Scholarlink

<p>In pharmaceutical manufacturing, equipment is often used for multiple products following a single standard cleaning process regardless of the previously manufactured product. Despite this, residue levels are frequently non-detectable. Regulatory guidelines, such as the EMA Q&amp;A on HBELs, mandate cleaning and visual inspection after each use, with analytical testing required after product changeovers unless otherwise justified. Consequently, cleaning validation (CV) data is typically left-censored, limited in sample size, and subject to multiple upper specification limits (USLs). These realities limit the applicability of traditional parametric capability indices and reduce the robustness of nonparametric methods. Moreover, they hinder the effective use of historical CV data in AI applications, which depend on large, standardized datasets for model training. To address these challenges, we propose a USL-normalization method that standardizes all residue measurements to a unified USL_Pct = 100. This transformation enables the aggregation of USL-heterogeneous datasets and supports robust nonparametric analysis. The cleaning process CQA (critical quality attribute) level upper capability index Ppu is calculated using the&#xa0;KDEDPonUSLND&#xa0;algorithm (kernel density estimation derived percentiles on USL-normalized data), which demonstrates improved robustness compared to traditional approaches. The cleaning process level Ppu is evaluated using either CQAWWC (CQA-wise worst case) or CQAWP (CQA-wise pooling) strategies, both integrated with the KDEDPonUSLND algorithm. The resulting CQAWWC_BAKEDPonUSLND and CQAWP_BAKEDPonUSLND models support an AI-driven CV Stage 3 Continued Cleaning Process Verification/Monitoring system, aligned with regulatory expectations.</p> Graphical Abstract <p></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

Specification Limit Normalization for Nonparametric Process Capability Analysis and Its Application in Cleaning Validation with AI-Enabled Monitoring Models

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月26日

查看全文
  • Xiande Yang,
  • Wendy Lou,
  • Mohamed Chan

摘要

In pharmaceutical manufacturing, equipment is often used for multiple products following a single standard cleaning process regardless of the previously manufactured product. Despite this, residue levels are frequently non-detectable. Regulatory guidelines, such as the EMA Q&A on HBELs, mandate cleaning and visual inspection after each use, with analytical testing required after product changeovers unless otherwise justified. Consequently, cleaning validation (CV) data is typically left-censored, limited in sample size, and subject to multiple upper specification limits (USLs). These realities limit the applicability of traditional parametric capability indices and reduce the robustness of nonparametric methods. Moreover, they hinder the effective use of historical CV data in AI applications, which depend on large, standardized datasets for model training. To address these challenges, we propose a USL-normalization method that standardizes all residue measurements to a unified USL_Pct = 100. This transformation enables the aggregation of USL-heterogeneous datasets and supports robust nonparametric analysis. The cleaning process CQA (critical quality attribute) level upper capability index Ppu is calculated using the KDEDPonUSLND algorithm (kernel density estimation derived percentiles on USL-normalized data), which demonstrates improved robustness compared to traditional approaches. The cleaning process level Ppu is evaluated using either CQAWWC (CQA-wise worst case) or CQAWP (CQA-wise pooling) strategies, both integrated with the KDEDPonUSLND algorithm. The resulting CQAWWC_BAKEDPonUSLND and CQAWP_BAKEDPonUSLND models support an AI-driven CV Stage 3 Continued Cleaning Process Verification/Monitoring system, aligned with regulatory expectations.

Graphical Abstract