Life Sciences Horizons Brochure 2025 - Flipbook - Page 15
14
2025 Horizons Life Sciences and Health Care
Using AI for device postmarket surveillance to meet FDA and EU obligations (continued)
Natural Language Processing (NLP).
Anomaly detection models.
Predictive analytics.
PMS generates vast amounts of free-text data, including MDR
reports, customer complaints, health care provider feedback,
online patient forums, and social media discussions.
Traditionally, analyzing this data required manual review, making
it slow and prone to human bias. Natural Language Processing
automates this process by extracting meaningful insights from
unstructured text. For example, NLP is capable of taking
unstructured information and creating order by mining text and
performing sentiment analysis by scanning complaint logs,
identifying recurring issues, emotional tone, and negative
sentiment patterns in user feedback.
Anomaly detection models are capable of identifying unusual
patterns or unexpected trends in complaint data that may indicate
underlying safety concerns. These models learn from historical
complaint data and identify outliers that deviate from expected
behavior, enabling early intervention. For example, unsupervised
learning models detect rare or unusual complaint patterns
without predefined rules, uncovering emerging risks. ML
algorithms can be set to continuously analyze incoming
complaint data, triggering alerts when deviations occur. Finally,
AI can be tasked with examining relationships between reported
failures to identifying systemic issues across multiple devices or
manufacturers.
By its nature, traditional complaint monitoring is reactive and
must be coupled with the company’s obligation for risk
management, which is designed to be predictive of possible
failure. With AI’s power of predictive analytics, manufacturers
may be able to better forecast potential failures, allowing
manufacturers to take earlier and more effective preventive
action. Using time-series forecasting, AI can analyze historical
complaint trends to predict when and where future failures might
occur. Risk scoring models could be used to assigns risk scores to
devices, prioritizing those most likely to experience defects or
adverse events. Failure mode prediction models could also be
used to supplement existing risk management tools to correlate
real-world usage data (e.g., device sensor readings, hospital
reports) with past complaints to anticipate malfunction risks.
NLP is also capable of identifying key medical terms, device
names, symptoms, and adverse event descriptions to classify
complaint types. AI can also group similar complaints, helping
manufacturers detect emerging safety signals early. Finally,
without the limits of language, NLP allows companies to monitor
complaints globally, translating and analyzing reports from
different languages and geographies in real time.
As AI becomes more integrated into regulatory compliance and
medical device monitoring, its potential to reduce the burdens of
monitoring adverse events, streamline reporting, and enhance
patient safety will only grow. After validation as part of the quality
management system, the combination of NLP, anomaly detection,
and predictive analytics allows for a more proactive, intelligent
approach to medical device PMS.
Fabien Roy
Partner
Brussels
Jodi Scott
Partner
Denver
