Multi-center Validation Study of a Large Language Model-based Intelligent Agent for Blood Cell Analysis

I. Study Background: Currently, in most medical institutions, the review of blood cell analysis still heavily relies on manual verification by laboratory staff. This process requires a comprehensive analysis of instrument parameters, alarm flags, historical comparison results, and, when necessary, microscopic examination. However, with the increasing volume of test samples and the high concentration of review tasks during peak hours, the traditional manual review model increasingly shows problems such as prolonged turnaround time (TAT), uneven workload distribution, and decreased consistency in reviews. In recent years, intelligent review systems based on Large Language Models (LLM) have shown potential in analyzing abnormal results and stratifying sample risks by integrating preset rules, clinical diagnostic information, and multi-dimensional laboratory data, which is expected to optimize the review workflow.

II. Study Objective: To evaluate the difference in overall sample review turnaround time between the experimental process and the control process during the formal study phase, and to test its superiority.

III. Subjects: The investigators need to recruit approximately 20,000 subjects, regardless of age or gender.

IV. Study Procedures: If participants agree to participate in the study, participants only need to allow us to use participants test results after participants have completed your routine blood test (CBC).

V. Risks and Benefits:

1. Risks: This study poses no risk to the subjects. The investigators only use the result data of patients after participants have had their routine blood test; there is no need for patients to undergo additional blood draws.
2. Benefits: It will shorten the turnaround time for routine blood test results and share the workload of doctors in reviewing these results.

VI. Privacy: All of participants information will be kept strictly confidential and will only be used for this scientific research.

Study Overview

• Status: Not yet recruiting
• Conditions: Complete Blood Count Review
• Intervention / Treatment: Other: LLM-Assisted Review Group

• Study Type: Observational
• Enrollment (Estimated): 20000

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Child; Adult; Older Adult
• Accepts Healthy Volunteers: Yes

• Sampling Method: Non-Probability Sample

Study Population

The study subjects were consecutive individuals undergoing routine blood tests at each center, with no restrictions on gender or age. After inclusion, samples entered the corresponding review process based on the study week, with preliminary and secondary reviews conducted by predefined workflows and qualified personnel, respectively.

Description

Inclusion Criteria:

• Subjects who underwent routine blood tests in the outpatient, emergency, or inpatient departments of the participating centers during the study period.

Corresponding samples must have complete instrument results, review trails, and report timestamp records.

Approved for inclusion by the Ethics Committee.

Exclusion Criteria:

• Samples collected during periods of instrument malfunction or interface transmission anomalies.

Missing key research data, particularly samples where the final review conclusion or key timestamps cannot be confirmed.

Subjects or their legal representatives explicitly refuse to participate in the study.

Study Plan

How is the study designed?

Number of groups / cohorts

2

Cohorts and Interventions

Group / Cohort	Intervention / Treatment
LLM-Assisted Review Group	Other: LLM-Assisted Review Group; This study introduces an intelligent auxiliary review system based on a medical Large Language Model (LLM), aimed at optimizing the traditional CBC report review process. The core functions and intervention mechanisms are as follows: Multi-source Data Integration: The system integrates seamlessly with the Laboratory Information System (LIS) to automatically retrieve patient demographics (age, sex), current CBC indices, historical results, and clinical diagnoses. Deep Analysis and Anomaly Detection: Unlike traditional rule-based auto-verification, this system leverages the reasoning capability of LLMs to perform multidimensional clinical logic checks. It identifies out-of-range values and interprets their clinical significance by combining them with patient history (e.g., distinguishing physiological fluctuations from pathological changes).
Standard Manual Review Group

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Time Frame
Overall Report Turnaround Time	one year

Collaborators and Investigators

• Sponsor: Huashan Hospital
• Collaborators: Peking University First Hospital; Lishui hospital of Zhejiang University

Study record dates

Study Major Dates

• Study Start (Estimated): May 14, 2026
• Primary Completion (Estimated): August 31, 2027
• Study Completion (Estimated): August 31, 2027

Study Registration Dates

• First Submitted: May 19, 2026
• First Submitted That Met QC Criteria: May 19, 2026
• First Posted (Actual): May 26, 2026

Study Record Updates

• Last Update Posted (Actual): May 28, 2026
• Last Update Submitted That Met QC Criteria: May 24, 2026
• Last Verified: May 1, 2026

More Information

Terms related to this study

• Other Study ID Numbers: KY2026-084