Decision Support System for Diagnosis and Progression of Heart Failure

Overview

Heart failure (HF) is a complex clinical syndrome associated with impaired heart function, poor quality of life for patients and high healthcare costs. Accurate risk stratification and early diagnosis in HF are challenging as signs and symptoms are non-specific. Here the investigators propose to address this global challenge by developing novel analytic methods for HF (STRATIFYHF). A prospective clinical study will collect patient-specific data related to medical history, a physical examination for signs and symptoms, blood tests including natriuretic peptides, an electrocardiogram (ECG), an echocardiogram (ultrasound of the heart), cardiovascular magnetic resonance imaging (MRI), demographic, socio-economic and lifestyle data along with novel technologies (cardiac output response to stress (CORS) test and voice recognition biomarkers) from individuals at-risk of developing HF and those with a confirmed diagnosis of HF. STRATIFYHF will use these data to develop, validate and implement the first artificial intelligence (AI)-based, Decision Support System (DSS) for assessing and predicting the risk of HF development, its early diagnosis and progression. STRATIFYHF will integrate 1) patient-specific data i.e. demographic, clinical, genetic, lifestyle and socio-economic, 2) an AI-based digital patient library and AI-driven algorithms for risk stratification, early diagnosis, and disease progression in HF, and 3) a highly innovative multifunctional AI-based DSS and mobile application for informing a patient-centred, personalised, prevention and treatment strategies for HF.

Aim and Objectives: This prospective study is part of the STRATIFYHF project which aims to validate a decision support system for risk prediction, diagnosis, and progression of HF. To achieve this, the investigators will undertake a prospective study to collect data and deliver proof-of-concept study with a duration of 36 months. Eight clinical centres will recruit 1,600 patients (i.e., 800 suspected and 800 confirmed HF patients) with recruitment target of 8-9 patients per month and up to 24-month follow-up. Design and Methods: The present study is a prospective, longitudinal, clinical observational study comprising eight clinical partners across Europe. The study will start on 1st July 2024 and last for 36 months. Research visits and clinical assessments: Eligible participants will attend the clinical research site of participating centres for two visits (i.e., baseline and follow-up clinical assessments at month 12). Each visit will last approximately 2.5 hours. An additional visit will be arranged in a subset of participants for a cardiac magnetic resonance imaging scan. Further monitoring of patient outcomes (e.g., hospitalization) will be performed via examination of medical records and telephone calls to patients. i) Consent and Screening Questionnaires Patients will be provided with the opportunity to ask further questions and requested to provide written informed consent. Review of the medical history will be performed. ii) Physical examination anthropometric and body composition Physical examination will be performed by a medically trained member of the research team. Body weight and height will be measured using a hospital-based scale and stadiometer. The amount of fat and muscle in the body will be assessed using bioimpedance or other appropriate methods. In case of contraindications only anthropometrics (weight, height, and waist circumference) will be measured. iii) Blood Samples Blood sample will be taken from the antecubital vein. The blood sample will be assessed for cardiac biomarkers (brain natriuretic peptides (NTproBNP), troponin, total protein, creatine kinase myocardial band), lipid profile (total cholesterol, HDL-cholesterol, LDL-cholesterol, triglycerides), full blood count, glucose, HbA1c, markers of renal function (albumin, urea, creatinine, eGFR), markers of liver function (aspartate aminotransferase, alanine transaminase, gamma-glutamyl transpeptidase, bilirubin), calcium, potassium, sodium, uric acid, markers of inflammation (C-Reactive Protein), and thyroid-stimulating hormone. iv) Quality of life questionnaires All participants will be asked to complete the Short Form-36 quality of life questionnaire. In addition, participants with diagnosed HF will be asked to complete the validated Minnesota Living with Heart Failure (MLHFQ) questionnaire. v) Health Economics Health economics analysis will be performed based on clinical records on participants' use of healthcare facilities related to HF (including visits to GP, Cardiology department, rehabilitation, and other specialist services), clinical investigations completed, and medication use. vi) Electrocardiography An ECG with integrated heart rate variability will be performed using a standard 12-lead electrocardiogram in the supine position. In addition, cardiac autonomic function i.e., heart rate variability integrated into the ECG device will also be assessed. vii) Arterial stiffness assessment (Coventry University only) Arterial stiffness, as a measure of arterial function, will be assessed using the non-invasive SphygmoCor device, which allows for both pulse wave analysis and pulse wave velocity to be performed non-invasively using the gold standard techniques. The measurement is simple and painless, taking only a few minutes to perform. While the participant is in a comfortable supine position, the researcher will place a tonometer (pencil-like sensor) gently against the wrist and will record blood pressure signal from the pulse. viii) Transthoracic Echocardiography (ultrasound of the heart) An echocardiogram is an ultrasound scan of the heart which detail's structure and function of the heart. Transthoracic echocardiography including colour and tissue Doppler will be performed at rest and in response to Valsalva manoeuvre. Real-time images acquired in the standard parasternal (long-axis) and apical (apical 4 chamber, apical 2 chamber, and apical long axis) views, for which three cardiac cycles recorded. Parasternal short-axis views acquired at three levels: basal (at mitral valve level), midpapillary, and apical (minimum cavity distal to papillary muscle level). Parasternal long axis of the right ventricle and right ventricular outflow tract will be monitored. Peak velocity of the left ventricular outflow tract will be recorded from the apical 5 chamber view by pulse Doppler, used to calculate pressure gradient. Apical 4-chamber view will be used for right ventricular evaluation. ix) Cardiac output response to stress (CORS) test Participants will then be connected to the bioreactance Non-Invasive Cardiac Output Monitor (NICOM, Starling, Baxter Inc., USA) which the investigators have previously evaluated. The method uses four pairs of electrodes applied at the front side of the upper and lower thorax (similar to ECG). Bioreactance is a novel method for continuous non-invasive cardiac function monitoring and estimates cardiac output by analysing the frequency of relative phase shift of electronic current delivered across the thorax. Measurements will be performed using the protocol the investigators have recently developed consisting of three phases: 3-min rest (supine) phase, 3-min challenge (standing) phase, and 3-min stress (exercise step-test) phase. x) Cardiac magnetic resonance imaging Each centre will undertake CMR assessment in a subset of 20-40 participants who have not had CMR over the previous 6 months or have no contraindications to CMR studies as determined by a cardiologist or qualified radiographer. Participants will undergo cardiac cine imaging, to evaluate cardiac morphology, systolic and diastolic function. These measurements will be taken according to standard imaging protocol of the clinical testing facility. xi) Voice recording Subtle changes in speech pattern have emerged as a tool to risk stratify, diagnose, and monitor cardiovascular conditions. Participants will be seated in a quiet room in the clinical research facility to complete the recording. Participants' voices will be audio recorded for up to five minutes while they read a standardised text aloud. The recorder will be started just before patients start to read the text and stopped at the end of the task. The standardised text will be translated in different languages to ensure diversity and inclusion is achieved. The audio-file will be transferred to a computer and then processed using a dedicated software which is available from https://www.fon.hum.uva.nl/praat/. The software extracts relevant voice features after low-level acoustic features and sudden impulsive noise have been removed using spectral noise gating and voice processing techniques. Extracted voice features are presented in numerical values and will be stored in a password protected computer for further analysis using machine learning and artificial intelligence. xii) Home-based monitoring Participants will have the opportunity to opt in to participate in the home-based monitoring. Thirty participants per centre who have provided consent for the mobile application, will undertake home-based monitoring for body weight, blood pressure, accelerometry, and electrocardiography during six months follow-up. Participants will be selected from the first 30 participants who have opted in to participate in this part of the study. Participants who do not have access at home to body weight scale or blood pressure monitor will be provided the same by the local research team. Accelerometry and electrocardiography will be monitored using a wrist-worn watch (ScanWatch 2, Withings, France). xiii) Focus groups and semi-structured interviews Interviews with up to 15 healthcare professionals and up to 20 study participants (i.e., 10 at risk of HF and 10 diagnosed with HF), will be conducted to explore the needs of patients and health care providers prior to the development of decision support system (DSS) and to evaluate the acceptability of the mobile application. Additionally, Coventry and Newcastle Universities will conduct interviews with up to 20 study participants in total to evaluate their perception of being at risk of heart failure, medical support available to them, and their experience of completing the CORS test. An online training programme to guide the use of the DSS will be developed with healthcare professionals. The training programme will be delivered face-to-face to general practitioners, cardiologists, and nurses by a trained researcher. The stand-alone online training programme will also be made available. Sample size and statistical analysis: Power calculation: With an anticipated drop-out to follow-up assessments rate of 20%, it is estimated that recruitment of 1,600 patients in prospective phase will be sufficient to achieve a desired specificity and sensitivity of \>95% within a 4% margin of error. This sample size will also provide sufficient power to evaluate accuracy of prediction and diagnosis of different types of HF based on the most recent guidelines (i.e., heart failure reduced ejection fraction; heart failure mildly reduced ejection fraction; heart failure improved ejection fraction and heart failure preserved ejection fraction) assuming a sensitivity of at least 90% with a margin of error of 12% and a prevalence of heart failure with preserved ejection fraction (HFpEF) of \~50% in patients with a diagnosis of HF. Logistic regression will be used to evaluate the accuracy of the DSS. Cross tabulation of test versus observed disease status will enable calculation of sensitivity, specificity, positive and negative predictive values, and likelihood ratios. 95% confidence intervals will be calculated using the binomial exact method. To assess the accuracy of the DSS to detect different types of HF a multinomial regression and classification trees will be used. The AUC-ROC (area under receiver operator characteristic) curve for each diagnostic test result will be separately assessed through univariate logistic regression models. Multiple logistic regressions will also be used to evaluate the DSS added diagnostic value. Reclassification measures will be used to assess how many patients are reclassified by adding the novel technologies (CORS test and voice biomarkers) to the existing investigations (combined into a multivariable model) after introducing a particular probability of disease presence threshold. Multiple logistic regressions will be used for the initial development of the DSS for primary and secondary care to predict the presence or absence and type of HF. Data on diagnostic efficacy outcomes, use of resources and unit costs of the resources used will be utilised for health economic assessment.