Artificial Intelligence Makes FDA’s Current Drug Approval Process Obsolete – Healthcare Blog
Steven Zecora
Artificial Intelligence (“AI”) has taken root in the field of drug discovery and development and is showing signs of transcending traditional research models. Congress should take note of these rapid changes and: 1) direct the Department of Health and Human Services (“HHS”) to phase out the government’s basic research grant program for non-AI applicants, 2) require HHS to redirect these funds to fund emerging AI applications, and 3) require HHS to revise the drug approval roadmap for AI-driven trials to reflect new capabilities in drug discovery and development.
background
The U.S. healthcare industry has four distinctive characteristics.
First, the industry's costs as a share of gross national product have risen from 8% in 1980 to 17% today, and are expected to exceed 20% by 2030. The federal government subsidizes about one-third of the cost. As health care costs continue to soar, especially with the federal deficit totaling $37 trillion, these subsidies are unsustainable.
Secondly, the industry's regulatory system is such that the approval of each drug requires an average of 18 years of basic research and 12 years of clinical research. The clinical cost of each newly approved drug currently exceeds $2 billion. The economics of drug discovery are so unattractive to investors that the federal government and philanthropic foundations fund nearly all basic research. The federal government allocates $44 billion annually to this endeavour. When this cost is spread over the 50 or so drug approvals per year, the cost per drug increases by about $880 million, bringing the total cost per drug approval to more than $3 billion. Worse, the process is getting slower and more expensive every year. Therefore, drug discovery under current research methods will not make a significant contribution to reducing overall healthcare costs.
Third, the Trump administration laid off thousands of employees at the Department of Health and Human Services, weakening the federal government’s role in health care. As a result, the agency can no longer effectively administer the rules and regulations it previously adopted and therefore cannot expect to guide drug discovery to reduce health care costs.
Fourth, on the positive side, AI software combined with the vast and growing computing power of supercomputers shows the potential to significantly reduce the cost of drug discovery and radically shorten the time to identify effective treatments.
Bringing artificial intelligence (AI) to drug discovery
Over the past decade, a handful of companies have been exploring advanced automation technologies to improve many aspects of the drug discovery process. Improvements can now be made in meeting regulatory documentation requirements, which currently add up to 30% of the cost of compliance. What’s more, AI can be used to accurately create comprehensive clinical documentation by referencing and cross-referencing raw data, and continuously updating and validating the documentation.
Top AI drug discovery companies include Insilico Medicine, Atomwise, and Recursion, which use AI to accelerate every stage of drug development, from target identification to clinical trials. Other notable companies include BenevolentAI, Insitro, Owkin and Schrödinger, as well as technology providers such as Nvidia that provide critical AI infrastructure for the life sciences sector.
For example, recursive use of biological experiments combined with machine learning can identify potential treatments faster than traditional methods. Additionally, it creates a platform of data and tools for biopharmaceutical and commercial users to use in drug discovery and development.
While various approaches are being explored, the real promise of AI in drug discovery lies in knowledge creation. By efficiently exploring biological variability, AI can significantly increase the number of experiments by studying trillions of interactions between variables. This ability will be particularly helpful in treating complex and expensive diseases (such as Alzheimer's disease, Parkinson's disease, autism) and people with multiple chronic conditions. In other words, AI can process large amounts of biological data, reveal hidden causal relationships, and generate new actionable insights. Governments should focus on and encourage these capabilities because they have the potential to improve the health of the nation's most vulnerable citizens and significantly reduce the cost of providing care.
Healthcare regulation must adapt to the age of artificial intelligence
The rapidly growing potential of artificial intelligence in drug discovery requires new regulatory models. The federal government’s goal should not be to apply current regulatory processes to new AI-driven research, but rather to develop a regulatory process that accelerates multivariable treatment combinations that effectively reduce costs.
For example, given that AI can be used to continuously update and validate documentation, all clinical work utilizing AI should be broken down into one extended trial rather than discrete Phase 1, 2, and 3 trials. Safety results can be reviewed and reported in real time as participants are added to the trial. Once a trial exceeds a certain number, such as 1,000 participants, and proves its effectiveness and meets specified safety protocols, it will be approved for rollout. In this approach, the government's role is to act as an auditor to verify the results of the trial. This function will include experimental validation, mechanistic understanding, and ethical oversight.
generalize
For years, the health care industry has failed the American people with high costs and poor performance. Existing drug discovery processes offer relatively little improvement to this equation.
On the other hand, emerging AI discovery and development models will reach the market years earlier than traditional basic research projects and at a fraction of the cost. To realize the full potential of new technologies, a new industry model is needed. That is, subsidies for basic research and regulation of clinical trials that harness AI for discovery must change.
Any basic research projects currently under review are at a significant disadvantage compared to AI-driven research projects and should not receive funding. Instead, the focus of government funding should be on AI-driven research, particularly for people with Alzheimer’s, Parkinson’s, autism, and multiple chronic conditions. These categories account for the majority of U.S. health care costs and are least likely to be cured through traditional research methods.
Additionally, regulators can leverage AI’s recording and continuous updating capabilities to integrate clinical trials into a continuous phase, where regulatory approval can be obtained when preset safety and efficacy conditions are met after a specified number of participants enter the trial.
Twenty-three years ago, Steve Zecola sold his web application and hosting business when he was diagnosed with Parkinson's disease. Since then, he has worked in consulting, taught at graduate business schools, and practiced extensively
