Comparing AI vs. Human Reading for Lung Cancer Screening : A Cluster Randomized Controlled Trial

Not Applicable

Recruiting

• Conditions: Lung Cancer; Artificial Intelligence (AI); Randomized Controlled Trial

• Registration Number: NCT06988579
• Lead Sponsor: The First Affiliated Hospital of Guangzhou Medical University

Brief Summary

AI diagnostic systems show great promise for improving lung cancer screening in community healthcare settings. While not originally designed for primary care, these tools demonstrate capabilities in nodule detection and workflow optimization. However, their effectiveness in resource-limited community centers requires thorough evaluation.

This cluster RCT compares AI-assisted versus manual CT interpretation across community health centers. Expert radiologists will establish reference standards, while an independent committee blindly evaluates cases from both groups. The study assesses diagnostic accuracy, operational efficiency, and cost-effectiveness, with blinded analysts resolving discrepancies through consensus to ensure reliable results.

Detailed Description

Artificial intelligence (AI) technologies, particularly advanced medical imaging analysis systems like the AI diagnostic platform evaluated in this study, demonstrate significant potential for enhancing lung cancer screening programs in community healthcare settings.Although this AI system was not originally designed specifically for primary care implementation, it has shown promising capabilities in various clinical applications, including nodule detection, malignancy risk stratification, and workflow optimization in radiology departments. However, its effectiveness in improving screening accuracy and operational efficiency in resource-limited community health centers remains to be thoroughly investigated.

Lung cancer screening and diagnosis involve complex clinical processes,including image interpretation, risk factor assessment, and follow-up decision-making. Implementing AI tools like this diagnostic platform in community screening programs could potentially improve detection rates, standardize interpretations, and optimize resource allocation. Nevertheless, the system has not been rigorously validated for use in primary care settings and may carry limitations in generalizability across diverse patient populations and imaging equipment variations. Inappropriate implementation could lead to diagnostic errors or inefficient resource utilization. Therefore, evaluating how such AI systems can effectively support community-based screening while maintaining diagnostic accuracy and cost-effectiveness is of paramount importance.

In this cluster randomized controlled trial, participating community health centers will be allocated to either an AI-assisted interpretation group or a conventional manual interpretation group. All screening cases will undergo standardized low-dose CT imaging, with results interpreted through the respective group's designated method. A panel of three expert radiologists will establish reference standards for all cases, while an independent review committee will blindly evaluate a subset of cases from both groups to assess interpretation consistency. The evaluation will focus on diagnostic performance metrics, operational efficiency parameters, and cost-effectiveness indicators. Two separate analyst teams, blinded to group assignments, will process and compare the outcomes using predefined statistical methods, with any discrepancies resolved through consensus discussions to ensure data reliability.

Study Timeline

• Status Verified: 2025-05
• Study First Submitted: 2025-05-16
• Study Start: 2025-05-19
• Study First Submitted QC: 2025-05-23
• Last Update Submitted: 2025-05-23
• Study First Posted: 2025-05-25
• Last Update Posted: 2025-05-25
• Primary Completion: 2025-10-31
• Study Completion: 2026-01-07

Recruitment & Eligibility

• Status: RECRUITING
• Sex: All
• Target Recruitment: 95916

Inclusion Criteria

1. Aged 45-74 years
2. Permanent resident of participating study communities
3. No prior history of lung cancer and no lung cancer screening within the past 3 months
4. Able to comprehend and voluntarily sign informed consent, with willingness to participate in long-term follow-up

Exclusion Criteria

1. Individuals with a confirmed diagnosis of lung cancer
2. Those with severe comorbidities contraindicating CT imaging
3. Inability to understand study protocols or provide informed consent due to cognitive impairment
4. Concurrent participation in other clinical trials that may interfere with study outcomes
5. Unable to comply with follow-up requirements

Study & Design

• Study Type: INTERVENTIONAL
• Study Design: PARALLEL

Primary Outcome Measures

Name	Time	Method
Diagnostic Accuracy	One year after entry	Sensitivity and Specificity: Comparison of AI-assisted versus manual interpretation in detecting malignant pulmonary nodules, validated against histopathological confirmation or 12-month clinical follow-up.; Early Detection Rate: Proportion of stage I/II lung cancers correctly identified by each method.
Interpretation Consistency	One year after entry	Inter-reader Agreement: Measured by Cohen's kappa (κ) between AI-assisted radiologists and the independent review committee (IRC).; Intra-reader Variability: Consistency of nodule classification in repeat readings (subset analysis).

Secondary Outcome Measures

Name	Time	Method
Cost-Effectiveness	One year after entry	Resource Utilization: Comparative analysis of staffing, equipment, and follow-up costs per detected cancer case.; Incremental Cost per QALY (Quality-Adjusted Life Year): Long-term economic impact modeling.

Trial Locations

Locations (1)

the First Affiliated Hospital of Guangzhou Medical University,; 🇨🇳 Guangzhou, Guangdong, China