药物警戒挑战的答案：人工智能驱动的不良事件检测

The Answer To YourPharmacovigilanceChallenges: AI-PoweredAdverse Event Detection ANURADHA PRABHAKAR, Associate Director, Product Management, Vigilance Detect, IQVIASANMUGAM ARAVINTHAN, Senior Director, Product Management, Vigilance Detect, IQVIA Table of contents Introduction1What concerns accompany AI implementation in drug and device safety?1What is Vigilance Detect and how does it improve AE detection?2What is IQVIA’s approach to AI governance?3What’s next?3About the authors4About IQVIA4 Introduction As AI and Machine Learning (ML) technologies expandtheir reach across a wide range of industries, thePharmacovigilance (PV) and device safety space is noexception. Due to the growing number of sources forAdverse Events (AE), pharma companies are navigatingpreviously unprecedented volumes of data from whichthey must identify AEs and extract Individual Case SafetyReports (ICSRs) to protect patient safety and ensureregulatory compliance. As a result, many companiesare turning to AI-powered solutions to help safety professionals process these large data volumes moreefficiently and accurately. But, to embrace the power of AI-enabled safetysolutions, pharma companies must be confident in theirchosen system’s ability to produce reliable results andmaintain alignment with the latest regulatory guidelines.The solution? Activating a proven, AI-supported PVplatform that is backed by rigorous AI governance andexpert-in-the-loop support to improve the accuracy andprecision of AE detection. What concerns accompany AI implementation in drug anddevice safety? Despite the efficiencies demonstrated by AI, pharma companies may be hesitant to use it in a safety-regulatedenvironment due to three primary concerns: Regulatory challenges Lack of reliability Data privacy and security risks The speed of AI’s evolutionand integration has outpacedconcrete and prescriptiveregulation. As a result, theindustry is navigating anevolving regulatory landscape.To protect their patients, theirproducts, and their businesses,stakeholders must implementAI in a way that is compliantwith current regulationsand flexible to adapt as newguidance emerges. Drug and device safety teamsdeal in highly sensitive patientdata that companies have a dutyto protect. Patient data must bekept secure to remain compliantwithGeneral Data ProtectionRegulation (GDPR)andHIPAAlaws, amongst others. ActivatingGenerative AI (GenAI) toprocess and interact with thisdata inherently raises concernsaround how to sufficientlymaintain data privacy andprotection and mitigatepotential security risks. Inaccuracies in AE identification,case narratives, and/orsignal detection can havedirect consequences forpatient safety. Generative AImodels can “hallucinate” —producing factually inaccurateor misleading outputs —which can present a majorrisk to patient safety andregulatory compliance. Vigilance Detect was originally purpose-built for safetysurveillance by safety professionals to accurately identifyand classify AEs, special scenarios, and product riskswithin a wide variety of data sources, including CRM,social media, Chatbot, audio, and PDF data. It does soacross millions of records annually and over the pastdecade in production, it has yielded results in 60-90%efficiencies over human review. Now, powered by GenAI,Vigilance Detect demonstrates dramatically enhancedprecision and accuracy. Though concerns about AI in patient safety are valid,they are not insurmountable. Global policy initiativesand regulatory frameworks are rapidly coming onlineto guide AI compliance in safety workflows. TheU.S.Food and Drug Administration(FDA) andU.K. Medicineand Healthcare products Regulatory Agency(MHRA)have released draft frameworks on AI usage, and theCouncil for International Organizations of MedicalSciences (CIOMS) XIV Working Groupis currentlydeveloping a framework that focuses on the use of AIin PV and safety. Companies do not have to navigate AIimplementation and compliance in a silo. With the rightmulti-disciplinary technology partner, they can use AI toimprove AE detection. This type of partner can provideessential support through embedded AI governance,experienced experts-in-the-loop, and strategic selectionof appropriate, fit-for-purpose use cases. By the numbers, Vigilance Detectpowered by GenAI has exhibited theability to reduce false positive ratesby nearly 80%. In one case study of2500 audio files, a Vigilance Detectprocess supported by GenAI resultedin 94% precision, 99% accuracy,and an 81% reduction in manualreview for the client. The platformstreamlines safety activities andaccelerates processing by reducingredundant data and manual review. What is Vigilance Detectand how does it improveAE detection? IQVIA’s Vigilance Detect data safety platform is aCFRPart 11validated Software as a Service (SaaS) safetysolution that leverages validated proprietary safetyontologies to perform a first layer of review to surfacesafety risks in large and diverse data sets. GenAI isused