? Unsure about an additional dataset. However would strengthen the paper if we had a matched cohort as sensitivity analysis.
  3. In those without stated FH, the FDR were not contacted - reviewers are concerned that frequently patients are unaware of FH, and recommended contacting every study participant. Two paths: 1) Ignore this and state we had a rigorous approach to FH questionnaires; 2) Pursue the ~300 patients with EOAFL and contact FDR to limit confounding.
  4. Provide a hypothesis and reasoning for the choice of variables. Particularly that of race. Would refer to genetics and poor representation in other studies.
  5. Concern for why we emphasized difference between Black race and White/Latinx race. Would compare Black to non-Black instead
  6. Hypothesis on inverse relationsihp between HTN/BB/CCB and EOAFL risk

October 23, 2023

Updates

AFEQT/SDOH: Current focus is on obtaining updated AFEQT, led by JS, with goal to complete AFEQT in next several weeks. In parallel, using EMR-based records to evaluate clinical outcomes. It will take roughly 30 hours of statistical work to incorporate the clinical and AFEQT data (roughly 1-2 weeks), once the AFEQTs are done.

AFL/FH: Our good news was that JAHA is willing to review our paper again with revisions. A majority of these will likely be completed within the next 4-5 days (deadline = 10/31) but have to decide on if new/additional FH data collection is needed. The most challenging critique was the potential bias in FH reporting during initial screening. To overcome this, the most thorough option would be to obtain FH data from every patient to avoid systemic errors/bias. However, a limited approach of verifying FH in just the EOAFL may be reasonable. Would like everyone’s thoughts here.

Phenotype AF: Two students have joined this effort, SS and SD, who will be working on initially on the DSP and ML aspects. Current priority is to finish training the models using the TTN data. Unfortunately, this has been delayed with Boards and interviews, but hope to speed this back up in the next 1-2 weeks.

AF/PRS: For our evaluation of the PRS in AF in ethnic cohorts, we are meeting with the eMERGE group (NW). We will need to understand if the PRS can be validated in our UIC cohort, and if there is an existing mechanism to integrate a GIRA with our own EHR.

September 18, 2023

EMR Data

This has been somewhat streamlined to extract data from the CCTS data pull, with full cohort results returned from CCTS.

  1. Need to extract AF registry patients and add SDOH, medication intensification, and procedure reports.
  2. Working with Tofovic/Nischal to extract echocardiogram data
  3. Extracted roughly 8k ECG, digitized, and segmented to individual beats, which provides roughly 40k samples for testing amongst the ~300 WES patients
  4. Working with Chen to complete power calculations based on our population and current literature

AFEQT/SDOH

  • Spoke with Jerry/Muriel about re-organizing the AFEQT data for manuscript
  • Will need to complete roughly n=600 follow-up AFEQT
  • Power calculation suggest would need n=200 Hispanic/Latinx
  • Timeline TBD (expect ~ 4-6 weeks)
  • Subsequently will need to extract patient outcomes (intensification, burden) from CCTS pull
  • Adding Angela Hussain to data collection as student researcher
  • Check in with Annette to help with Spanish-speaking requirements

AF/Genetics

Met with AFGen group (Patrick Elinor, Emilia Benjamin), encouraged computational approach in genetics. Currently working to filter and annotate our WES dataset, but expanding to novel gene regions (e.g. neurocardiac receptors) that are not included in general cardiomyopathy/arrhythmia panels.

  1. Variant evaluation/assessment of different reference genomes
  2. Variant filtering based on SIFT/PolyPhen
  3. Annotation based on multiple sources (USCS, NCBI, etc)
  4. Inclusion of broader, related genes (e.g. ADRB2 not included in recent evaluation)

September 11, 2023

AFEQT

Current manuscript has n=453 individuals split amongst races, with some strength of assocation b/w SES/SDOH and ∆AFEQT.

  • Have n=760 baseline AFEQT data points, but need roughly n=600 additional follow-up AFEQT scores
  • Have n=1506 currently in AF Registry, thus 743 more patients that can be added
  • Would allow us to be powered for mediation analysis, and stratified analyses
  • Able to incorporate intensification and recurrence rates now, as well as census tract (vs. zip) and language barriers now

Next:

  1. Complete the additional 600 AFEQT follow-ups
  2. Extract data from CCTS to “flush out” registry data
  3. More specific/appropriate analyses for SDOH including mediation

PRS for AF/eMERGE

  1. We are meeting with NW and the eMERGE group to validate PRS in AA/HL population here at UIC this week.
  2. Met with Hua Yua Chen to go over PRS power calculations
  3. Generating multiple sample size/power calculations based on C-statistics seen in other AF/PRS papers → however may not apply equivalently in AA/HL population as primarily derived from Caucasian populations

September 8, 2023

AF/Genetics

Completed computational genetics coursework on variant analysis. Although some work has already been done on this, going through the basics will allow us to have in-house variant annotation pipeline

  1. Uploaded our ~2G of patients to cluster
  2. Generated functions to split VCF by individual case
  3. Identified variant locations

Next: Evaluate variants using PolyPhen and SIFT databases

ECG/pAF

Recently was asked to review/adjudicate ECG for the PhysioNet challenge on AF prediction algorithms. In line with my interest at phenotying AF based on ECG markers.

  1. Segmentation of individual beats has been automated
  2. Augmenting the data to individual beats (similar to van de Leur et al. (2021))

Next: Training on labeled AF patients with TTN variants

Figure 1: Annotation of individual beats

AF/SDOH

  • eMERGE network meeting this week
  • Processing UIC/CCTS data
  • Uploaded January to June ECG data from MUSE
  • Will still need updates from CCTS for 2023 remaining clinical data
  • Have to solidify definitions for therapy intensification and AF outcomes

August 21, 2023

Phenotyping AF

Identifying phenotypes of AF that are at risk of progression (e.g. developing higher burden, persistent AF, TICM, etc)

  1. Extracted clinical covariates for ~6k patients, stratified by time/onset (this is ~7 million diagnoses)
  2. Extracted medications by time/duration of prescription
    1. Filtered down to all cardiovascular medications
    2. Classified medications that are relevant to AF (AC/AP, AAD)
    3. 2.5 million prescriptions were analyzed
  3. Evaluated ECG related to this population and initiated segmentation/feature extraction
  4. Goals: Pull procedure reports to assess burden, and process text-based echocardiography data

ECG and Genetic Variants

  1. Completed the software needed to annotate ECGs on single leads
  2. Will have to evaluate efficacy amongst 12-lead systems (e.g. 5000 x 12 array)
  3. Using semantic segmentation approach to classify waveforms
  4. Have started cluster-based pipeline development for re-calling VCF to identify variants based on…
    1. Multiple different population panels (different allele frequencies)
    2. Filter on read depth and quality
    3. Will need to learn about other filtering approaches

SDOH/AF/HSR

  • Current knowledge gap is interaction between race/ethnicity and socioeconomic status in management of AF (Essien et al. 2021)
  • Similar to AF Registry + Cluster, can use the definition of treatment intensification as part of health care differences
  • Parent project to AF/language
  • Unlike other cohorts (e.g. ARIC had n=3200 but only n=500 were Black), have more diverse population pool

August 14, 2023

AFL/FH

Completed revisions to paper with tentative plans for re-submission.

  1. Called n=31 families to confirm AFL/AF diagnoses and clarify CHF and CVA histories
  2. Re-analyzed pedigrees with multiple definitions of +FH in FDRs as a sensitivity analysis
  3. Major findings remain same (and robust to adjustment):
    • +FH doubles risk for “premature” or earlier onset of AFL
    • EOAFL as corrolary increases likelihood of AF and AFL in FDRs
    • Black race a/w EOAFL, partially mediated through OSA/ETOH (most prominent in this subgroup)
  4. Resubmission to… JAHA, AHJ, IJC, BMJ Heart

AF/SDOH

  • n=4537 patients (still missing updated CCTS pull of about 2k)
  • 92% have NDI via census tract
  • 52% have social histories completed
  • 81% are English speaking, 10% Spanish
  • Medications/Labs/Vitals/Visits/Comorbidities being evaluated currently

AF Phenotyping

Figure 2: P wave detection

Working on ECG beat segmentation and annotation using machine learning approach across multiple leads. Will be applicable across ECG-based projects. Necessary step prior to creating median beats.

July 31, 2023

AFL/FH Reviewer Concerns

  1. How many FDR were directly contacted?
  2. What constitutes an “atrial arrhythmia” in family members? Often asymptomatic. Needs clear definition
  3. Why LOAFL versus EOAFL, instead of EOAFL versus controls?
  4. Why is FH/Race significant only in a single racial group?
  5. Overstatement of importance of FH over 4q25 papers in isolated AFL
  6. How are CVA events labeled as cardioembolic?
  7. Why include CHF as part of arrhythmia risk factors in FDR?
  8. Define excess ETOH usage
  9. Pedigree needs to be labeled more specifically/informatively

July 24, 2023

Active projects

  1. AFL/FH: Revisions to be addressed in next draft. Will identify additional journals to target.
  2. HRV/CVD: EHJ
  3. ECG/PAF: Active project involving computational approach at identifying sub-phenotypes of paroxysmal AF based on ECG-driven parameters with subsequent work on clinical patterns and trajectory

Phenotyping AF

Project is a “big data” project using over clinical data from 2010 to 2023, supplementing the AF Registry with data mining and machine learning approaches.

Composed of three elements:

  1. ECG data in form 12-leads of 10 second signals, sampled at 500 Hz from 2010 to now
  2. Clinical structured documentation including vitals, medications, labs, echo/study reports from 2010 to now
  3. Clinical unstructured data from MD documentation

Data overview

ECG data:

  • 500k ECGs collected from 2010 to 2013
  • 70 Gb of raw XML data (converts to ~50 Gb of WFDB)

Raw clinical data:

  • 43 Gb raw data (70% is from clinical notes alone)
  • 145k unique patients with CVD diagnoses (cardiomyopathy, arrhythmia, vascular dz, etc)

AF-specific data

ECG data:

  • 33k ECG with a diagnosis of an atrial arrhythmia
  • 6k unique patients with a diagnosis of AF confirmed by ECG
  • Of those 6k patients, 95k ECG taken in total (both sinus + AF)

Clinical data:

  • Of the 6k patients c- AF on ECG, 3.9k have clinical data available from CCTS
  • Aggregate file size for this 3.9k is < 1 Gb in total
  • 600k ICD diagnoses, 4.5 million lab results, 500k visits, 1.5 million vital sign measurements, 2.9 million medications

Project status

Currently have working pipeline for extracting relevant diagnoses, medications, labs, etc, for additional projects. Can email ashah282@uic.edu for questions or project proposals.

Can see current proposal here

June 12, 2023

Backbone for aims

  1. Sympathetically-mediated AF occurs predominantly earlier in paroxysmal AF, in hearts with minimal scar burden.
  2. To cause episodes of AF would require abnormalities in conduction from both intracardiac and extracardiac causes
  3. Abnormalities from both intracardiac and extracardiac causes are either acquired or genetic
  4. These factors may also lead to different rates of progression of AF

Question: Do certain genetic variants contribute to the progression of paroxysmal AF to persistent AF?

Trajectory of pAF

  • Kerr et al. (2005) from a Canadian Registry followed n = 757 for ~8 years, 5 year recurrence was >60%
    • Positive predictors = age, +AS, +MR, LAE, reduced LVEF
    • Negative predictor = rapid heart rate during AF
  • Post catheter ablation, 5 years recurrence was 60% (1st ablation), and 20% after (>2 ablations) (Takigawa et al. (2014))
    • Predictors = LAVI and patient age
  • Holmqvist et al. (2015) noted progression based on age + increased heart rate in sinus (CHA2DS2VASc had minimal discriminatory power)
  • Andrade et al. (2023) evaluated 3 year follow in ablation versus AAD, found HR = 0.51
  • Acutely, post-cardiac surgery AF can be predicted by decreased HRV (Veselá et al. (2023))

Predictors of pAF progression

CUrrently progression is poorly defined…

  • Clinical: need for rhythm control intervention (ECV, PVI), hospitalization for symptoms or heart failure, % burden increase by an inflection point
  • ECG: Decrease HF-HRV, increased P wave duration/dispersion, decreased P wave amplitude (during AF, decreased F-wave amplitude)

Persistent = n continous ECG in AF, diagnosis by cardiology, >7 day duration of episodes (or less)

Approach for evaluation

  1. Establish clinical progression of disease with multiple versions of the definition
  2. Obtain ECG data (segmentation to evaluate sinus beats)
  3. Assess genetic variants ↔︎ ECG characteristics to compare with outcomes

Figure 3: Manhattan plot Verweij et al. (2020)

Figure 4: Cluster of genetic variants by ECG component Verweij et al. (2020)

Figure 5: Individual variants with ECG components Verweij et al. (2020)

Next steps

June 5, 2023

Aims Progress

Goal: Identify characteristics that define sympathetically-mediated paroxysmal AF (as compared to scar-mediated pAF) in trajectory of paroxysmal → persistent

  • Pending mentorship meeting (dates requested for Alvaro, Amit, Dawood)
  • Met with Lampert - be much more specific in ANS factors in pAF
    1. Evaluate ECG during AF: coarse AF v. fine AF → triggering event (ADHF, surgery, etc…)
    2. Evaluate ECG during sinus: P wave morphology/area/dispersion and AF burden
    3. Evaluate candidate genes (ANS neurohormonal pathways) in Emory Cohort (which has high vagolysis burden) as initial study; validate these genes in AF registry with burden/rate of AF progression

AF Phenotyping

  • Pending CCTS to upload data (hundreds of GB in size)
  • Complete setting up cluster computational pipeline for NLP of unstructured text
  • Specific goal is to identify timepoints: onset of AF, transition to persistent AF
  • Initiate analysis of ECG data
    1. Convert to annotation format (WFDB)
    2. Identify median P waves per ECG

May 22, 2023

ECG analysis

K23 AIM #1 is to identify paroxysmal AF into low-scar burden (and high vagolytic triggers) in AF registry.

May 12, 2023

Updates

  1. ECG data is being pulled from MUSE in XML format
  2. CCTS data pull to be completed today
  3. EGM signal processing - work in progress with goal of automated beat generation of multi-channel leads (R/C++ package)

May 8, 2023

Goals

  1. Phenotyping of AF, as current grouping is inadequate. Focus on risk category changing from paroxysmal to persistent (baseline concept of multi-wavelet reentry)
  2. ECG- and EGM-based analysis of AF as additive components
  3. CARTO/Rhythmia/EnSite cardiac mapping software to determine LA scar burden

HRx abstract

Figure 6: Automated EGM annotation of single lead, with high-fidelity signal resolution. 75% file size reduction and 500% read/write speed-up.

April 17, 2023

Goals

  1. Phenotyping of AF, as current grouping is inadequate. Focus on risk category changing from paroxysmal to persistent
  2. ECG- and EGM-based analysis of AF as additive components
  3. CARTO/Rhythmia/EnSite cardiac mapping software to determine LA scar burden

April 10, 2023

AF Ontology

Generation of key features that are related to arrhythmias:

  1. Clinical history and trajectory
  2. Echocardiographic findings
  3. ECG-based features
  4. Family and social history
  5. Potential genetic markers

Figure 7: Example from HF evaluation and progression

Figure 8: Utilizing patient features to create a matrix-based cluster

Figure 9: Support-vector based clustering, multidimensional K-clustering in feature-space

Phenotype-based WES cohort

Age Comorbidities Proportion
- none 10-15%
≤ 55 * 20%
≤ 65 ≤ 1 20%
≤ 65 ≤ 2 10%
≤ 65 ≤ 3 10%
≥ 65 ≤ 1 10-15%

Alternatively can aim for two major groups within paroxysmal cohort definition:

  1. ≤ 65y + structural heart disease (70%)
  2. ≤ 65y - structural heart disease (30%)

Aim for 20-30% to have had PVI to be able to integrate intracardiac findings.

Table 2: Autonomic inflexibility and CV mortality in HF and LF bands
Table 3: Improved model concordance with addition of autonomic inflexiblity

March 31, 2023

AF Phenotyping

NLP combined with DSP can help in identifying sub-phenotypes of AF…

  1. Polygenic risk score assessment based on phenotypes and sub-phenotypes
  2. Current phenotype approaches limited to structured text
  3. Unstructured data can be extracted from clinical notes using NLP tools: BERT, Sci/RoBERTa, meta mapper, etc

AWS HPC + CCTS data + AF registry + EP lab + biomarkers

Figure 10: Complexity of phenotyping is based on how structured or unstructured the data may be

Figure 11: Identifying a true phenotype is complex and most often incorrect unless corroborated through other mechanisms

Figure 12: Phenotypke KB current standard for diagnoses of AF

EPS signal processing

Utilize combination of surface ECG and intracardiac EGM to…

  1. Identify intracardiac features, e.g. multichannel morphologies
  2. Evaluate atrial abnormalities, such as conduction left → right activation
  3. Feed features into LSTM or convolutional neural network to understand different AF phenotypes

Figure 13: Annotated ECG to identify P wave, QRS, and T wave fiducial points

  • Summation of P wave force vectors across surface ECG (signal-averaged)
  • Identify terminal P wave force abnormalities
  • Evaluate changes in P wave morphology over time

March 20, 2023

Updates

  • AFL/FH:
    • manuscript was sent for initial revisions (03/03/23)
    • feedback received on 03/17/23 from Jordan
    • second draft to be returned on 3/20/23
  • Aim 1: “Big data” approach using claims-based data, IRB/CCTS approval obtained
    1. NLP/ML area of focus will be initially on AF (only performed x 1)
    2. Approach for recurrent events, Weibull distribution expected
  • HRV/CVD :
    • priority for this week
    • HF/LF HRV values showing >10 fold HR for bottom versus top quartile (with almost >95% sensitivity)
  • Genetic analysis:
    • ACER will start AWS workspace this week

Help Wanted

  • Ablation database: working with Wissner to revise ablation database
    • manual data reconciliation
    • OCR/OMR-based PDF conversion approach
    • ablation outcomes x genetics
  • AF phenotyping: key part of Aim 1
    • Computational Biorepository for Cardiovascular Disease: CCTS data pull active, all clinical notes, data points, etc since 2010 on CVD
    • AF ontology needed using a NLP/LLM (e.g. BERT)
  • NLP/ML for EHR data:
    • CBCD: overlap of DCM/AF registries, CCTS data pull of all clinical notes since 2010 on CVD
    • Key part of Aim 1 is to identify AF phenotypes in those with paroxysmal AF
    • AF Ontology: working with Andrew Boyd on AF-NLP framework (SciB)

February 27, 2023

Updates

  • AFL/FH: will need age of diagnosis of event of family history prior to final analyses
  • CAR: analysis pending by Konda, however unclear if we can strongly accept null-hypothesis

Non-linearity in ANS dysfunction

  1. Proportional hazard assumptions
    1. Defined using time-variance, time-interaction, Martingale residuals
    2. Satisfied PH assumptions
  2. Non-linearity of HRV and CV mortality
    1. Spline analysis with up to 5 knots
    2. Parametric threshold analysis (survival modeling)
    3. Non-parametric (binary outcome) threshold analysis
    4. Identified non-linearity of response
  3. Re-analyzed data using cut-point (as shown in Figure 14)
    1. Almost 100% accuracy in classifier of low-risk cohort
    2. Almost 10-fold hazard in identify high-risk cohort

Figure 14: Increase of ~10 fold in CV mortality in 1/4 patients identified by abnormal resting and reactive vagal tone. Robust classifier of resilience.

Genetic analysis

  1. WES-generated VCF files on n=28 patients from EO-AFL + FH subgroup
  2. Due to size limitations (no access to HPC) converted VCF data …
    1. VCF to Apache Arrow
    2. Arrow conversion to feather and to parquet for header, annotation, and columnar genotype data
    3. Analyze as in-memory array
    4. Re-vert to ASCII-based format for GZip formated VCF
  3. Filtering of data
    1. Limited to coding regions
    2. Filtered for read depth > 20
    3. Compared to dbSNP builds using SIFT and PolyPhen
    4. Pending further annotations based on arrhythmia-panels

Specific aims

Have revised aims, and have drafted the x1 page specific aims to be shared with mentors.

Mentors: D Darbar, AJ Shah, A Alonso, M McCauley, A Boyd

February 20, 2023

AFL/FH

  • Genetic basis for FH may be more promising than association with FH broadly, but would require additional WES to be sent
  • Would need to repeat VCF analysis for an arrhythmia panel (instead of cardiomyopathy panel)

CBCD

Computational biorepository for cardiovascular disease

  • Will need a large computational biorepository
    • Claims data from both CPT/ICD codes
    • Clinical documentation (raw text from clinical notes)
    • Medication history
    • Study data e.g. XML of ECG, echo reports, coronary angiograms, device interrogations, etc
  • CCTS to pull data, DRA to be submitted today

February 6, 2023

Updates

  • AFL/FH: WES needed prior to completing revisions
  • ARIC: potential option to pursue genetic profiling in ARIC per Alvaro, need formal proposal
  • K23: drafting specific aims in 1-pager format
  • VCF: data able to be analyzed/cleaned, however requires more CPU to perform

Figure 15: Chromosome 1 summary

January 9, 2023

Updates

AFL/FH:

  • Pedigrees/genetics completed (thanks to Ana, Shashank, Mike)
  • Outline/draft to be re-written
  1. HRV/CV Mortality paper rejected, need to re-think strategy with senior authors
  2. K23 aims to include 1) EP lab as translational component, 2) arrhythmia risk prediction as computational component

AFL/genetics

  • ~87 individuals with +FH
  • ~81 individuals with WES
  • ~18 individuals with VUS in total
  • 1 individual with VUS + FH

Multiple Pedigrees

AFL/surrogate disease burden ## TTN Mutant

AFL/mutant # December 19, 2022

Updates

AFL/FH:

  • Pedigrees TBD
  • Draft completed, pending feedback
  • Supplemental tables needed?

AF/Recurrence:

  • Scott added additional patients from UIC (September 2020 to now)
  • Data collection pending

K23 Aims

Objective: Clinical EP researcher using computational neurocardiology techniques to study arrhythmia mechanisms

  1. Evaluating vagolysis and its effect on triggered arrhythmia mechanisms, both ventricular (e.g. SCD) and atrial (triggered AF)
  2. Using computational approach to phenotype triggered onset arrhythmias, e.g. atrial fibrillation

Mentorship Team: Amit J. Shah, Dawood Darbar, Rachel Lampert, Viola Vaccarino, Mark McCauley, Andrew Boyd, Alvaro Alonso

K23 Approach

Vagal tone and triggered arrhythmias

  • Manuscript on stress-induced vagolysis and CV mortality under review
  • ANS dysfunction during/with ischemia work-in-progress
  • [Create murine/translational model for vagolysis and triggered arrhythmias]{.arrhythmia[2]}

Arrhythmia phenotyping

  • AFL/FH manuscript work-in-progress
  • AF/race project work-in-progress
  • [Classification of arrhythmia phenotypes using large data set (MVP/VA research data, UIC, ARIC)]{.computational[3]}

Murine Model of Vagolysis

  • Atria are heavily innervated by ANS ganglia
    • Sympathovagal balance locally mediated through adrenergic lysis of cholinergic activity
    • Neuropeptide Y (NPY) causes cholinergic inhibition through Y2R
  • Pro-arrhythmic murine model (compared to healthy controls)
    • Ex-vivo vagal-sparing Langendorf preparation
    • Catecholamine infusion and NPY Y2R antagonists to modulate arrhythmic state
  • Measure atrial conductive properties as outcome
    • Baseline, with catecholamine infusion, VNS stimulation, and Y2R antagonism

December 12, 2022

AF Catheter Ablation and Recurrence

Data/power increase:

  • Inclusion of VA data
  • Additional data pull from EPIC by Scott Uphouse (pending)
  • Inclusion of UIC billing code data?

Atrial Flutter and Family History

Pedigrees:

  • How should these be incorporated into the manuscript?

Genetics:

  • 305 patients that have undergone whole exome sequencing, but unclear how to match these to the AFL data
  • Should we genotype the rest of the EO-AFL patients?

ECG/EGM analysis

November 21, 2022

Updates

  • AHA IPA: Completed draft, pending revisions
  • AFL Paper: Draft to be done by Wednesday
  • AF Ablation + Biorepository: Pending IRB
  • Stress and CVD Mortality: Submitted to Circulation, revisions Pending

November 14, 2022

Early Onset Atrial Flutter

  1. Draft of paper expanded to include results
  2. Prominent findings of family history and association with arrhythmia
  3. Key tables and figures established

Next steps:

  • Find agreement in key results and figures
  • Identify main finding
  • Complete discussion section

October 10, 2022

AF Registry

Status:

  • Current overview
  • Not yet powered
  • Will need to revise out approach

October 2, 2022

AF Registry

  • Current number: 219
  • Recurrence rate: 46%
  • Death = 12
  • MACE = 70 (including repeats)

GIS-based MAP of right atrium for mapping and localization purposes for publications.

September 26, 2022

AF Registry

Analysis

  • Genetic analysis in PLINK complete
  • Models are set up for analysis once data intake is done

Status

  1. Have n = 238 currently
  2. Reviewed 95 charts thus far (CAR team)
  3. Of n = 108, 75 have had recurrence, 33 without (rate of about 70% of anyarrhythmia)
  4. Of ECG based confirmation, 43 AF recurrence events (53%)

September 12, 2022

AF Registry

REDCap data dictionary is finalized. Next steps are:

  1. Data entry for outcomes
  2. Finalization of new patients to add from AF ablation registry

Genetics

Currently able to:

  • Format large SNP data sets into appropriate tables
  • Utilize PLINK and MERLIN to process Chen’s genetics data
  • Run from R to help analyze findings

Next steps:

  • Confirm ancestry (for practice) percent likelihood
  • Then, once outcome data is complete, analyze differences by race

Analytical methods:

  • Time to binary outcome of recurrence
  • Recurrent event analysis for AF recurrence
  • Can revise with Bayesian modeling in STAN as well

August 22, 2022

AF Registry

Status

  • REDCap is in progress
  • Have an additional list of registry patients that may/may not have had ablation

Next Steps

  1. Complete 100 REDCap patients
  2. Identify 300 patients that have potentially had ablation

CARTO Maps

  • Ablation quantification
  • Activation mapping
  • EGM annotation
  • Voltage mapping
  • Conduction velocity
  • Earliest/latest activation
  • Geometry data

August 15, 2022

UIC-wide data summary

  • N ~ XXX had ablation
  • Date range from >= 2010

AF registry data

  • N=~300 had ablation
  • Date range from >= 2010

Next Steps

  1. Generate list of MRN for data collection
  2. Use new REDCap to obtain ECG/ablation data
  3. Run preliminary analysis on first 100 patients
  4. Consider expansion to JBVA/ACMC data

August 1, 2022

AF Registry

Consider additional variables:

  • Recurrent events and adjudication (HF, cardiac admission, MACE)
  • Holter/ECG data and repeat collection (P-wave morphology, Wilson’s vector gradient, amplitude of AF waves “coarseness”)

Will check with the MESA and ARIC data base on how repeat events were defined.

Murine EP Studies

Human EP Studies

July 25, 2022

Forest Plots

# Forest Plots
library(vlndr)
m <- rx(
    Petal.Length ~ X(Sepal.Length) + Petal.Width + S(Species),
    label = list(
        Petal.Length ~ "Length of Petals", 
        Species ~ "Genus of the Flower"
    ),
    pattern = "direct"
    ) |>
    fmls(order = 2) |>
    fit(.fit = lm,
            data = iris,
            archetype = TRUE) |>
    mdls()
tbl <- tbl_forest( #<<
    object = m,
    formula = Petal.Length ~ Sepal.Length,
    vars = "Species",
    columns = list(
        beta ~ "Estimate",
        conf ~ "95% Confidence Interval",
        n ~ "Number of Samples"
    ),
    axis = list(
        lim ~ c(0, 3),
        breaks ~ c(0, 1, 2),
        lab ~ "ß (95% CI)",
        title ~ "Petal Length by Sepal Length"
    )
)

A simple way to make sub-group analysis Forest plots, working on improving customization.

July 18, 2022

AF Registry

The AF registry is described here. Current issues:

  1. Data quality - consistency between reviewers
  2. Missingness - decisions on acceptable thresholds
  3. Variable selection - echo findings, EP studies, medications, labs, symptoms
  4. Management - REDCap, shared excel sheet
  5. Adjudication/review - outcomes, clinical follow-up length
Andrade, Jason G., Marc W. Deyell, Laurent Macle, George A. Wells, Matthew Bennett, Vidal Essebag, Jean Champagne, et al. 2023. “Progression of Atrial Fibrillation After Cryoablation or Drug Therapy.” New England Journal of Medicine 388 (2): 105–16. https://doi.org/10.1056/NEJMoa2212540.
Benjamin, Emelia J., Kevin L. Thomas, Alan S. Go, Patrice Desvigne-Nickens, Christine M. Albert, Alvaro Alonso, Alanna M. Chamberlain, et al. 2023. “Transforming Atrial Fibrillation Research to Integrate Social Determinants of Health: A National Heart, Lung, and Blood Institute Workshop Report.” JAMA Cardiology 8 (2): 182–91. https://doi.org/10.1001/jamacardio.2022.4091.
Essien, Utibe R., Jelena Kornej, Amber E. Johnson, Lucy B. Schulson, Emelia J. Benjamin, and Jared W. Magnani. 2021. “Social Determinants of Atrial Fibrillation.” Nature Reviews. Cardiology 18 (11): 763–73. https://doi.org/10.1038/s41569-021-00561-0.
Holmqvist, Fredrik, Sunghee Kim, Benjamin A. Steinberg, James A. Reiffel, Kenneth W. Mahaffey, Bernard J. Gersh, Gregg C. Fonarow, et al. 2015. “Heart Rate Is Associated with Progression of Atrial Fibrillation, Independent of Rhythm.” Heart (British Cardiac Society) 101 (11): 894–99. https://doi.org/10.1136/heartjnl-2014-307043.
Kerr, Charles R., Karin H. Humphries, Mario Talajic, George J. Klein, Stuart J. Connolly, Martin Green, John Boone, Robert Sheldon, Paul Dorian, and David Newman. 2005. “Progression to Chronic Atrial Fibrillation After the Initial Diagnosis of Paroxysmal Atrial Fibrillation: Results from the Canadian Registry of Atrial Fibrillation.” American Heart Journal 149 (3): 489–96. https://doi.org/10.1016/j.ahj.2004.09.053.
Leur, Rutger R. van de, Karim Taha, Max N. Bos, Jeroen F. van der Heijden, Deepak Gupta, Maarten J. Cramer, Rutger J. Hassink, et al. 2021. “Discovering and Visualizing Disease-Specific Electrocardiogram Features Using Deep Learning.” Circulation: Arrhythmia and Electrophysiology 14 (2): e009056. https://doi.org/10.1161/CIRCEP.120.009056.
Takigawa, Masateru, Atsushi Takahashi, Taishi Kuwahara, Kenji Okubo, Yoshihide Takahashi, Yuji Watari, Katsumasa Takagi, et al. 2014. “Long-Term Follow-Up After Catheter Ablation of Paroxysmal Atrial Fibrillation.” Circulation: Arrhythmia and Electrophysiology 7 (2): 267–73. https://doi.org/10.1161/CIRCEP.113.000471.
Verweij, Niek, Jan-Walter Benjamins, Michael P. Morley, Yordi J. Van De Vegte, Alexander Teumer, Teresa Trenkwalder, Wibke Reinhard, Thomas P. Cappola, and Pim Van Der Harst. 2020. “The Genetic Makeup of the Electrocardiogram.” Cell Systems 11 (3): 229–238.e5. https://doi.org/10.1016/j.cels.2020.08.005.
Veselá, Jana, Pavel Osmančík, Dalibor Heřman, Sabri Hassouna, Radka Raková, Tomáš Veselý, and Petr Budera. 2023. “Prediction of Post-Operative Atrial Fibrillation in Patients After Cardiac Surgery Using Heart Rate Variability.” BMC Cardiovascular Disorders 23 (1): 290. https://doi.org/10.1186/s12872-023-03309-5.