Page 11 - ITPS-8-3

INNOSC Theranostics and
Pharmacological Sciences AI in medical device safety

3.1. Real-time data monitoring and predictive and anomalies that may indicate device malfunctions. ML
analytics in patient safety algorithms have been applied in clinical trials to predict
AI has proven to be a highly effective tool for improving adverse outcomes, including fatalities and other critical
patient safety through predictive analytics and continuous events. For example, one study developed five ML models
using data from 28,340 clinical trial reports. The best-
real-time data monitoring. AI-driven systems enable performing model, logistic regression, achieved a receiver
early identification of potential safety issues and prompt 30
risk-reduction measures by automating the analysis of operating characteristic score of 0.7344. This finding
real-world data, such as patient outcomes, adverse event demonstrates how ML can assist researchers in estimating
reports, and electronic health records. These systems, risks and predicting unfavorable events, thereby facilitating
8,9
leveraging big data analytics and ML algorithms, provide the development of more effective measures to protect
actionable insights into device safety profiles and emerging participants.
trends by identifying patterns and anomalies. Adverse events are classified using various ML
techniques, such as ensemble learning, unsupervised
In both clinical and home settings, AI-driven systems
have significantly improved the accuracy of real-time learning, and deep learning. Algorithms such as Random
monitoring and predictive capabilities through the Forests, Support Vector Machines, and Neural Networks
27
application of ML techniques. These systems have the analyze complex health data to classify these occurrences.
capacity to continuously collect and evaluate data from These classification models are essential in the field of
medical devices by identifying irregularities and managing
various sources, including physiological signals, enabling risks in interconnected medical systems. 31
timely intervention and the early detection of potential
health issues. 27,28 To ensure accurate event detection, ML models require
extensive data validation during training. The training
AI technologies also facilitate the identification
of patient decline in its pre-symptomatic stages by data is thoroughly examined and adjusted, frequently
converting streaming clinical data into real-time visual through cross-validation methods, to maximize model
performance under real-world conditions. The reliability
risk assessments, thereby improving triage processes and of these models in detecting and reporting safety issues
patient care strategies. 29
in medical devices is enhanced by continuous updates
Moreover, AI systems use predictive models based on and reassessments based on new data inputs. As AI
31
real-time patient data to generate automated alerts for continues to evolve, further advancements in the efficiency
health-care providers. These notifications are essential and personalization of medical device safety monitoring
for preventing complications, ensuring timely responses, are anticipated, particularly with the integration of new
and improving patient outcomes. When abnormalities are domains such as proteomics and genomics with ML. 32
detected, the constant flow of data processed by AI models
enables real-time monitoring of clinical conditions, patient 3.3. AI for post-market surveillance of medical
vitals, and device performance, thereby enabling prompt devices
action. AI-powered systems are revolutionizing post-market
AI-based predictive analytics also support medical surveillance by providing reliable methods for monitoring
professionals in complex clinical decision-making, device performance and safety across patient populations
particularly in high-stress situations such as the coronavirus after deployment. One example is the Data Extraction
disease 2019 (COVID-19) pandemic. During the and Longitudinal Trend Analysis (DELTA) network
pandemic, AI applications provided critical information study, which demonstrates how computerized safety
for patient risk assessments, allowing health-care systems surveillance systems can monitor cardiovascular devices.
to efficiently manage patient loads. These systems By continuously analyzing routinely collected data, this
continue to evolve, offering personalized treatment plans system enables the prompt identification of adverse event
and enhancing diagnostic precision through predictive rates associated with specific device classes. It surpasses
analytics. 28 conventional retrospective analysis techniques by
offering real-time analysis, significantly accelerating the
3.2. ML in detecting adverse events identification of safety issues. 33
ML models have shown significant promise in identifying In addition, the potential of AI in real-time health
and reporting safety concerns related to medical devices, monitoring is demonstrated by AI-enabled devices, such as
particularly in the detection of adverse events. These smartwatches with atrial fibrillation detection capabilities,
models are trained on large datasets to detect patterns which have been approved by the United States Food and

Volume 8 Issue 3 (2025) 5 doi: 10.36922/itps.6204