There is no doubt that artificial intelligence (AI) is currently en vogue. In 2023, overall AI investments reached $67.2 billion in the U.S. Within the global life sciences market, AI was valued at $1.5 billion in 2022 and is estimated to reach about $9.17 billion in the next 10 years.
Many AI and machine-learning applications are already being deployed in drug discovery and development. Not only can AI process, analyze, and interpret huge datasets, it can also be used for structural drug design and toxicity and efficacy predictions.
However, a persistent issue in drug development is the high cost. For instance, the overall cost of developing a cancer drug and bringing it to market is estimated to be around $2.8 billion. Between 50% and 70% of that cost goes toward the clinical stage of drug development.
Founder and CEO
Altis Labs
“The major driver of the extremely high costs in drug development is the fact that 95% of drugs fail at the clinical stage,” said Felix Baldauf-Lenschen, founder and CEO of the Canadian company Altis Labs. “This means a new approval basically has to pay for all of the previous failures.”
“Some of the highest impact applications of AI in clinical trials are the ones that get to the heart of this challenge. They can aid in identifying the most efficacious drugs earlier in the clinic and get these to market faster and more cheaply while identifying drugs that won’t be efficacious and deprioritize them sooner.”
Using AI to improve trial design
At the clinical stage of drug development, AI has the potential to address many of the current bottlenecks in clinical trials. Even before the start of a trial, AI can be used to predict patient outcomes and the probability of trial success. This allows sponsors to improve their trial design beforehand, increasing the likelihood of successful transitions between trial phases, shortening trial duration, and increasing the chance of regulatory approval.
For instance, the Israeli company QuantHealth has developed a clinical trial simulator based on a dataset of 350 million patients and more than 700,000 therapeutics. It enables drug developers to simulate a clinical trial at scale and predict trial outcomes with 86% accuracy. Learning from these results, drug developers can adjust their clinical trial design to boost their chances of success.
Our AI models then automatically process those scans and generate standard-of-care outcome predictions and over 200 spatial imaging biomarkers,
Using real-world imaging data, Altis Labs has developed AI models that enable imaging-based outcome prediction in clinical trials. Baldauf-Lenschen explained: “Sponsors are processing imaging data from their clinical trials using our models that were trained on millions of de-identified longitudinal images and associated clinical, molecular, treatment, and survival outcomes of patients treated at Canadian health systems over the past decade.”
“Our AI models then automatically process those scans and generate standard-of-care outcome predictions and over 200 spatial imaging biomarkers, which are used by our customers, including AstraZeneca and Bayer, to predict efficacy in the trials they are conducting. Said another way, they can generate synthetic control data to measure treatment effect far beyond the simplistic and oftentimes misleading tumor size measurements that the industry has had to rely on for the past 40 years.”
Better trial matching means greater success
The success of a clinical trial is also impacted by patient enrollment, which remains challenging, work-intensive, and costly. In fact, data analysis by Clinical Trials Arena in 2022 showed that the most common specified reason for trial termination between 2010 and 2021 was low patient recruitment rate.
However, the analysis also showed that the clinical trial termination rate has been steadily decreasing, mainly driven by technological advancements such as AI. For example, patient enrollment can be supported by AI-driven natural language processing tools that are trained on clinical trial protocols and real-world patient data. Suitable patients can then be selected in a process called “trial matching.”
Based on eligibility criteria for a particular trial, the technology can prioritize clinical trials that best suit a patient. In a recent study published in the journal JCO Clinical Cancer Informatics, researchers at the University of Chicago developed a patient-centric clinical trial matching tool that accomplished automated matching “in approximately four seconds, as compared with several hours using manual derivation.”
Matching the right patients with the right trials is essential. Patients who enroll in clinical trials have very likely tried all available treatments without getting the desired effect. Therefore, the last resort is often to try a new experimental therapy. Using AI, patients have a high chance of being enrolled in the right trial and even being treated with relevant experimental therapies.
To this end, the California-based company Unlearn is using AI to increase the success rate of trials with only a small number of participants. It has developed an AI model that uses control data from past trials to build a digital twin for each patient. Unlearn is working with pharmaceutical and biotech companies and academic institutions to improve and use this AI model in clinical trials.
The main advantage of Unlearn’s digital twins is that the control group can be much smaller than in traditional trials, which means that more patients have a chance of receiving the experimental (and potentially beneficial) treatment instead of a placebo.
Improving the patient experience
To ensure that patients who are enrolled in a clinical trial have the best possible experience, the North Carolina-based company IQVIA uses AI to analyze clinical trial data in a patient-centric manner.
Director of AI Engineering
IQVIA
“As AI becomes increasingly integrated into clinical trials, maintaining patient centricity and enhancing the patient experience are paramount,” Wing Lon Ng, director of AI engineering at IQVIA told Inside Precision Medicine.
“It is crucial to engage patients from the outset and throughout the trial. This involves integrating patient feedback into trial planning and design, and ensuring that trials are respectful of and responsive to patient preferences and unmet needs. Early and continuous patient engagement helps build trust and improves adherence, ultimately leading to better trial outcomes.”
For this purpose, IQVIA is training its AI models with diverse patient experience data, collected through literature reviews, interviews, focus groups, clinical outcome assessments, surveys, and digital health technologies.
“We make use of electronic clinical outcome assessments (eCOAs), where data is captured electronically and uploaded immediately, providing near real-time insights into the patient experience. This accelerates access to crucial data, allowing sponsors to make timely decisions that can enhance patient safety and trial efficiency. Furthermore, eCOA platforms send reminders and alerts to patients, encouraging timely data submission and creating a culture of accountability,” Ng explained.
Similarly, the company AiCure has built an AI-based patient engagement platform that clinical trial participants can access through their smartphones. This allows the clinical teams to remotely monitor, support, and analyze trial participants and related data, receive participants’ feedback in real time, and remind them to take their medication on time to prevent people from dropping out of the trial.
Key challenges
The reason for implementing these technologies is to improve on existing approaches, find more accurate ways to measure treatment responses, use continuous measurements to get a richer picture of the effect of a new drug, and allow measurements to be made in ways that are convenient for participants.
Despite the many advantages that AI brings to clinical trials, several challenges remain to be addressed in the future. One of the main challenges is the novelty of AI.
“Many are reluctant to replace traditional approaches with novel technologies in clinical trials,” explained Marsha Samson, PhD, lieutenant commander of the U.S. Public Health Service and senior regulatory analyst at the FDA.
“The reason for implementing these technologies is to improve on existing approaches, find more accurate ways to measure treatment responses, use continuous measurements to get a richer picture of the effect of a new drug, and allow measurements to be made in ways that are convenient for participants. None of these benefits are guaranteed and the challenge is to develop enough data to give us confidence that we will be improving on existing methods.”
Another key problem in the life sciences is data sharing and standardization, which can greatly impact the use of AI in clinical trials.
Lieutenant Commander
U.S. Public Health Service and Senior Regulatory Analyst
FDA
“It takes a lot of high-quality data to train and validate AI models,” Baldauf-Lenschen explained. “There’s a huge bottleneck, particularly across the healthcare industry, wherein data is often of low quality and dispersed in silos. For example, in our case, it can be tremendously challenging to tie together longitudinal imaging data with diagnostic information, patients’ treatment history, and survival outcomes, which is what we need to train and validate AI models that predict survival outcomes.”
Ng said: “AI systems require high-quality, diverse, and large datasets to function effectively. Inconsistent, incomplete, or biased data can lead to unreliable outcomes. Establishing standardized data collection protocols and investing in data cleaning and integration solutions will ensure robust datasets.”
Tying into this is the importance of training AI with data that represents diverse patient populations. “The significant risk of AI and other algorithmic tools is that we perpetuate inequities in our healthcare systems by using biased data to train these computer programs,” said Samson.
It is important to eliminate biases because AI algorithms can inadvertently maintain and amplify existing biases if the data they are training from does not adequately represent the diversity of the population.
“By leveraging patient experience data, sponsors can identify and address the specific needs of various demographic groups, ensuring that trials are inclusive and representative. This helps in designing trials that are better aligned with patient needs and in identifying potential risks and areas for improvement of diversity and inclusion. This not only meets regulatory requirements but also enhances the generalizability and relevance of trial results,” Ng added.
Regulators’ views on the use of AI in clinical trials
Regulatory authorities around the world are addressing many of these challenges and publishing guidelines on the use of AI in clinical trials. By working closely with drug development companies, regulatory bodies can develop clear rules and frameworks for the use of AI in clinical trials. This can ensure that ethical standards are met, data privacy is upheld, and existing laws are complied with.
“The fundamental principles of good clinical practice and evidentiary standards remain consistent for clinical trials, regardless of the use of AI,” said Samson. “Globally, countries are developing guidelines and risk-based frameworks to integrate AI into clinical trials to ensure patient safety and data integrity while fostering innovation. Differences may exist in establishing risks, data privacy regulations, and other factors. Continued collaboration and discussion globally can assist with harmonization, when possible, to support the use of AI in clinical trials.”
Head of the Data Analytics and Methods Task Force
European Medicines Agency (EMA)
“Different jurisdictions have different legal frameworks that may lead to differences in approach,” added Peter Arlett, head of the data analytics and methods task force at the European Medicines Agency (EMA). “Worldwide medicines regulators use fora such as the International Coalition of Medicines Regulatory Authorities and bilateral discussions to keep each other informed and discuss these topics of common interest.”
Looking into the future
While the application of AI in clinical trials is still in its infancy, the growing number of clinical trials around the world means that more and more data will be generated and available in the public domain. As a result, biotech and pharma companies have become increasingly invested in reaping the benefits of AI and machine learning models.
“The use of AI is likely to increase and become a common feature of clinical trials,” said Arlett. “On the one hand, this will be due to the evolving nature of AI and on the other hand, as guidance from regulators is published, the pharmaceutical industry will be able to implement the governance and procedures needed to fully leverage and scale up their use of AI.”
Read more:
- The AI Index Report 2024, Stanford University
- Artificial Intelligence (AI) in Life Sciences Market Size, Share, and Trends,
Precedence Research - Efficiency, effectiveness and productivity in pharmaceutical R&D, Nature Reviews Drug Discovery
- Costs of Drug Development and Research and Development Intensity in the US, 2000-2018, JAMA Network
- Artificial Intelligence Applied to clinical trials: opportunities and challenges,
Springer Nature - Clinical Trial Simulator, QuantHealth
- QuantHealth, Interview With CEO Orr Inbar, FINSMES
- Trial Termination Analysis Unveils a Silver Lining for Patient Recruitment, ClinicalTrials Arena
- Harnessing artificial intelligence to improve clinical trial design,
Nature Communications Medicine - Automated Matching of Patients to Clinical Trials: A Patient-Centric Natural Language Processing Approach for Pediatric Leukemia, JCO Clinical
Cancer Informatics - Clinical Trials Powered by Patients’ Digital Twins, TwinRCTs, Unlearn
- Reflection paper on the use of artificial intelligence in the lifecycle of medicines, European Medicines Agency
Larissa Warneck-Silvestrin is a freelance science journalist based in Berlin, Germany. She has a BSc in biology from Friedrich-Schiller University in Jena, Germany, and an MSc in science communication from the University of Kent in Canterbury, U.K. Larissa has written in English and German for several media outlets, including Labiotech, Deutsche Welle, and Inside Precision Medicine. She specializes in biotechnology, health, medicine, innovation, and biology.
