Artificial intelligence in drug development predicted to grow: DIA-Tufts study

By Melissa Fassbender

- Last updated on GMT

(Image: Getty/aurielaki)
(Image: Getty/aurielaki)

Related tags DIA Artificial intelligence AI Drug development

In which areas is artificial intelligence most mature? What are the challenges to widespread adoption? Tufts and DIA recently teamed up to explore these questions.

It’s widely documented that it takes billions of dollars and more than a decade to develop a new drug. Artificial intelligence​ (AI), while not the panacea, has emerged as one enabling technology that could reduce both the cost and time to market.

“The excitement around AI in the life sciences industry continues to grow, and more and more companies are investing in AI and adopt its use in the discovery of tomorrow’s medicines,”​ said Kayla Valdes, PhD, associate director, scientific programs, DIA.

But why now? Valdes cited the increasing size of datasets used in the drug development process for various activities, such as target identification, patient recruitment, or post-market safety surveillance.

“With all this data, comes the need for better data analysis capacity – this is where AI comes in,”​ she told us. “Biotech and pharma are in the midst of adopting AI with the end goal of reducing costs of drug development.”

To better gauge how companies are using AI to reach this goal, the Drug Information Association (DIA), in collaboration with the Tufts Center for the Study of Drug Development (CSDD) recently conducted a study, further results from which will be shared next week during DIA’s 2019 Global Annual Meeting.

DIA brought together eight pharmaceutical and biotech companies – Amgen, Bayer, Eli Lilly, Genentech/Roche, Johnson & Johnson, Merck & Co, Novartis, and Pfizer – to define nomenclature and outline the AI landscape in 2018. The organization then conducted the survey, which garnered responses from 217 organizations, including pharma, biotech, and service providers.

According to the study, AI currently is being used in every major health care function, most often being used in clinical operations functions (61%), followed by pharmacovigilance, safety, and risk management (57%), in addition to information technology (IT) (55%).

Approximately 40% of respondents reported using machine learning and natural language processing (NLP) – subsets of AI – for patient recruitment and selection and analysis of real world evidence (RWE).

“AI implementation in clinical trial decision-making can help researchers gather, analyze, and gain insights from clinical trial data quickly. As one of the most data heavy tasks in drug development, clinical trials create large datasets that capture a vast number of variables of patient data. AI can help generate significant results from this complex data,”​ said Valdes.

As to how AI implementation is managed, 42% of respondents said it is not centrally managed, though 20% indicated that it is managed by R&D, and 12% said it falls under the purview of the chief information officer.

Additionally, the majority (62%) of respondents currently partner with an organization to implement AI. These partnerships are with technology or data providers (95%), academic organizations (58%), and contract research organizations (CROs) (56%).

Partnerships were most often related to clinical operations (71%), pharmacovigilance or safety (68%), and drug discovery and preclinical work (56%).

Not without challenges

The biggest hurdles to AI use were described as a lack of skilled employees in addition to safety, regulatory, and compliance concerns, per the report. Lack of validation also has stood in the way of broad adoption, as many remain skeptical.

Yet, 59% of respondents said they plan to expand AI staff through next year. Such jobs include data scientists, computer scientists, IT specialists, and AI architects. For the time period from 2021 to 2023, 19% indicated AI staff increases were planned, with 7% holding off until after 2023.

According to the report, only 15% said that their organization had no plans to add AI staff.

Mary Jo Lamberti, research assistant professor and associate director of sponsored research at Tufts CSDD, who led the analysis, told us, “one of the greatest challenges besides staff lacking appropriate skill sets was having data that is unstructured or too siloed requiring excessive effort to adapt to AI use.”

“Despite the challenges to AI implementation, however, we expect to see increased utilization given development risk and also the desire for innovative and novel approaches to develop drugs,”​ Lamberti said.

Valdes noted the importance of support from the US Food and Drug Administration (FDA), which has been stressing clinical trial modernization and begun using AI as a regulatory science tool. With regulator backing, “this technology [AI] will indisputably be more openly accepted and implemented across the biopharmaceutical industry at an exponential rate,”​ she added.

Still, adoption is in its infancy. Standards, validation, in addition to regulator acceptance are all key to effective implementation, Valdes said.

“As industry works to improve their AI infrastructure, incorporates and trains the data scientist, and form partnerships with other organizations developing algorithms,” ​she said, “questions about the uncertainty and applicability of AI in drug development will be answered.”

Related news

Show more

Related products

show more

Saama accelerates data review processes

Saama accelerates data review processes

Content provided by Saama | 25-Mar-2024 | Infographic

In this new infographic, learn how Saama accelerates data review processes. Only Saama has AI/ML models trained for life sciences on over 300 million data...

More Data, More Insights, More Progress

More Data, More Insights, More Progress

Content provided by Saama | 04-Mar-2024 | Case Study

The sponsor’s clinical development team needed a flexible solution to quickly visualize patient and site data in a single location

Using Define-XML to build more efficient studies

Using Define-XML to build more efficient studies

Content provided by Formedix | 14-Nov-2023 | White Paper

It is commonly thought that Define-XML is simply a dataset descriptor: a way to document what datasets look like, including the names and labels of datasets...

Related suppliers

Follow us


View more