AI Use in Pharma Shows Promise, Prompts Caution

It’s hard to underestimate the reach of artificial intelligence (AI) across the health care and pharmaceutical industries. While the ultimate impact of the technology on payers and providers may be debatable, pharma companies have made broader inroads into exploring ways to enhance their efforts, including in drug discovery. Still, caution in some areas is warranted, according to some studies.

For the Managed Care Biologics & Injectables Index: Q3 2023, from Aug. 13, 2023, to Sept. 29, 2023, Zitter Insights polled 35 commercial payers covering 117.7 million lives, 103 physicians and 83 practice managers about their familiarity with Open AI’s ChatGPT and AI tools in general.

Among the respondents, 33 payers with 115.9 million commercial lives, 77 physicians and 59 practice managers said that they are familiar with ChatGPT/AI. Those respondents ranked data retrieval and prior authorization as the top potentially useful areas for them, while data privacy and data breaches were flagged as the areas of most concern (see chart).

Both AIS Health and Zitter Insights are divisions of MMIT.

Among the three stakeholder groups, payers reported having the most familiarity with ChatGPT, with physicians having the least amount. And while almost two-thirds of commercial payers said they are having preliminary discussions about implementing ChatGPT or similar AI tools, less than one-quarter said they are actively researching and implementing them.

AI Is ‘Game Changer in Pharma’

Pharma companies, however, may be further along in their use of and comfortability with AI.

For many reasons, AI “is a game changer in pharma,” asserts Dinesh Kabaleeswaran, senior vice president of consulting and advisory services for MMIT. “In a recent survey conducted by MMIT, more than two-thirds of commercial market access personas have indicated that their organizations are leveraging AI for several use cases. From a commercial standpoint, [AstraZeneca’s] initiative to identify diseases risks through unstructured patient notes by employing AI and NLP [natural language processing] could pave the way for greater innovation and better survival rate outcomes. One of the more common developments that we continue to hear more about is integrating AI with patient communication and disease management strategies.”

Indeed, multiple respondents to the Zitter Insights survey cited disease and drug management as areas in which they are actively researching and/or implementing AI.

Even before the commercialization phase, AI is having an impact in earlier stages of pharma development. Over the past year, “we observed significant traction in the role of AI in advancing drug discovery and development,” says Namrita Negi, head of the Life Sciences Knowledge Center at Deloitte Consulting LLP. “Big Tech companies made significant inroads in this space through investing, collaborating and introducing services for generative AI-based drug discovery, advancing their AI-driven molecular modeling capabilities, hence promising breakthroughs in understanding disease pathways, drug design and genomics.”

Negi adds that “another key milestone for the industry in 2023 was when the first drug discovered and designed with generative AI entered Phase II trials”: Insilico Medicine’s INS018_055, which is being assessed in idiopathic pulmonary fibrosis.

AI has been a potent tool in drug discovery, declares Rumiana Tenchov, Ph.D., D.Sc., an information scientist at CAS, a division of the American Chemical Society. “For example, AI can predict structure-function relationships for small-molecule drugs, identify targets and screen candidates by conducting molecular dynamics simulations. Similarly, it can predict protein structure and function to identify new therapeutic candidates.”

This expertise, she says, can mean “more therapies at lower costs. Pharmaceutical and biotechnology companies make large investments in developing AI capabilities, and companies like Alphabet and Nvidia have expanded into drug research. Traditional drug discovery is a notoriously time-consuming and expensive process, but AI tools are revolutionizing virtually every step of the drug discovery process, offering substantial potential to reshape the pace and finances of the industry.”

When it comes to identifying potential targets, AI can be trained to use large datasets in order to “understand the biological mechanisms of diseases and to identify novel proteins and/or genes that can be targeted to counteract those diseases,” Tenchov tells AIS Health. “Combined with systems like AlphaFold, AI can proceed further by predicting the 3D structures of targets and speed up the design of appropriate drugs that bind to them.”

Researchers can also forgo traditional chemistry methods — and their costs — to physically test candidate drug compounds by instead using “high-fidelity molecular simulations that can be run entirely in silico,” she notes. In addition, certain systems can forecast important properties “such as toxicity, bioactivity and the physicochemical characteristics of molecules,” bypassing simulated testing of candidates.

“While traditional drug discovery has historically involved the screening of large libraries of candidate molecules, AI is shifting this paradigm,” declares Tenchov. “Some systems are capable of generating promising and novel drug molecules entirely afresh.”

Studies Identify Limitations

Still, other areas have seen mixed results with AI, according to two recent articles in JAMA Oncology.

The first assessed large language model (LLM) chatbots’ use in providing treatment recommendations for breast, prostate and lung cancer adhering to National Comprehensive Cancer Network (NCCN) guidelines. Researchers found that all of the outputs that had a recommendation had at least one NCCN-adherent treatment, but more than one-third of the outputs also recommended at least one treatment that did not adhere to NCCN guidelines.

Study authors concluded that “clinicians should advise patients that LLM chatbots are not a reliable source of treatment information. Language learning models can pass the US Medical Licensing Examination, encode clinical knowledge, and provide diagnoses better than laypeople. However, the chatbot did not perform well at providing accurate cancer treatment recommendations.”

For the second study, researchers examined the information provided by four AI chatbots about the top five search queries for skin, lung, breast, colorectal and prostate cancers. While researchers assessed the responses’ quality as good and did not detect misinformation, they found that understandability of the responses was moderate, and actionability was poor.

Researchers concluded that the chatbots “generally produce accurate information for the top cancer-related search queries, but the responses are not readily actionable and are written at a college reading level. These limitations suggest that AI chatbots should be used supplementarily and not as a primary source for medical information.”

For more information on the Zitter Insights data, contact Jill Brown Kettler at jkettler@mmitnetwork.com. Contact Kabaleeswaran at dkabaleeswaran@mmitnetwork.com, Negi via Julie Landmesser at jlandmesser@deloitte.com and Tenchov via Zornitsa Ivanova at ZIvanova@cas.org.

0

0

0 Comments

0

0

• Spotlight on Market Access