Over the past few years, pharmaceutical companies have steadily integrated artificial intelligence (AI) into many aspects of clinical development. Today, the impact of AI is felt from the benches to clinics and beyond.
According to a survey by the Tufts Center for The Drug Development (Tufts CSDD), a third of respondents report partial or full implementation of AI, supporting clinical trial planning, design, implementation and regulatory submission. According to the same study, using AI saves 18% on average time savings on clinical trial implementation tasks and activities.
Since 2015, 75 AI-discovered molecules have entered clinics, of which 67 were participating in ongoing trials as of 2023. In 2023, when candidates for the treatment of INIRICO Medicine’s idiopathic pulmonary cytopathy (IPF), INS018_055, emerged, the first drug became clinical disease II.
These examples only damage the surface of AI’s benefits over Pharma. McKinsey has identified 12 use cases that demonstrate the ability of AI to significantly improve the quality, speed and efficiency of clinical development. These use cases showed reduced costs, accelerated registration, and higher success rates as a result of incorporating AI.
Use AI with caution
What Tufts CSDD investigates says there are more successful implementations and use cases. Still, this does not entirely bypass the unique challenges that are hindering AI adoption in pharma.
One of the biggest obstacles is the fear that AI could potentially publish sensitive patient data. Companies are addressing this risk in part by employing anonymization and identification removal, data masking, and pseudonymization to remove personally identifiable information from datasets before they are used in AI applications. Additionally, models trained with identified data are further protected or encrypted to further protect patient-level data. For example, the use of more sensitive data points, such as date of birth, can be avoided by using a proxy instead of age, etc.
Another concern is the different qualities of data used to train AI-driven models. In clinical development, AI models trained with bad data may increase prediction errors and introduce additional delays in trial timelines. Therefore, it is a company that is important to have data born from a reliable source with strong data management practices and receive the required quality assurance and transformation before it is used to train AI models.
Human bias in data occurs when one answer or outcome is more intentionally or unintentionally encouraged than another. If AI models are constructed or trained with biased data, these biases can be perpetuated to their outcomes, exacerbating existing inequality. In one study, when researchers intentionally trained AI assistants using biased data, the accuracy of assistant diagnosis was reduced by 11.3%. IBM says bias recognition must be incorporated into each data processing step, and continuous monitoring and testing using real data may be caught and modified before it is incorporated into the AI model. For example, data points such as ethnicity and race should only be used as data filters to identify eligible participants when the protocol is restricted from an epidemiological perspective. In fact, such data points should not be used as predictors for clinical trial operational indicators that must be based on objective, measurable, and more epidemiologically relevant criteria, such as diagnosis.
Pharma companies need to ensure that compliance tools reach face value so that AI applications can remain within regulations. This is challenging given that several regulatory guidelines, such as the FDA’s “good machine learning practices,” slow the rapid advancement of AI, for example, taking into account the widespread use of generated AI. Using modern and comprehensive compliance tools to evaluate, shape and monitor data and AI models, you can provide AI with rugged regulatory guardrails.
Smooth sky first
Pharma continues to adopt AI in 2025, with some companies likely to adopt use at every stage of the development process. Fairfield Market Research forecasts show that Pharma’s global AI market will reach revenues of over $4.45 billion by the end of 2030, with a robust combined annual growth rate (CAGR) of 19.1% from 2023 to 2030.
This AI deployment could be encouraged by the newly obsessed Trump administration. Shortly after taking office, President Trump helped unveil a $500 billion joint venture between Openai, Oracle and Softbank, which invest in AI infrastructure. During the announcement, Oracle CEO Larry Ellison suggested that some of the projects would link to digital health records and promote the possibility of AI to develop new treatments for diseases such as cancer.
Several ways pharmaceutical companies use AI technologies in 2025 include evaluating epidemiological suitability and their potential for reuse and optimization, planning and optimization of clinical trial design and execution, improving diversity and inclusion of clinical trial registrations, and evaluating the possibility of streamlining and strengthening the regulatory disclosure process.
Several ways Pharma uses AI in clinical trial planning and optimization include predictive and normative modeling, identification of drug candidates, rearrangement of approved drugs for the treatment of other indications, and design and optimization of clinical trial protocols. In particular, protocol design is considered a promising frontier for the use of AI in drug development, given the burden of time and effort required to create a trial protocol, and the delays brought about by protocol modifications in clinical trial timelines.
Merck & Co. Companies like these are already using AI in their development workflows, such as helping and accelerating medical writing, and will leverage AI agents to automate repetitive tasks such as data cleaning and preliminary analysis.
As sponsors implement clinical trial enrollment, the future will see more pharmaceutical companies leveraging AI to improve diversity and inclusion, especially as they may find patients who meet diverse trials, especially by design, and acceptable inclusion/exclusion criteria to accelerate clinical trial timelines.
For example, rather than waiting for a patient to come to the trial, Johnson & Johnson uses AI to find clinical research sites and researchers with eligible and appropriate patients to be helped by the J&J drugs they are studying. J&J also uses data and AI to diversify clinical trials by finding providers that are likely to be treated with a wide variety of patients and prioritized enrollment of eligible patients from those providers. The use of AI in regulatory applications is another area that is likely to see an expansion in 2025. So, save weeks of dialogue. It also investigates the use of AI in automating the production of large volumes of reports and documents required by regulators.
Photo: Zorazhuang, Getty Images
Luca Parisi, Director of Clinical Analytics and Data Science at Citeline, is a recipient of the 2022 Citeline and 2024 Norstella Innovation and the 2023 Norstella Global Hackathon Awards. He leads a team of data scientists developing and delivering industry-leading, sophisticated clinical analytics solutions and services that enhance and optimize clinical trial plans and implementation strategies. He holds a PhD in AI for clinical applications and an MBA with AI expertise in decision intelligence.
This post will be viewed through the Medcity Influencers program. Anyone can reveal MedCity News’ healthcare business and innovation perspectives through MedCity influencers. Click here to find out how.
