Artificial Intelligence in Healthcare: A New Era of Innovation and Ethical Considerations

Authors:

Mustaqhusain Kazi, Global Head of Roche Informatics Strategy and Digital Innovation; Chairperson of the Board at Alliance for AI in Healthcare (AAIH)

Stacie Calad-Thomson, Business Development Leader, NVIDIA Healthcare and Life Sciences; Vice-Chairperson of the Board at Alliance for AI in Healthcare (AAIH)

Abstract

AI has demonstrated its potential to substantially impact healthcare systems across the value chain and will continue to transform drug research & development, drive equitable healthcare, and deliver better outcomes for patients.

This opinion piece from AAIH explores relevant use cases and what regulatory bodies should consider when evaluating AI technologies being used in healthcare. We will share what AAIH is doing to influence policy, engage stakeholders, educate, and learn together as we drive AI adoption to benefit the industry, patient well-being, and global humanity.

Introduction

Just as the industrial revolution reshaped the future of entire industries, Artificial Intelligence (AI) stands out as an equally transformative force in healthcare and life sciences. While the full potential of AI in this field remains to be seen, its applications are already revolutionizing several key areas. These include: 1) accelerating drug research and discovery for better medicines designed faster, 2) clinical trial optimization to promote more equitable healthcare, and 3) enhancing patient outcomes through clinical insights and care delivery, something often referred to as “personalized medicine”. But what are the real-world examples of AI's impact, and how are regulatory bodies responding to this technological evolution?

AI-Based Drug Discovery

The traditional drug discovery process is slow, expensive, and inefficient, with over 95 percent of programs failing to reach the clinic and those that do have less than a 10 percent success rate in becoming an approved drug (1). A recent analysis of pharma R&D productivity over the last 20 years shows that an average of $6.16 billion is spent per drug approval (2). AI is transforming this landscape and reducing costs by accelerating scientific productivity through better identification of potential drug targets and candidates with a higher probability of success.

Many solutions are providing significant advancements in AI-based target discovery and drug design. With large clinical-genomic datasets and automated experimental labs capable of creating massive and high-quality data, AI identifies new connections underpinning disease biology and proposes new, more effective drug targets, opening new avenues for therapeutic innovation and personalized medicine. Additionally, AI models can sift through billions of compounds and optimize structures for multiple parameters in parallel, like efficacy, safety, and pharmacology, significantly reducing the time and cost required for preclinical research. A recent example of this is NVIDIA’s generative virtual screening blueprint featuring accelerated NIM microservices that can design optimized small molecules smarter and faster (3).

Many AI drug discovery companies (Recursion, insitro, Insilico Medicine, and others) have invested in automated, robotic laboratory infrastructure used to run massive experiments, and all of their data is driving the creation of biology maps in ways similar to digital city maps. This newfound efficiency in understanding human biology expedites the translation of scientific breakthroughs into life-saving treatments, where AI solutions to design drugs also provide added efficiency. When you combine this speed to experimental discovery and validation with open source AI tools like Google DeepMind’s AlphaFold (4) or Meta AI’s ESMFold (5) AI protein structure prediction technologies, new drugs are being designed in months rather than the historical 2-6 years that this phase traditionally covers (6).

Clinical Trial Optimization

AI is driving a more selective, efficient, and decentralized approach to clinical trials that should increase the success of those trials. The ability to analyze large amounts of data, for instance with Large Language Models (LLMs) {7}, offers the opportunity to identify suitable candidates for trials, speeding up the recruitment process and ensuring a more diverse and representative sample. Large Language Models are also allowing us to quickly review vast amounts of scientific literature to confirm the protocol is based on the latest research findings and best practices. Using digital biomarkers as primary and secondary endpoints also offers us a more nuanced understanding of patient responses, while AI-driven patient identification and recruitment help match the right participants for each trial.

Finally, the concept of digital patient twins is also becoming a reality with the advent of software as medical devices (that collect data needed to simulate a patient), digital therapeutics, remote patient monitoring, adherence, and retention. Together, these technologies speed up the trial process and enhance the results' quality. For example, the company Unlearn has developed deep learning-based disease progression models, trained on historical data, which they use to create digital twins that are accepted by both the EMA and FDA (7,8,9). These digital twins can predict how their corporeal counterparts are likely to respond to various treatments to achieve the best treatment outcome in clinical trials and bring new treatments, therapies, and cures to market faster.

Clinical Insights, Outcomes & Real-World Evidence

For patients, the integration of real-world data with new data modalities like genomics, clinical imaging, and digital insights from wearables will provide a more comprehensive 360-degree view of health, enabling more targeted interventions and a better understanding of disease progression and treatment efficacy. Doctors and clinicians now have access to large language models (LLMs) like OpenAI GPT4 (10), Google’s Med-PaLM (11), and Writer’s Palmyra Med (12), (available as a NIM microservice from NVIDIA{13}), the first AI systems to pass the US Medical Exam. Doctors are also assisted by FDA-approved radiology diagnostics that can more accurately interpret scans than experienced pathologists, providing patients with better care options and outcomes (14). With the FDA approval of hundreds of AI/ML-enabled medical devices and Software as a Medical Device (SaMD) {15}, patients now have more options to monitor their health directly.

Digital therapeutic companies, a subset of SaMD, have already received FDA clearance for wearables and AI-applications that monitor everything from blood pressure to blood glucose to kidney function (16). Another example is Flatiron Health which provides Clinical Insights, EHR and Real World Evidence tools, and partners with cancer centers to deliver a better patient experience, strengthen practice health, and close the gap between care and research.

Regulatory Considerations

There is no question that AI's impact on healthcare is profound and far-reaching and will continue to have a critical influence on a field that touches all our lives. Whether used to predict public health concerns, enhance patient care, shorten the commercialization timeline, or even initiate drug discovery, its potential is immeasurable. However, as with any transformational technology, it will require careful consideration of regulatory, ethical, and organizational factors. To protect patient privacy and health information, be inclusive of diverse health representations, and ensure no harm comes from misinterpretation of models or inaccurate models, integrating AI into healthcare requires a multifaceted approach that balances innovation with ethical, legal, and practical considerations. Regulatory bodies must work closely with industry, healthcare professionals, and other stakeholders to create a framework that promotes responsible AI development and use but does not unduly stifle innovation. The dynamic nature of AI technology also means that these regulations must be flexible and adaptable to keep pace with ongoing advancements in the field.

Below are five key areas that life science and healthcare organizations should consider for responsible and ethical AI:

Reproducibility: Regulations must ensure that AI models are transparent, standardized, and verifiable, producing consistent results across different runs.

Fairness: Regulatory bodies must enforce measures to detect and mitigate biases, promote diverse data sources, and establish ethical guidelines to prevent discrimination.

Reliability: Ensuring consistent and accurate performance requires quality assurance, continuous monitoring, maintenance, and risk management.

Ethical Considerations: Regulations must address patient privacy, informed consent, and transparency, holding developers accountable for errors.

Security: Data privacy can only be fully guaranteed with properly managed ownership of data flow and model deployment and usage.

Conclusion

Ongoing efforts to guide and support these outcomes, from groups such as the Alliance for Artificial Intelligence in Healthcare (AAIH) and the Pistoia Alliance, are essential to ensuring that AI's promise is fulfilled, delivering better outcomes for patients and the healthcare industry. By working together with government entities, regulatory agencies, and patient advocacy groups, we can ensure that the world recognizes that the future of healthcare is here and AI is at its heart.