Jul 18 2019
In a review published on July 17th, 2019 in the journal Trends in Pharmacological Sciences, scientists analyzed how artificial intelligence (AI) could influence drug development in the next 10 years.
Big pharma and other drug developers are struggling with a problem: the age of blockbuster drugs is ceasing. At the same time, incorporating new drugs to their portfolios is slow and costly. It requires on average 10-15 years and $1.5-2B to commercialize a new drug; about half of this time and investment is spent on clinical trials.
Although AI has still not had a major impact on clinical trials, AI-based models are helping trial design, AI-based methods are being used for patient recruitment, and AI-based monitoring platforms aim to increase study adherence and curtail dropout rates.
AI is not a magic bullet and is very much a work in progress, yet it holds much promise for the future of healthcare and drug development.
Stefan Harrer, Computer Scientist and Researcher, IBM Research-Australia
He is also the lead author of the study. As part of the review and based on their study, Harrer and coworkers reported that AI can possibly increase the success rate of clinical trials by:
- Efficiently computing biomarkers that echo the efficacy of the drug being tested
- Recognizing and characterizing patient subpopulations ideally suited for particular drugs. Under a third of all phase II compounds advance to phase III, and one-in-three phase III trials fail-not because the drug is unsuccessful or unsafe, but because the trial lacks a sufficient number of patients or the right kinds of patients.
- Start-ups, regulatory bodies, large corporations, and governments are all investigating and driving the use of AI for enhancing clinical trial design, Harrer says.
What we see at this point are predominantly early-stage, proof-of-concept, and feasibility pilot studies demonstrating the high potential of numerous AI techniques for improving the performance of clinical trials.
Stefan Harrer, Computer Scientist and Researcher, IBM Research-Australia
The authors also identify more than a few areas showing the most practical promise of AI for patients. For example:
- AI-enabled systems might enable patients more access to and control over their private data.
- Coaching via AI-based apps could happen before and during trials.
- AI could track individual patients' adherence to protocols uninterruptedly in real time.
- AI methods could help direct patients to trials of which they may not have been conscious
- Above all, Harrer says, the use of AI in precision-medicine methods, such as applying technology to advance how accurately and efficiently specialists can diagnose, treat, and manage neurological diseases, is encouraging. "AI can have a profound impact on improving patient monitoring before and during neurological trials," he says.
The review also appraised the probable implications for pharma, which included:
- Computer vision algorithms that could potentially determine appropriate patient populations through a variety of inputs from handwritten forms to digital medical imagery.
- Applications of AI analysis to failed clinical trial data to expose insights for future trial design.
- The use of AI competencies such as Deep Learning (DL), Machine Learning (ML), and Natural Language Processing (NLP) for correlating diverse and large data sets such as medical literature, electronic health records, and trial databases to help pharma optimize trial design, recruiting, and patient-trial matching as well as for checking patients during trials.
The authors also identified numerous crucial takeaways for scientists:
- "Health AI" is an emergent field connecting pharma, medicine, data science and engineering.
- The next generation of health-related AI experts will need a wide array of knowledge in algorithm coding, analytics, and technology integration.
- Continuing work is necessary to evaluate data privacy, security and accessibility, as well as the ethics of using AI methods to sensitive medical data.
Since AI approaches have just started to be applied to clinical trials in the past 5 to 8 years, it will most probably be another several years in a standard 10- to 15-year drug-development cycle before AI's influence can be accurately measured.
In the interim, rigorous research and development is needed to ensure the feasibility of these advances, Harrer says.
Major further work is necessary before the AI demonstrated in pilot studies can be integrated in clinical trial design. Any breach of research protocol or premature setting of unreasonable expectations may lead to an undermining of trust—and ultimately the success—of AI in the clinical sector.
Stefan Harrer, Computer Scientist and Researcher, IBM Research-Australia