By: Sas Maheswaran, Vice President of Strategic Consulting, CluePoints
Pharmaceutical companies are increasingly outsourcing to specialist providers in the hope of harnessing artificial intelligence (AI) to optimize clinical trial design, enhance data analysis and streamline processes.
From 2022 to 2024 there was a 30% increase in AI-driven clinical trial partnerships, and, by 2030, AI is expected to be integrated into 60-70% of global clinical trials1. Over the same period, the global AI in clinical trials market is expected to triple to $8.5 billion, with the Asia Pacific market experiencing the highest growth2.
However, there are still challenges which need to be addressed if we are to fully unlock the potential of AI. These include regulatory restrictions, concerns over data quality and security and the cost of implementation.
In this article Sas Maheswaran, Vice President of Strategic Consulting at CluePoints, discusses how AI is being applied to clinical trials, its benefits and the considerations which still need to be addressed to unlock its full potential. They also share real-world examples showcasing how artificial intelligence is being turned into human intelligence.
Using AI in Clinical Trials
The two AI applications with the greatest potential to redefine clinical trials are machine learning (ML) – both supervised and unsupervised – and deep learning (DL). Supervised ML uses labeled datasets to train algorithms to classify data or predict outcomes accurately. Unsupervised ML uses algorithms to analyze and cluster unlabeled datasets without the need for human intervention. DL uses multilayered neural networks for complex decision making and can evaluate and refine outputs for increased precision.
Some of the key areas where these technologies are being used include trial design, patient recruitment and selection, data management and integration, monitoring and adverse event detection. AI is also being used to enhance drug discovery and precision medicine and has huge potential for medical and safety review (MSR).
The Benefits of AI
AI increases efficiency throughout the trial life cycle – from reducing workload in the trial recruitment process by 90%3 to cutting unclear risk signals by 25%4. The FDA has recognized the potential for ML to “significantly enhance” data integration efforts5. ML’s ability to sift through vast datasets and identify patterns makes it ideal for streamlining previously labor-intensive and inefficient manual processes like medical coding and medical safety review. This not only increases efficiency but also unlocks new insights and enhances data-driven decision making.
Ultimately, these efficiencies and new insights result in improved patient safety and better outcomes. AI enables early detection of adverse events and site issues, timely interventions on critical-to-quality issues, and tailored support and treatment plans for patients.
Considerations for AI Use in Clinical Trials
While AI brings several benefits across the clinical trial life cycle, we also need to be aware of the challenges which still stand in the way of implementation – and how to overcome them. Reported barriers to AI adoption include poor data quality, the cost of implementation and regulatory frameworks6 with smaller companies likely to be disproportionately affected7.
AI is only as effective as the data it is trained on. Poor quality data can lead to misleading results, costly trial delays and even compromised patient safety. It is vital to ensure robust datasets are used for model training and validation. Organizations should also consider combining new technologies with approaches which thrive on increasing data sources such as risk-based quality management (RBQM).
To continue to support the development of innovative therapies globally we need to ensure AI solutions are affordable, scalable and compatible with existing systems. The industry also needs to proactively engage with regulators to create more agile frameworks which can adapt to rapidly changing technological and data landscapes.
As well as the technical challenges of AI implementation, we also need to consider how to build trust among both staff and patients. While adoption is on the increase, concerns remain around data privacy and the loss of skilled workers. We need to be clear that AI is a tool which can be used to streamline data analysis, unlock new insights and free up skilled professionals to focus on critical-to-quality issues.
Effective change management, including training and showcasing applications which improve clinical trial efficiency for the benefit of both industry and patients, is vital. Below we share some examples of how AI is being combined with advanced statistics to turn data into positive outcomes in clinical development.
Intelligent Query Detection
Traditionally, issues in clinical data have been identified by using programmatic code – ‘listings’ – which present potential issues to data managers for review. However, Sponsor A reported that only around 30-50% of listing generated issues require a query. The remaining 50-70% are unnecessary noise for data managers to sort through. Manual query detection also lacks a feedback loop to tell the listing whether a query was raised or not. This leaves data managers reviewing hundreds of pages of noise and repeating the same activity trial after trial.
To validate the accuracy of a ML model for intelligent query detection, a head-to-head comparison was run. The ML model was run on a study it had not seen before and had not been trained on. Resulting data issues were then reviewed by expert data managers. This quantitative assessment found the model was up to 95% accurate across 10 of the most common data anomaly issues found in Adverse Event and Concomitant Medication datasets.
In addition, an application allowing data managers to view data issues identified by the model and subsequently raise queries, reduced the average time it took data managers to identify and raise a query from 5-8 minutes to an average of 1.5 minutes.
Medical Coding
Traditional medical coding systems use synonym lists to help match terms from medical records to standardized medical codes. However, traditional coding tools typically successfully code just 50-60% of input terms. While this can be increased with the use of a synonym library, this is labor-intensive to build and maintain and medical coders still spend considerable time assigning codes. Typically, this results in 85% accuracy on initial review and codes are subject to secondary review, in which an approver accepts or rejects decisions.
In contrast, a DL model can integrate with existing systems to offer precise, coding suggestions for adverse events and concomitant medications at up to 99% accuracy and automatically handle regular upgrades of the WHODrug and MedDRA dictionaries, with up to 80% accuracy for completely new terms. Instead of outputting only one dictionary entry, a DL model can suggest several entries to review together with a confidence score.
DL can also leverage the semantics by using embeddings that encode the meaning of words. For example, ‘ache’ and ‘pain’ have very similar vectors which allows the model to understand that their meanings are close to each other. This helps the model properly select the right dictionary entry from many choices and deal with high variability from the input terms in expressing the same concept. Combined, these features reduce the time taken to code medical data and adverse events by around 75%.
Medical and Safety Review
As well as sharing examples of how AI is already transforming global clinical trials, we also need to continue to innovate and explore new opportunities.
Traditional manual data analysis of patient safety outcomes can be prone to inefficiency and error. Common challenges include infrequent data updates, lengthy visualization preparation, slow outlier identification and insufficient data query processes. Technology is already helping to streamline this process, via comprehensive visualization libraries, the use of predefined filters for quick and accurate record identification and enhanced query management.
However, creating an AI-human workflow offers the opportunity to enhance MSR even further by surfacing critical trends and enabling intuitive, on-demand analysis. For example, AI could flag patients breaching Hy’s Law thresholds, streamlining the safety focus. At the same time generative business intelligence (BI) could allow reviewers to request “Show Grade 3 AEs by cycle” and receive tailored visuals which eliminates manual effort and accelerate insights.
The Future of AI in Clinical Trials
AI is already reshaping the clinical trials industry and that transformation looks set to continue over the next five years. New technologies will enable the development of more personalized medicines, empower more responsive trial designs and improve informed decision making by unlocking new data insights.
It is vital we leverage this potential to create smarter, more efficient clinical trials, going beyond data points and algorithms to improve outcomes for patients worldwide.
References
- https://www.clinicalleader.com/doc/global-ai-in-clinical-trials-market-trends-current-partnerships-0001
- https://www.researchandmarkets.com/report/ai-based-clinical-trials
- https://pmc.ncbi.nlm.nih.gov/articles/PMC10636341/
- https://cluepoints.com/natural-language-processing-improves-risk-signal-documentation-in-clinical-trials/
- https://www.fda.gov/media/167973/download
- https://essay.utwente.nl/106899/1/Mesem_BA_BMS.pdf
- https://www.spglobal.com/ratings/en/regulatory/article/241001-ai-in-pharmaceuticals-promises-innovation-speed-and-savings-s13254002
About the Author
Sas Maheswaran has 18 years of clinical development experience at sponsors, sites and software vendors. He is responsible for helping sponsors and CROs operationalize AI transformations across the clinical development spectrum, with particular focus in overcoming a decade of challenges in realizing Risk Based Quality Management (RBQM) as a cross functional obligation.