Using Retinal Photograph Based AI to Predict Incident Coronary Heart Disease (DeepCHD Plus)

Tsinghua University

Status: Not yet enrolling

Conditions: Coronary Heart Disease (CHD)

Treatments: Diagnostic Test: AI-derived probability of coronary heart disease.; Diagnostic Test: PCEs derived ASCVD risk

Study type: Interventional

Funder types: Other

Identifiers

NCT06695273

DeepCHD Plus

Details and patient eligibility

About

To determine whether an integrated retinal AI decision support can improve predictive accuracy of coronary heart disease (CHD), the investigators are conducting a randomized controlled study of AI guided prediction of CHD compared to clinical prediction by physicians (e.g., usingPCEs), both using clinical intuition as baseline.

Full description

This is a randomized controlled trial (RCT) evaluating the effectiveness of an AI-based decision support tool in CHD risk prediction and decision making by physicians. Prospective cohort study participant cases will be randomly assigned to either guideline group (e.g., PCEs) or AI group after baseline assessment (clinical intuition):

There are three settings: (1) Clinical Intuition (baseline assessment) Physicians' make decision about prevention strategy initiation (e.g., statin initiation) without any external assistance. Assessment relies solely on the physician's clinical judgment and experience. (2) Guideline-Based Group (Guideline Group) Physicians use a PCE table to calculate the 10 year ASCVD risk. This approach aligns with current clinical guidelines to assist in decision-making. (3) AI-Assisted Group (AI Group) Physicians receive CHD probability estimates from an AI model based on retinal photographs. The AI tool provides individualized obstructive CHD probabilities, leveraging retinal biomarkers associated with cardiovascular risk.

Primary Objective To evaluate whether AI-guided decision support could improves diagnostic accuracy of CHD to a greater extent than standard clinical assessments, both compared to clinical intuition. The accuracy could be assessed by the extent of prevention initiation (e.g., prescribing statins) corresponding with actual CHD outcomes observed.

Secondary Objective To assess whether AI-guided decision support reduces the time required to complete CHD assessments and decision making.

Participants, Readers and Randomization:

Participants: Participants in prospective cohort studies, with 10-year follow up.

Readers: Physicians performing evaluations of CHD probability and make primary prevention recommendations.

Randomization: Participants will be randomized into one of the groups (PCEs or AI) after clinical assessment at baseline using block randomization to ensure balanced group sizes.

Enrollment: 1,570 estimated patients

Sex: All

Ages: 40 to 75 years old

Volunteers: Accepts Healthy Volunteers

Inclusion criteria

• Individuals without uncontrolled vascular risk factors
• Age range: 40-75 years old
• Can accept and cooperate with the examination and potential follow-up work after being selected for clinical trials

Exclusion criteria

• Severe lung disease and cancer or surgery patients
• Statin user or pre-existing cardiovascular disease
• Individuals with severe liver and kidney dysfunction and electrolyte imbalance

Trial design

Primary purpose: Screening

Allocation: Randomized

Interventional model: Parallel Assignment

Masking: Single Blind

---

1,570 participants in 2 patient groups

AI-Assisted Group (AI Group)

Experimental group

Description:

Physicians receive CHD probability estimates from an AI model based on retinal photographs. The AI tool provides individualized CHD probabilities, leveraging retinal biomarkers associated with cardiovascular risk.

Treatment:

Diagnostic Test: AI-derived probability of coronary heart disease.

Guideline-Based Group (Guideline Group)

Active Comparator group

Description:

Physicians use a PCE calculator to calculate the 10 year ASCVD risk. This approach aligns with current clinical guidelines to assist in decision-making.

Treatment:

Diagnostic Test: PCEs derived ASCVD risk