Laura Gerab da Veiga and Diego Ures1
AI technology has become increasingly known as a tool to maximize company operations, facilitate procedures, analyze data more precisely, among other brilliant features. But its potential in the pharmaceutical industry may go beyond these definitions.
While searching for rare diseases’ cures with the AI’s looming advance, drug development, clinical trials, diagnosis, precision medicine, and healthcare software will face costs and time optimization. This movement calls for a faster and more effective approval process to facilitate treatments and access to drugs. The regulatory optimization’s immediate result is the further growth in R&D and innovative drugs’ market access.
It should come as no surprise that large pharmaceutical companies have invested in acquiring AI or built partnerships for its development. Good examples are the “Roche-Owkin” deal or the “Bayer-Merck” alliance.
Novartis has gone a step ahead and, in 2002, established the NIBR, the Novartis Institute of BioMedical Research. The Institute uses AI in drug trials, discovering new medicines, and analytics, resulting in a 15% reduction in patient participation during clinical trials. More than this, it accelerates the process and gains valuable time. NIBR is fastening the discovery process of new drugs thanks to Big Data and their Machine-Learning algorithms.
The reduced R&D timeframe becomes even more relevant as 90% of clinical drugs do not make it to the trial stage. The ones that do still must be FDA-approved, placing an additional barrier for market access. This thorough process can take up to 12 years for ultimate sanction, entailing massive costs (over US$ 1 billion), which is entirely beyond reach for most R&D projects. New developments are simply left behind.
Simplifying AI regulation for its adoption is not a simple task. AI technology is complex and needs skilled professionals to operate it. Still, there is not yet a sufficient qualified labor force in place. Moreover, most pharmaceutical companies have not gone through the necessary internal adjustments to make the most use of AI tech. The industry must make large investments in infrastructure as most of the R&D data is not digitized.
The private sector has already observed the need for regulatory improvement, and governments seem to be moving in that direction. The conversation focuses on supervised or unsupervised learning, integrating AI with EMR (Electronic Medical Records), ethics knowledge, transparency, data security, and encryption. The US OMB (Office of Management and Budget) recently formulated a guide to government agencies about AI technologies’ governance. While highlighting the importance of cooperation in regulatory convergence, the document also shows uneasiness of over-regulating the tool, which could become a barrier for further developments.
The “Guidance for Regulation of Artificial Intelligence Applications” also addresses non-regulatory approaches. It encourages developing this technology and overcoming barriers, based on a US perspective, to improve safety, fairness, welfare, transparency, while preserving economic and national security.2 The document has 10 main principles for regulatory approaches: public trust in AI, public participation, scientific integrity and information quality, risk assessment and management, benefits and costs, flexibility, fairness and non-discrimination, disclosure and transparency, safety and security, and interagency coordination.
The non-regulatory approaches suggest three pillars: sector-specific policy guidance or frameworks, pilot programs and experiments, as well as voluntary consensus standards.
In a nutshell, agencies should always evaluate privacy risks (data confidentiality), generating trustworthy AI applications. Consequently, society benefits from a higher trust and agency accountability, improving visibility through an informative process. Furthermore, the information used should have high quality and compliance, and companies would take a risk-based approach to implement their AI strategy. Useful data allows for the costs and benefits measurement whereas showing the appropriate regulation level to be implemented. These aspects will only work if there is a flexibility to technology updates, care with problems like unlawful discrimination, and coordination with other agencies. The alignment of approaches and policies allows AI to be more consistent and fair.
There are also two specific regulations applicable to this context, both from primary health authorities, the FDA3 and the EMA4. The requirements contained in said regulations can be summarized in three words: “Fit for use”. That means that a computerized system must be validated to ensure its safe usage for health products, treatments, and, ultimately, patients. The FDA is aligned with the OMB memorandum: they explicitly state that all the regulations should not be a barrier to new technologies’ development. As Brazil adheres to the Pharmaceutical Inspection Convention and Pharmaceutical Inspection Co-operation Scheme (PIC/S), with more than 50 participating countries, these requirements are met through its ANVISA Resolution 301.
It is clear that this segment already has positive results with AI applications. One of the fundamental bases for a clinical trial is that the participants follow the rules. Otherwise, it distresses the research results and can be a deciding factor for the treatment success rate. For instance, to overcome the challenge of the patients’ possible negligence, Bayer uses a PVAI (Pharmacovigilance AI) technology called Genepact to confirm if “the correct person swallowed the right pill”5 by analyzing the contributors’ videos ingesting the tablets.
Another promising segment is the medicines development for rare and complex diseases such as ALS, Alzheimer’s, Parkinson’s, and diabetes. Chemical world leader BASF will use AI to discover and develop treatments for diabetes, with analysis of new and healthier ingredients and foods.
Verge Genomics, focused on developing treatments based on advanced technologies for neurodegenerative diseases, uses machine learning to build genomics intelligence. It uses these advances to keep track of the impact results in a pre-clinical phase. This technology not only saves almost US$ 2.6 billion before the clinical trials investment but also accelerates the process of making the medicine publicly available.
With good prospects of this technology, companies like Healx and Therachon are directing funding from US$ 10 to US$ 60 million for AI to develop drugs for rare and complex conditions. For the 10 most giant pharmaceuticals, the savings on R&D could reach around US$ 70 billion every year.
Another successful example is Cyclica, a company that “combines computational biophysics and machine learning in an integrated platform to develop safer, more effective medicines for patients”6. Their technology is based on a cloud system, using AI to create, investigate, filter, and analyze multiple kinds of drugs. The technology facilitates new medicines’ identification and approval by creating a decentralized partnership and making the market access process more efficient while improving R&D.
The partnership between Bayer and Merk brings another benefit from AI. The technology enables faster identification of chronic thromboembolic pulmonary hypertension (CTEPH). This disease has similar symptoms to asthma and COPD, making the diagnosis hard to identify. The tool recognizes pulmonary vessels, lung perfusion, cardiac check-ups, and the patient’s history, making the diagnostic more efficient, thus enabling a more successful treatment.
Regarding the benefit of finding more reliable patients for clinical trials, AI uses potential contributors’ data and matches them up with answers from the recruitment questionnaire. After analyzing this information, the platform recommends the best R&D fits. Using AI in this context dramatically impacts the costings because a pharma company usually invests from U$ 161 million to U$ 2 billion per medication development.
It is also necessary to “translate” AI solutions to maximize their use through software that makes the data interpretation more accessible. Two examples of this technology are IBM’s “Watson” and Apple’s “Researchkit”. They can put together relevant information and synthesize them, pointing to the collected data’s direct results, creating transparency, and providing a machine learning that has the “ability to explain or to present in understandable terms to a human”7.
In conclusion, AI technology makes the pharma R&D process a lot more efficient and faster, which entails quicker access for those in need, especially in the developing world. The tool brings real-world data into the analysis, showing and interpreting the impacts, results, and solutions, proportioning a proven treatment/drug real value. And with this kind of information provides market access for very promising cures.
A final key aspect that interferes directly in the market access is the KOLs. Key opinion leaders are mostly represented by physicians, pharmacists, and healthcare providers. They impact and participate in the research presentations, congresses, and clinical trials, having a lot of power to accelerate – or not – the drug approval process. Aiming to improve this relationship, AI also has a solution. The technology collects data from those professionals, mapping, without any bias, obtaining a scalable list, updating real-time, maintaining, and investing in the relationship with the leading opinion influencers.
1 Diego is Sidera’s Partner for Investment Attraction and Market Access, and Laura is Undergraduate Business Administration student at Insper, and an intern at Sidera’s Global Internship Program.
2 Adapted from Memorandum for the Heads of Executive Departments and Agencies.
3 U.S. Food and Drug Administration.
4 The European Agency for Medicines.