AI Poised To Revolutionize Drug Development

There has been much discussion recently about how artificial intelligence and machine learning (AI/ML) will revolutionize pharmaceutical research. Substantial progress has been made in the discovery and identification of new drugs enabled by AL/ML. Now the clinical testing process is being revitalized by advances in technology. The US Food and Drug Administration (FDA) is paving regulatory groundwork with a new discussion paper regarding AI/ML in drug development.

Rapid advances in AI-guided automation will revolutionize how scientists discover new medicines in the laboratory according to the McKinsey Global Institute. Scholars are tracking how AI/ML use has been increasing in the pharmaceutical industry, including drug discovery, drug repurposing and improving pharmaceutical productivity.

The next frontier is drug development, including driving innovation in clinical trials. AI will be used to improve clinical trial design, management, and outcomes, allowing for more efficient use of resources while also providing more accurate results.

“There is a lot that will be transformative about AI and ML in drug development,” says Fareed Melham, SVP and Head of Medidata AI, an early leader applying advanced technology to the drug development industry. “This is a new way to understand data and gives us a better ability to search and unlock knowledge from legacy materials.”

AI Tools For Data Deluge

The scale and speed of AI for processing and analyzing vast quantities of information is far superior to traditional systems which have become overburdened by escalating growth of medical knowledge and clinical trial data. A single trial can generate an average of 3.6 million data points, roughly three times the volume of data collected by late-stage studies a decade ago.

We have been witnessing a drastic increase in data digitization in the medical sector, and the ability to intelligently leverage data insights has been called the pharmaceutical industry’s next blockbuster. One report estimates that the industry could spend more than $4.5 billion on digital transformation by 2030. This investment is partial due to the rapid increase in medical data. It has been documented that medical knowledge has been expanding exponentially. Whereas the doubling time was an estimated 50 years back in 1950, it accelerated to 7 years in 1980, 3.5 years in 2010, and now medical data is expected to double up to 5 times per year or about every 70 days.

The data surge to increasing protocol complexity with industry’s focus on rare diseases as well as greater use of biomarkers and patient stratification being major drivers or increased data requirements. Phase 3 clinical trials generate an average of 3.6 million data points, roughly three times the volume of data collected by late-stage studies a decade ago. Most protocols involve an average of 263 procedures per patient, supporting approximately 20 endpoints, which is an increase of 44% since 2009. In addition, the average number of investigative sites conducting phase 2 and 3 protocols increased 33%.

Experts have said that the rise in data complexity contributes to high failure rates. Complex development processes and difficult disease mechanisms are leading to larger amounts of data and more patients in clinical trials. Success rates are frighteningly low. The likelihood of a new drug advancing to the next stage or regulatory approval is just under 14% for all therapeutic areas. Phase III trials have success rates from a low of nearly 36% for oncology to a high of over 85% for vaccines.

In order for healthcare researchers to keep up with the overwhelming growth in data, new tools such as AI have to be more widely used. AI is a technology-based system involving various advanced software tools and networks that can mimic certain human functions. One of its key attributes is the ability to handle large volumes of data with enhanced automation. AI uses several method domains, such as reasoning, knowledge representation, solution search. ML uses algorithms that can recognize patterns within a set of data that has been further classified. A subfield of the ML is deep learning (DL), which engages artificial neural networks (ANNs). The popular new applications in generative AI, such as ChatGPT, are based on algorithms that can be used to create new content, including audio, code, images, text, simulations, and videos, as described by McKinsey.

AI In The Lab

AI/ML has the potential to revolutionize drug discovery and accelerate the development of new drugs. It can make drug development cheaper and faster while improving the probability of approval. Involvement of AI in the development of a pharmaceutical product has been used to aid rational drug design AI/ML is being used in different parts of drug discovery such as drug design, synthesis, screening, polypharmacology, and repurposing. It has been proposed to reduce the time and capital required to take a drug from the laboratory to clinical trials. AI is expected to deliver value in small-molecule drug discovery by accessing new biology, improving or novel chemistry, increasing success rates, and speeding up discovery processes. AI can also help in preclinical development by testing potential drug targets on animal models and predicting how a drug might interact with them. Deep learning algorithms can analyze the structure of molecules and predict their properties such as their solubility, bio-availability, and toxicity. The prediction of the interaction of a drug with a receptor or protein is essential to understand its efficacy and effectiveness, allows the repurposing of drugs, and prevents polypharmacology. By analyzing large amounts of data, predicting drug efficacy, and optimizing drug design, AI can help researchers identify new drug targets, design new drugs, and optimize the drug discovery process.

AI In Biomarkers

Although AI/ML is getting more attention recently with the new FDA discussion paper and the broader zeitgeist around more novel AI technologies, many of these biopharmaceutical companies have been investing in their data science teams and the evidence to validate their ML models for AL-powered digital biomarkers for the better part of a decade. The industry is poised to have broader adoption with the increased regulatory support.

“Developers of predictive models for digital biomarkers have been using AI/ML extensively for the better part of a decade,” said Andy Coravos, the CEO and co-founder of HumanFirst. “More than 95% of sensor-derived digital measures have used some type of machine learning to develop the technology whether as part of the signal processing or to validate the models.” According to Humanfirst’s Atlas platform, which categorizes more than 2000 DHTs deployed across 1700 clinical trials, more than 100+ sponsors, including top biopharma companies like Janssen (Johnson & Johnson), Roche, and Regeneron have used AI-powered digital biomarkers and endpoints across a range of Phase 1 through 4 clinical trials.

Digital endpoints drive not only scientific progress but also R&D efficiencies. At the J.P. Morgan Health Care Conference this past year, Roche CFO Alan Hippe gave a presentation about how the organization using novel digital endpoints for prasinezumab Ph II (PASADENA) where the trial duration with the digital measure was two times faster and the sample size was 70% leaner, resulting in 2x return ratio.

AI In The Clinic

AI is being proposed to improve clinical trials and drug development in multiple ways. According to a life sciences digital innovation survey, 76% of respondents are currently investing in AI for clinical development. AI-enabled data collection and management can reduce the time and effort required for clinical trials, accelerate the drug development process, and help companies get new treatments to market more quickly.

“AI/ML will initially be a great help in the automation of many tasks,” says Melham from Medidata AI. “It will be supervised, but the technology will help reduce much of the human time that is now spent on creating analytics or charts, as well as documentation of the large amount of data collected in clinical trials.”

AI technologies can be used to create structured, standardized, and digital data elements from a range of inputs and sources. These tools can interpret vast data elements, feed downstream operating systems, and help populate required reports and analyses.

• Protocol design. AI-enabled study design could help optimize and accelerate the writing of appropriate targeted clinical trial protocols. It would promise to decrease the number of amendments, increase the likelihood of success, and improve overall efficiencies, in addition to reducing patient burden. AI algorithms can analyze historical clinical trial data to identify potential areas for protocol optimization, such as selecting appropriate endpoints, sample sizes, and study durations. By leveraging AI’s ability to analyze complex data, researchers can design more efficient and informative trials.
• Data collection. AI could help develop innovative ways of collecting trial data and reducing the need for patients to visit hospital sites. Body sensors and wearable devices such as bracelets, heart monitors, patches, and sensor-enabled clothing, can monitor vital signs and other information from patients’ homes. Advanced algorithms would help reveal real-time insights into study execution and patient adherence.
• Patient screening. AI powered algorithms could facilitate identifying which patients would be appropriate for clinical trials based on their specific personal medical characteristics and aligned to trials’ enrollment criteria. In this way, patients with subtle combinations of symptoms might be identified and diagnosed early and provided options for clinical trials. AI/ML can be used to mine vast amounts of data, such as data from clinical trial databases, trial announcements, social media, medical literature, registries, and structured and unstructured data in EHRs, which can be used to match individuals to trials. AI/ML has been explored and used as part of a clinical investigation in the prediction of an individual participant’s clinical outcome based on baseline characteristics, and predictive models can be used to enrich clinical trials and may be possible to reduce variability and increase study power. Such models can also be used for participant stratification which could lead to predict the probability of a serious adverse events.
• Dosing. AI/ML can be used to characterize and predict pharmacokinetics (PK) profiles after drug administration. It can also be used to study the relationship between drug exposure and response.
• Real-time monitoring and safety. AI-powered systems can enable continuous monitoring of patients during clinical trials, providing real-time insights into their health status and potential adverse reactions. This can help ensure participant safety and allow for timely intervention when needed.
• Adherence and retention. AI/ML can be used to monitor and improve adherence during a clinical trial through remote tools, such as smartphone alerts and reminders. It also has the potential to improve retention by increasing participants’ access to relevant trial information by enabling tools, such as AI chatbots, voice assistance, and intelligent search.
• Documentation. One of the significant work loads of clinical development is the need to summarize large amounts of information, and then analyze for regulatory submissions. For example a new drug application (NDA) can have over 100,000 pages. AI automation can greatly facilitate and accelerate documentation, analysis and submission.
• Data management. As pharmaceutical companies produce hundreds of thousands of pages of reports and documentation for regulators, AI can help automate the production of much of that information, and must be share across the sponsor company and external partners such as clinical research organizations, clinical trial sites, academic partners and investigators, diverse and complex real world data (RWD) extracted from electronic medical records (EHRs), medical claims, and disease registries. It would be used for a range of data cleaning and curation purposes, including duplicate participant detection and imputation of missing data values.
• Digital twins and external control arms. AI/ML can also be used in the context of creating digital twins,which are essential in silico duplicates of relevant patient characteristics. This is an emerging method that can be utilized to build replicas of an individual that can dynamically reflect molecular and physiological changes and potentially predict pharmacological effects including safety events. Such a concept would be beneficial in external control arms (ECAs) where aggregated digital twins could provide a comprehensive, longitudinal, and computationally generated clinical record to understand what would have happened in a traditional placebo control arm of a clinical trial.

Regulatory AI Leadership

The FDA released a discussion paper covering AI/ML in drug development in June 2023, Using Artificial Intelligence and Machine Learning in the Development of Drug and Biological Products. The FDA’s stated goal is an initial communication with stakeholders, including academic groups, researchers, and technology developers, that is intended to promote mutual learning and discussion. The agency provides some much needed definitions including that machine learning (ML) is considered a subset of AI that allows “ML models to be developed by ML training algorithms through analysis of data, without models being explicitly programmed.”

The discussion paper outlines a wide array of how AI/ML is being implemented across drug discovery for target identification and compound screening and also in clinical research for recruitment, selection of trial participants, dose/socing regime optimization, adherence, retention, site selections, trial data collection and clinical endpoint assessments. Two other sections look at AI/ML applications for post market safety surveillance and advanced pharmaceutical manufacturing. One of the key sections of the FDA discussion paper is references for how AI/ML are being applied to real-world data (RWD) and data from digital health technologies (DHTs) in support of drug development.

The FDA states that, “AI/ML is being utilized to analyze vast amounts of data from both interventional studies (also referred to as clinical trials) and non-interventional studies (also referred to as observational studies) to make inferences regarding the safety and effectiveness of a drug. Additionally, AI/ML has the potential to inform the design and efficiency of non-traditional trials such as decentralized clinical trials, and trials incorporating the use of RWD extracted from EHRs, medical claims, or other data sources.”

Clinical Research Change Long Overdue

Traditional development for clinical trials is an unsustainably expensive process. As a rule, the cost of phase two clinical trials can be anywhere between $7 million and $20 million, while the average cost of phase three can surpass as much as $50 to $100 million. According to academics, the overall price tag to get one drug all the way to approval is well over a billion dollars.

Longer timelines contribute to the cost of developing new drugs. It often takes 10 to 12 years to bring a potential medication through research and development (R&D) to market. The clinical-trial phase averages five to seven years often due to the traditional flow of data through clinical operations systems, which can be a complicated maze of manual effort, rework, and inefficiency.

The first randomized control trial of streptomycin in pulmonary tuberculosis was carried out in 1946. Life sciences executives sometimes lament that the same processes that used over 50 years ago are still the standard. Now AI has the potential to significantly improve clinical trials in several ways. It promises to assist in various stages of the drug discovery and development process. It can help identify potential drug targets, simulate drug interactions, and optimize lead compounds. Additionally, AI can analyze scientific literature to support researchers in assessing the latest findings, potentially accelerating the development of new treatments. In drug development, AI can process and analyze vast amounts of clinical data, including patient records, genetic information and imaging data. By utilizing machine learning techniques, AI algorithms can identify patterns, detect correlations, and make predictions that could aid in identifying potential treatment outcomes, adverse events, or drug interactions.