Scholarlink

<p>Clinical laboratories rely on internal quality control (QC) systems to ensure the reliability of examination results that support clinical decision-making. Conventional QC approaches, including Westgard multirule systems and Sigma-based evaluation, are effective in detecting analytical deviations but provide limited support for structured interpretation and documentation. Variability in how QC events are interpreted and recorded may affect documentation consistency and traceability within ISO 15189-aligned environments. This study evaluated a ChatGPT-5-based digital quality assistant (DQA) implemented as a supervised decision support tool for QC interpretation and documentation in clinical chemistry laboratories. A structured, two-site evaluation was conducted using examination-phase QC interpretation events (n = 1000) and documentation workflow records. Performance was assessed across four predefined objectives: interpretation agreement with expert adjudication and established rule-based approaches, documentation quality and traceability, system reproducibility and explainability, and operational workflow impact. All outputs were reviewed and verified by laboratory personnel prior to use. The DQA demonstrated higher agreement with expert-adjudicated interpretations, achieving 90.0% classification accuracy compared with 79.6% for algorithmic rule-based interpretation and 74.0% for LIS auto-flagging, with a Cohen’s κ of 0.77. Classification error rates were lower (10.0% vs 20.4% and 26.0%), and false negative and false positive rates were reduced (11.6% and 9.3%, respectively). Performance remained stable across instruments (90.97% vs 89.12%) and control levels (88.85%–91.21%). Documentation completeness increased from 45.1% to 73.1%, traceability scores improved from 3.01 to 4.27, and documentation timeliness decreased from 53.1 to 22.1&#xa0;min. Reproducibility testing showed 98.6% identical outputs under repeated conditions. Operational indicators improved, including reductions in corrective action and preventive action (CAPA) closure time (17.24 to 8.04&#xa0;days) and rerun rates (38.8% to 21.2%). When used under continuous human supervision, the ChatGPT-5-based digital quality assistant (DQA) showed potential to enhance QC interpretation consistency, improve documentation quality and traceability, and support workflow efficiency. These findings suggest that ChatGPT-assisted DQA systems may serve as complementary tools alongside established statistical QC frameworks within ISO-aligned clinical laboratory practice.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Explainable language model reasoning for laboratory quality control: validation of the ChatGPT-5 as a digital quality assistant

  • 期刊论文

  • 发表日期：

    2026年4月20日

查看全文
  • Ahmed Naseer Kaftan

摘要

Clinical laboratories rely on internal quality control (QC) systems to ensure the reliability of examination results that support clinical decision-making. Conventional QC approaches, including Westgard multirule systems and Sigma-based evaluation, are effective in detecting analytical deviations but provide limited support for structured interpretation and documentation. Variability in how QC events are interpreted and recorded may affect documentation consistency and traceability within ISO 15189-aligned environments. This study evaluated a ChatGPT-5-based digital quality assistant (DQA) implemented as a supervised decision support tool for QC interpretation and documentation in clinical chemistry laboratories. A structured, two-site evaluation was conducted using examination-phase QC interpretation events (n = 1000) and documentation workflow records. Performance was assessed across four predefined objectives: interpretation agreement with expert adjudication and established rule-based approaches, documentation quality and traceability, system reproducibility and explainability, and operational workflow impact. All outputs were reviewed and verified by laboratory personnel prior to use. The DQA demonstrated higher agreement with expert-adjudicated interpretations, achieving 90.0% classification accuracy compared with 79.6% for algorithmic rule-based interpretation and 74.0% for LIS auto-flagging, with a Cohen’s κ of 0.77. Classification error rates were lower (10.0% vs 20.4% and 26.0%), and false negative and false positive rates were reduced (11.6% and 9.3%, respectively). Performance remained stable across instruments (90.97% vs 89.12%) and control levels (88.85%–91.21%). Documentation completeness increased from 45.1% to 73.1%, traceability scores improved from 3.01 to 4.27, and documentation timeliness decreased from 53.1 to 22.1 min. Reproducibility testing showed 98.6% identical outputs under repeated conditions. Operational indicators improved, including reductions in corrective action and preventive action (CAPA) closure time (17.24 to 8.04 days) and rerun rates (38.8% to 21.2%). When used under continuous human supervision, the ChatGPT-5-based digital quality assistant (DQA) showed potential to enhance QC interpretation consistency, improve documentation quality and traceability, and support workflow efficiency. These findings suggest that ChatGPT-assisted DQA systems may serve as complementary tools alongside established statistical QC frameworks within ISO-aligned clinical laboratory practice.