Predicting Patient-level New Onset Atrial Fibrillation

Atrial Fibrillation Clinical Trial

Official title:

Predicting Patient-level New Onset Atrial Fibrillation From Population-based Nationwide Electronic Health Records: A Precision Medicine Investigation Using Artificial Intelligence

Clinical Trial Details — Status: Completed

Administrative data

• NCT number: NCT04657900
• Other study ID #: 120029
• Status: Completed
• Start date: November 2, 2020
• Est. completion date: October 31, 2023

Study information

• Verified date: May 2024
• Source: University of Leeds
• Contact: n/a
• Is FDA regulated: No
• Study type: Observational

Clinical Trial Summary

Atrial fibrillation (AF) is a major cardiovascular health problem: it is common, chronic and incurs substantial health-care expenditure as a result of stroke, sudden death, heart failure and unplanned hospitalisation. There is a compelling argument for the early diagnosis of AF, before the first complication occurs, but population-based screening is not recommended. Strategies to identify individuals at higher risk of new onset AF are required. previous risk scores have been limited by data and methodology. The investigators will use routinely collected hospital-linked primary care data and focus on the use of artificial intelligence methods to develop and validate a model for the prediction of incident AF. Specifically, the investigators will investigate how population-based data may be used for precision medicine using a deep neural networks learning model. Using clinical factors readily accessible in primary care, the investigators will provide a method for the identification of individuals in the community who are at risk of AF, as well as when incident AF will occur in those at risk, thus accelerating research assessing technologies for the improvement of risk prediction, and the targeting of high-risk individuals for preventive measures and screening.

Description:

Atrial fibrillation (AF) is a common chronic condition with substantial impact on health outcomes. Many cases of AF are detected too late - as a manifestation of stroke, heart failure, or other complication. Whilst earlier detection of AF offers the potential to prevent premature cardiovascular disease, population screening is not recommended.

Atrial fibrillation (AF) is a leading cardiovascular health problem. It is the most common sustained cardiac arrhythmia, affecting 1-2% of the population of Europe and the USA, with a lifetime risk of one in four in the general population. It has an increasing prevalence as the population ages. Consequently, these estimates are likely to increase, and presently are underestimated given that AF may long remain undiagnosed. AF incurs 1-3% of healthcare expenditure as a result of stroke, sudden death, heart failure, unplanned hospitalisation, and associated complications. The resultant emerging AF epidemic and its associated costly complications (including, but not limited to stroke, depression, heart failure, acute coronary syndrome, cognitive decline and unscheduled hospitalisation) has ensured that AF is now a major threat to healthy longevity. The early diagnosis of AF, ideally before manifestation of the first complication, remains a major public health challenge. While for some patients AF may present with symptomatic palpitations, for others the first diagnosis of AF may be when they present to healthcare professionals with stroke, acute cardiac decompensation or co-morbidity exacerbation - a stage that is unnecessarily late in the disease trajectory. This is because many patients with AF may not have AF-associated symptoms. Given that nearly one third of patients admitted to a stroke ward have AF at the time of their admission to hospital and that oral anticoagulants reduce the risk of stroke by up to two thirds in those with AF who are at higher risk of stroke, there is a compelling argument for the earlier detection of AF. To that end, opportunistic screening for AF (pulse palpation followed by ECG in patients with an irregular pulse) in patients aged 65 years and over is now recommended in national and international guidelines. International guidelines also recommend the use of a 12 lead ECG and ambulatory rhythm monitors (within increasing duration according to perceived risk of AF), escalating to implantable leadless AF recorders in patients with suspected but undiagnosed AF - and each with implications for healthcare costs and patient satisfaction. Whilst there are promising results from systematic screening of elderly populations for AF using self-operated devices, presently there is no recommendation in the United Kingdom (UK) for population-wide systematic screening for AF because it is not yet clear if those identified as at risk would benefit from early diagnosis. Indeed, research is needed to understand better the detection rates, diagnostic accuracy, outcomes of such programs, as well as to define in what sub-populations AF screening would offer the greatest patient and public health value.

The identification of AF has important patient and clinical ramifications. Those patients at higher risk of stroke (CHADSVASC score ≥ 2) without a contra-indication should be offered stroke prophylaxis with an oral anticoagulant. Moreover, most patients with AF will have stroke risk factors, making them eligible for an oral anticoagulant, and many will have concomitant cardiovascular disease (such as hypertension, valvular heart disease or heart failure) making them eligible for further investigation or treatment. Equally, in those with AF who are low risk for stroke (and therefore do not qualify for oral anticoagulation), surveillance for increasing stroke risk is advisable. Predicting precisely if and when a person will have new onset AF may allow phenotype and temporal-specific (thus more effective) screening, as well as identify putative risk markers for AF aetiology. For example, patients presently in sinus rhythm, but at higher risk of stroke and predicted to develop AF at a specific time-point in the future may benefit from screening for AF nearer the forecasted date. Equally, modifiable risk factors for the development of AF and for risk of stroke may be proactively addressed in light of knowledge of higher risk of new onset AF, and new risk factors studied for causality. Other possible research opportunities may include the study of patients who do not have and are not predicted to have AF, and the evaluation of lifestyle, device technology and pharmacotherapeutic strategies to reduce the risk of AF in patients at high predicted risk of new onset AF.

To date, a number of AF risk prediction tools have been developed, including those from the CHARGE-AF consortium, Framingham Heart Study, the CHADS score, the CHADSVASC score and the CHEST score, among others. The CHEST score (structural heart disease, heart failure, age ≥ 75 years, coronary artery disease, hyperthyroidism, Chronic Obstructive Pulmonary Disease (COPD), and hypertension) derived from 471,446 subjects from the Chinese Yunnan Insurance Database and validated in 451,199 subjects from the Korean National Health Insurance Service was found to predict future incident AF. Of the 4764 participants in the Framingham Heart Study, age, sex, body-mass index, systolic blood pressure, treatment for hypertension, the time from the onset of the P wave to the start of the ventricular depolarization (QRS) complex (PR interval), clinically significant cardiac murmur, and heart failure were found using survival modelling to be components of a score predicting incident AF at 10 years. However, each of the studies to date are limited by one or more of, their use of geographically remote data, historical data, small datasets, lack of temporal information, crude risk modelling with consequent suboptimal model performance and/or predictor variables not readily available to the General Practitioner. Understandably, none have reached widespread clinical practice. Artificial intelligence facilitates the use of vast quantities of event data and the associated temporal information (such as that in primary care datasets), handles large numbers of predictors with automatic variable selection techniques, accommodates nonlinearities and interactions among variables, enables a live learning approach (whereby the prediction model is automatically updated), and can use population-wide data to predict if and when there will be new onset AF for an individual. A range of Artificial Intelligence (AI) techniques have been applied to EHR data and have demonstrated better diagnostic and prediction power over traditional statistical approaches in large scale EHR data. Yet, as highlighted recently, it is important to identify models that are clinically useful. For example, a study which developed an AI-enabled ECG algorithm that predicted AF from ECGs with normal sinus rhythm, whilst an important step forward may not be applicable in the community setting where routine ECGs are not always available. Thus, developing a predictive algorithm for new onset AF from routine primary care electronic health records data using AI techniques could offer the opportunity for early translation to clinical practice. The investigators will develop and validate a deep neural networks learning model, utilising large scale linked electronic health records (EHR) from primary care, to predict the risk of new AF. The prediction algorithm will be trained and tested for its accuracy and robustness in predicting future AF events using Clinical Practice Research Datalink (CPRD)-Global initiative for chronic Obstructive Lung Disease (GOLD), and will be externally validated using similar databases CPRD-AURUM but at different geographic locations. The new predictive algorithm will be compared against a range of classic machine learning techniques as well as traditional statistical predictive modelling methods. Pending a successful model of improving the predicting accuracy of at least 5% compared with existing models, the algorithm could be made readily available through free to use software.

Recruitment information / eligibility

• Status: Completed
• Enrollment: 140000
• Est. completion date: October 31, 2023
• Est. primary completion date: October 31, 2023
• Accepts healthy volunteers: No
• Gender: All
• Age group: 18 Years and older

Eligibility

Inclusion Criteria:

• Diagnosed AF after 1 January 2009 (Identified using Read codes (for the CPRD patient profile) and ICD-10 codes (for HES events)

• In Clinical Practice Research Datalink -Global initiative for chronic Obstructive Lung Disease (CPRD-GOLD) and eligible for data linkage.

• Have at least 1-year follow-up in the period between 1st Jan 1998 and 31st December 2018.

Exclusion Criteria:

• Under 18 at date of the first registration in CPRD

• Diagnosed with AF before 1st Jan 1998

• In CPRD-GOLD and not eligible for data linkage

• Has less than one year follow up in CPRD

Related Conditions & MeSH terms

• Atrial Fibrillation

Intervention

Other:
• Observational
Observational - no intervention given

Locations

Country	Name	City	State
United Kingdom	University of Leeds	Leeds	West Yorkshire

Sponsors (2)

Lead Sponsor	Collaborator
University of Leeds	British Heart Foundation

Country where clinical trial is conducted

United Kingdom, 

References & Publications (25)

Alonso A, Krijthe BP, Aspelund T, Stepas KA, Pencina MJ, Moser CB, Sinner MF, Sotoodehnia N, Fontes JD, Janssens AC, Kronmal RA, Magnani JW, Witteman JC, Chamberlain AM, Lubitz SA, Schnabel RB, Agarwal SK, McManus DD, Ellinor PT, Larson MG, Burke GL, Laun — View Citation

Aronson D, Shalev V, Katz R, Chodick G, Mutlak D. Risk Score for Prediction of 10-Year Atrial Fibrillation: A Community-Based Study. Thromb Haemost. 2018 Sep;118(9):1556-1563. doi: 10.1055/s-0038-1668522. Epub 2018 Aug 13. — View Citation

Attia ZI, Noseworthy PA, Lopez-Jimenez F, Asirvatham SJ, Deshmukh AJ, Gersh BJ, Carter RE, Yao X, Rabinstein AA, Erickson BJ, Kapa S, Friedman PA. An artificial intelligence-enabled ECG algorithm for the identification of patients with atrial fibrillation — View Citation

Beam AL, Kohane IS. Big Data and Machine Learning in Health Care. JAMA. 2018 Apr 3;319(13):1317-1318. doi: 10.1001/jama.2017.18391. No abstract available. — View Citation

Camm AJ, Kirchhof P, Lip GY, Schotten U, Savelieva I, Ernst S, Van Gelder IC, Al-Attar N, Hindricks G, Prendergast B, Heidbuchel H, Alfieri O, Angelini A, Atar D, Colonna P, De Caterina R, De Sutter J, Goette A, Gorenek B, Heldal M, Hohloser SH, Kolh P, L — View Citation

Chamberlain AM, Agarwal SK, Folsom AR, Soliman EZ, Chambless LE, Crow R, Ambrose M, Alonso A. A clinical risk score for atrial fibrillation in a biracial prospective cohort (from the Atherosclerosis Risk in Communities [ARIC] study). Am J Cardiol. 2011 Ja — View Citation

Conen D. Epidemiology of atrial fibrillation. Eur Heart J. 2018 Apr 21;39(16):1323-1324. doi: 10.1093/eurheartj/ehy171. No abstract available. — View Citation

Ehteshami Bejnordi B, Veta M, Johannes van Diest P, van Ginneken B, Karssemeijer N, Litjens G, van der Laak JAWM; the CAMELYON16 Consortium; Hermsen M, Manson QF, Balkenhol M, Geessink O, Stathonikos N, van Dijk MC, Bult P, Beca F, Beck AH, Wang D, Khosla — View Citation

Friberg L, Rosenqvist M, Lindgren A, Terent A, Norrving B, Asplund K. High prevalence of atrial fibrillation among patients with ischemic stroke. Stroke. 2014 Sep;45(9):2599-605. doi: 10.1161/STROKEAHA.114.006070. Epub 2014 Jul 17. — View Citation

Fuster V, Ryden LE, Cannom DS, Crijns HJ, Curtis AB, Ellenbogen KA, Halperin JL, Le Heuzey JY, Kay GN, Lowe JE, Olsson SB, Prystowsky EN, Tamargo JL, Wann S, Smith SC Jr, Jacobs AK, Adams CD, Anderson JL, Antman EM, Halperin JL, Hunt SA, Nishimura R, Orna — View Citation

Gulshan V, Peng L, Coram M, Stumpe MC, Wu D, Narayanaswamy A, Venugopalan S, Widner K, Madams T, Cuadros J, Kim R, Raman R, Nelson PC, Mega JL, Webster DR. Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Re — View Citation

Huang Z, Dong W, Duan H, Liu J. A Regularized Deep Learning Approach for Clinical Risk Prediction of Acute Coronary Syndrome Using Electronic Health Records. IEEE Trans Biomed Eng. 2018 May;65(5):956-968. doi: 10.1109/TBME.2017.2731158. Epub 2017 Jul 24. — View Citation

January CT, Wann LS, Alpert JS, Calkins H, Cigarroa JE, Cleveland JC Jr, Conti JB, Ellinor PT, Ezekowitz MD, Field ME, Murray KT, Sacco RL, Stevenson WG, Tchou PJ, Tracy CM, Yancy CW; ACC/AHA Task Force Members. 2014 AHA/ACC/HRS guideline for the manageme — View Citation

Kirchhof P, Auricchio A, Bax J, Crijns H, Camm J, Diener HC, Goette A, Hindricks G, Hohnloser S, Kappenberger L, Kuck KH, Lip GY, Olsson B, Meinertz T, Priori S, Ravens U, Steinbeck G, Svernhage E, Tijssen J, Vincent A, Breithardt G. Outcome parameters fo — View Citation

Kirchhof P. The future of atrial fibrillation management: integrated care and stratified therapy. Lancet. 2017 Oct 21;390(10105):1873-1887. doi: 10.1016/S0140-6736(17)31072-3. Epub 2017 Apr 28. Erratum In: Lancet. 2017 Oct 21;390(10105):1832. Dosage error — View Citation

Kolek MJ, Graves AJ, Xu M, Bian A, Teixeira PL, Shoemaker MB, Parvez B, Xu H, Heckbert SR, Ellinor PT, Benjamin EJ, Alonso A, Denny JC, Moons KG, Shintani AK, Harrell FE Jr, Roden DM, Darbar D. Evaluation of a Prediction Model for the Development of Atria — View Citation

Li YG, Pastori D, Farcomeni A, Yang PS, Jang E, Joung B, Wang YT, Guo YT, Lip GYH. A Simple Clinical Risk Score (C2HEST) for Predicting Incident Atrial Fibrillation in Asian Subjects: Derivation in 471,446 Chinese Subjects, With Internal Validation and Ex — View Citation

Obermeyer Z, Emanuel EJ. Predicting the Future - Big Data, Machine Learning, and Clinical Medicine. N Engl J Med. 2016 Sep 29;375(13):1216-9. doi: 10.1056/NEJMp1606181. No abstract available. — View Citation

Rahimian F, Salimi-Khorshidi G, Payberah AH, Tran J, Ayala Solares R, Raimondi F, Nazarzadeh M, Canoy D, Rahimi K. Predicting the risk of emergency admission with machine learning: Development and validation using linked electronic health records. PLoS Me — View Citation

Riley RD, Snell KI, Ensor J, Burke DL, Harrell FE Jr, Moons KG, Collins GS. Minimum sample size for developing a multivariable prediction model: PART II - binary and time-to-event outcomes. Stat Med. 2019 Mar 30;38(7):1276-1296. doi: 10.1002/sim.7992. Epu — View Citation

Schnabel RB, Aspelund T, Li G, Sullivan LM, Suchy-Dicey A, Harris TB, Pencina MJ, D'Agostino RB Sr, Levy D, Kannel WB, Wang TJ, Kronmal RA, Wolf PA, Burke GL, Launer LJ, Vasan RS, Psaty BM, Benjamin EJ, Gudnason V, Heckbert SR. Validation of an atrial fib — View Citation

Schnabel RB, Sullivan LM, Levy D, Pencina MJ, Massaro JM, D'Agostino RB Sr, Newton-Cheh C, Yamamoto JF, Magnani JW, Tadros TM, Kannel WB, Wang TJ, Ellinor PT, Wolf PA, Vasan RS, Benjamin EJ. Development of a risk score for atrial fibrillation (Framingham — View Citation

Shah NH, Milstein A, Bagley PhD SC. Making Machine Learning Models Clinically Useful. JAMA. 2019 Oct 8;322(14):1351-1352. doi: 10.1001/jama.2019.10306. No abstract available. — View Citation

Sultan AA, West J, Grainge MJ, Riley RD, Tata LJ, Stephansson O, Fleming KM, Nelson-Piercy C, Ludvigsson JF. Development and validation of risk prediction model for venous thromboembolism in postpartum women: multinational cohort study. BMJ. 2016 Dec 5;35 — View Citation

Wolf PA, Abbott RD, Kannel WB. Atrial fibrillation: a major contributor to stroke in the elderly. The Framingham Study. Arch Intern Med. 1987 Sep;147(9):1561-4. — View Citation

* Note: There are 25 references in all — Click here to view all references

Outcome

Type	Measure	Description	Time frame	Safety issue
Primary	To develop and validate a deep learning hierarchical model for predicting the risk, and where appropriate period, of new onset AF	Predictive factors will be identified using Read codes (diagnoses), measurements and Prod codes (medications) in CPRD; ICD10 codes and statistical classification (OPCS) codes in Hospital Episode Statistics (HES); and ICD 10 codes (ICD9 codes for the period before 2001) in Office of National Statistics (ONS) data. All variables will be considered as potential predictors, and may include: sociodemographic variables: age, sex, ethnicity, index of multiple deprivation; all (repeated) hospitalised disease conditions during follow-up; clinical assessments, such as ECG, heart rate, height, weight,; medications prescribed,; lifestyle factors (e.g. smoking status, alcohol consumption); all biomarkers collected during follow-up The temporal information of all clinical assessments, hospitalised events, medications will be included.	Between 1st Jan 1998 and 31st December 2018
Primary	To identify and quantify the magnitude of predictors of new onset AF	The proposed deep learning model can extract informative risk factors from EHR data.; Specifically, a risk factor selection strategy proposed in Huang et al will be adapted to identify informative risk factors. The model will provide weights of the identified risk factors to help understand the significance of risk factors at different risk levels. The impact of the number of risk factors on the performance of AF risk prediction will be assessed through the curves of both area under curve (AUC) and prediction accuracy plotted against the number of risk factors. Some predictors, such as BMI, blood pressure, frequency of General Practitioner (GP) visits, strength of prescribed medication, may change over time. The incremental prognostic values of including these variable trajectories will be explored and the impact on predictive accuracy will be assessed.	Between 1st Jan 1998 and 31st December 2018