重新定义上市后监控：人工智能如何改变质量、安全和全球合规

Redefining Post-MarketSurveillance: How AITransforms Quality, Safety, Table of contents Introduction1Strengthening complaint handling: Volume, quality, andtimeliness2Improving Adverse Event Reporting (AER): Consistency and report quality2Expanding data gathering: Structured and unstructuredsources3Understanding new failure modes and human factorinsights4Process optimization: Efficiency, consistency, andcompliance5The IQVIA SmartSolve®difference6Conclusion6About the author6 Introduction Post-Market Surveillance (PMS) has become one of thefastest-evolving areas within the medical technology andpharmaceutical industries. Global regulatory authoritiesexpect timely visibility into emerging risks, clearerevidence linking real-world data to a product’s clinicalperformance and risk profile, and stronger connectivity AI can be a significant enabler for quality, regulatory,and safety professionals working in PMS, helping teamstransform data into information and action. A range ofAI solutions deployed to support activities mandatedby global regulation can amplify the human-in-the-loop professional, enabling teams to keep pace with vary by organization, division, and product type. Humanoversight remains essential, but the time requiredfor initial drafting decreases substantially, increasing Strengthening complainthandling: Volume, quality, Complaint handling remains central to PMS, butthe volume and variability of incoming informationcan overwhelm even well-structured, experienced Improving Adverse EventReporting (AER): Consistency Adverse event reporting demands consistency, clarity,and timeliness in submissions to global authorities.Variability in narrative text and coded fields across Automated intake and normalization AI tools can parse free text narratives from call-centerinteractions, emails, social media, and technical servicereports and automatically map them to standardized Consistent coding AI can recommend appropriate problem codes andpatient-impact descriptors based on historical patterns.Reviewers retain full decision authority, while automated Language translation capabilities When organizations use language-translation capabilities,cases can be captured in a local language while allowingan investigator to review the information in another Enhanced drafting with GenAI GenAI can prepopulate key sections of AER reports usingapproved templates and established precedents. Thisimproves consistency, reduces administrative burden,and helps global teams maintain alignment in structure Case triage and clustering Machine-learning techniques can identify similar cases,group related events, and prioritize those likely to havethe greatest safety relevance. This allows teams to focus Timely completion By automating checks for completeness and coherence,AI reduces the risk of missing data elements and shortensreport preparation time. This supports compliance with Generative AI for investigation summaries Generative AI can accelerate investigations by draftingconcise summaries that integrate historical information, Unstructured data unlocking Expanding datagathering: Structured and NLP-driven AI enables organizations to analyze a broaderrange of sources essential to a holistic understandingof product quality and safety. These include call-centeraudio and transcripts, PDF field reports and service Traditional PMS processes rely heavily on structureddata from complaint systems and vigilance databases.However, many early signals originate from unstructured Always on surveillance AI enables continuous, real-time monitoring of sourcesrather than periodic manual reviews. Anomalies orunusual patterns can be flagged promptly, giving Structured data optimization NLP AI can support normalization, deduplication, andtrend identification across structured datasets, includingcomplaint logs, NC and CAPA records, technical service There are a wide range of data sources for AI/NLP solutions Identifying the most suitable ‘proactive’ data sources is a key success criterion complications, off-label use concerns, or unexpectedperformance challenges. Patterns related to user errors,misinterpretation of instructions, and training gaps canprovide valuable insights into where an organization Understanding newfailure modes and human Major product issues often emerge first as weak signals:a few unusual service notes, recurring user stepsidentified in call-center interactions, or an early reportof an issue not yet known to be systematic. AI can help Cross source signal correlation Connecting trends across multiple sources, includingliterature, service logs, complaints, and customerfeedback, supports early hypothesis generation for Literature and scientific mining AI can continuously scan scientific publicationsand clinical reports for early indicators of new Patient safety, product quality, market access Content review •Literature publications Market Insights •Social media channels•Internal and external