Ishiguro Chieko, Nonaka Takahiro
Laboratory of Clinical Epidemiology, Department of Data Science, Center for Clinical Sciences, Japan Institute for Health Security, Tokyo, Japan.
Department of Clinical Research Governance, The University of Tokyo Hospital, Graduate School of Medicine, The University of Tokyo, Tokyo, Japan.
Front Pharmacol. 2025 Sep 2;16:1642490. doi: 10.3389/fphar.2025.1642490. eCollection 2025.
In pharmacoepidemiological research, misclassification is a concern with claims-based algorithms (also called computable phenotypes). Validating them is crucial, particularly within regulatory settings. However, the extent of their application remains unclear globally.
This study aimed to investigate the frequency and trends of validated claims-based algorithms use in post-marketing database studies.
We reviewed all Japanese risk management plans published until January 2023, identifying four issue types [Effectiveness Issues (EI), Important Identified Risks (IIR), Important Potential Risks (IPR), and Important Missing Information (IMI)] that were planned to use a claims-based algorithm in post-marketing database studies. We then calculated the proportion of issues intending to use a validated claims-based algorithm, and performed subgroup analyses by issue type.
Of 68 issues (3 EI, 47 IIR, 13 IPR, 5 IMI), 15 (22.1%) planned to use a validated algorithm, all for outcome definitions; 10 to conduct new validation studies and 5 to refer to existing studies, including studies with high positive predictive value and sensitivity. Subgroup analyses by issue type showed that the proportions were 100% for EI, 17.0% for IIR, 30.8% for IPR, and 0% for IMI.
Validated algorithm use was the highest for effectiveness issues but limited for safety, suggesting that results from these post-marketing database studies for safety issues may not provide sufficient evidence, highlighting the need to promote the use of validated claims-based algorithms. Future studies should use more recent data, compare the use of validated algorithms between Japan and other countries, and explore barriers to their adoption.
在药物流行病学研究中，错误分类是基于索赔算法（也称为可计算表型）所关注的问题。对其进行验证至关重要，尤其是在监管环境中。然而，其在全球范围内的应用程度仍不明确。
本研究旨在调查在上市后数据库研究中使用经过验证的基于索赔算法的频率和趋势。
我们回顾了截至2023年1月发布的所有日本风险管理计划，确定了四种问题类型[有效性问题（EI）、已识别的重要风险（IIR）、潜在重要风险（IPR）和重要缺失信息（IMI）]，这些问题类型计划在上市后数据库研究中使用基于索赔的算法。然后，我们计算了打算使用经过验证的基于索赔算法的问题比例，并按问题类型进行亚组分析。
在68个问题（3个EI、47个IIR、13个IPR、5个IMI）中，15个（22.1%）计划使用经过验证的算法，均用于结果定义；10个进行新的验证研究，5个参考现有研究，包括具有高阳性预测值和敏感性的研究。按问题类型进行的亚组分析表明，EI的比例为100%，IIR为17.0%，IPR为30.8%，IMI为0%。
经过验证的算法在有效性问题上的使用最高，但在安全性方面有限，这表明这些上市后数据库研究针对安全性问题的结果可能无法提供充分证据，凸显了推广使用经过验证的基于索赔算法的必要性。未来的研究应使用更新的数据，比较日本和其他国家经过验证的算法的使用情况，并探索采用这些算法的障碍。
