Broad spectrum of cardiovascular diseases (e.g., coronary heart disease, heart failure, cardiac arrhythmias, valvular heart diseases) based on retrospectively analyzed cardiological routine data.
Conditions
Interventions
Group 1: Retrospective analysis of pseudonymized, multimodally linked cardiological routine data from adult, non-pregnant patients who underwent cardiological diagnostic or therapeutic care between 20
Sponsors
Universitätsklinikum Heidelberg
Eligibility
Sex/Gender
All
Age
18 Years to No maximum
Inclusion criteria
Inclusion criteria: Adult patients who underwent cardiological diagnostic procedures at the Department of Internal Medicine III – Cardiology, Angiology and Pulmonology of Heidelberg University Hospital (UKHD) between 2006 and 2025.
Exclusion criteria
Exclusion criteria: Pregnant patient
Design outcomes
Primary
| Measure | Time frame |
|---|---|
| As this is a retrospective, exploratory methodological study, no patient-related endpoints in the sense of a prospective clinical efficacy assessment are defined. The endpoints serve the methodological and diagnostic evaluation of AI-based analysis methods on routine clinical data. Primary Endpoint 1: Detection of Cardiological Pathologies by AI Models WHAT: Automated detection of predefined cardiological findings, including: regional or global myocardial wall motion abnormalities, indications of ischemic heart disease, clinically relevant arrhythmias, detection of cardiac implants (e.g., pacemakers, implantable cardioverter-defibrillators). WHEN: Retrospectively at the time of each clinical data collection during routine care (2006–2025); there are no fixed study time points (T0/T1); analysis is based on the available examination time points. HOW: Multimodal AI-based analysis of existing routine data, including echocardiography, electrocardiograms, cardiac CT, structured electronic health record data, and unstructured clinical texts. The reference standard is the documented clinical evaluation in the patient record (e.g., echocardiography report, discharge letter). Primary Endpoint 2: Quantitative Cardiac Functional Parameters WHAT: Automated extraction of specific quantitative cardiac functional parameters, including: ventricular volumes, left ventricular ejection fraction, additional segmentation-based measures. WHEN: Retrospectively at the time of each imaging examination (e.g., transthoracic echocardiography or cardiac CT). HOW: AI-based image analysis (segmentation and parameter extraction) from existing imaging data; comparison with clinically documented reference values where available. | — |
Secondary
| Measure | Time frame |
|---|---|
| Secondary Endpoint 1: Association Between AI-Based Findings and Clinical Events WHAT: Occurrence of clinical events, including: cardiovascular or all-cause mortality, cardiological rehospitalizations, myocardial infarctions, repeat imaging examinations (e.g., follow-up transthoracic echocardiography or cardiac CT). WHEN: Retrospectively over the time course following the respective index examination within the available clinical follow-up period. HOW: Analysis of structured electronic health record data (diagnosis and procedure codes) and unstructured clinical texts; statistical correlation with AI-generated predictions. Secondary Endpoint 2: Model Performance and Robustness WHAT: Performance of the developed AI models, including discrimination ability, sensitivity, specificity, and robustness against missing data modalities. WHEN: During retrospective model development and validation. HOW: Quantitative evaluation using standardized metrics (e.g., ROC-AUC, F1-score) on internal and external datasets. | — |
Countries
Germany, Netherlands, Spain
Contacts
Public ContactSandy Engelhardt
Universitätsklinikum Heidelberg
Outcome results
None listed