On December 22, 2022, the Peking University Global Health and Development Forum 2022 was held with the main theme of Digital Transformation and Development Divides. Co-organized by the Beijing Forum, Asian Development Bank and PKU Institute for Global Health and Development, this Forum brought together world leading scholars, policy researchers and industry leaders from both China and international communities to share their insights and recommendations on the thematic topics, attracted over 10 thousands online viewers participated in the event. XIAO Rui-ping, Peking University Chair Professor and Dean of the College of Future Technology and Associate editor of New England Journal of Medicine delivered a keynote speech at the session of Digital Transformation in Healthcare.
Thank you, Professor Xu, for the generous introduction. It is my great honor to be invited to speak here today. I would like to extend my gratitude to the organizers for their kind invitation. I will now share the screen with you all. Due to the ongoing COVID-19 pandemic, my voice may be a bit rough. I apologize in advance.
The topic of my talk today is the use of artificial intelligence in healthcare and medicine. As many of you may already be aware, there is a growing global focus on this area. I will cover four different aspects in my presentation.
Firstly, I will discuss medical journals and the publication of AI research. Specifically, I will delve into how medical journals evaluate AI research. We use different criteria to evaluate an AI clinical trial. For example, we consider whether the clinical trial is rigorous and useful or even transformative for patients, as well as whether it is transparent and reproducible.
I would also like to bring attention to a couple of articles that have been published in BMJ about AI publication guidelines. The first is the START-AI Protocol, which emphasizes accuracy and completeness in reporting and was developed to improve the transparency of reporting data. The second is the CONSORT-AI Extension, which aims to promote transparency and completeness in the reporting of clinical trials with AI for interventions, helping editors, reviewers, and the general public to understand, interpret, and critically appreciate the design of clinical trials and the risk of bias.
Furthermore, at our journal, the New England Journal of Medicine, we strive to publish the best articles that focus on AI in medicine. To do this, we build partnerships with the research community, collaborate with doctors and specialists, and diversify our publishing strategies, including branching out to new journals and the newly established journal NEJM Evidence, which features a lot of AI in medicine.
For instance, in 2020 we published an article that focuses on an AI system used to detect papilledema from ocular fundus photographs. The article described how a deep learning system was able to successfully differentiate among optical disks with papilledema from normal disks and disks with other abnormalities. This serves as an example of the methodology that can be widely used for other diseases.
Next, I would like to focus on the use of AI in clinical practice. As we are aware, AI has enormous potential across the entire human lifespan. Historically, researchers have proposed and developed numerous clinical decision-making systems, starting from the mid-20th century. In the 1980s, many rule-based approaches were produced and were widely used in clinical practice. More recently, with the successful application of deep learning, we have seen a significant advancement in the field.
For the entire human lifespan, there are a plethora of different aspects and applications of AI in clinical practice. Firstly, for diagnosis, AI can be used in both imaging-related and phenotype and biomarker-related applications. Secondly, in genome interpretation. With the advent of deep sequencing and other high-technology, genome interpretation has become increasingly important for diagnosis and drug development.
Machine learning for biomarker identification and discovery is widely used in cancer research and clinical practice. Additionally, AI can be utilized for clinical outcome prediction and patient monitoring. Additionally, wearable devices have become a major industry for monitoring the inferior health status of patients. Lastly, autonomous robotic systems are increasingly being used in clinical practice, such as surgery.
One of the most important and successful applications of AI in clinical practice is cancer diagnosis based on imaging. It is an active field, however, there are several limitations, particularly regarding early diagnosis. There is a high rate of false positives, which can lead to the over-diagnosis of early-stage cancer. Therefore, great care must be taken when using AI in this application.
In 2018, Dr. Woo and his colleagues developed a machine learning-based approach for somatic mutation discoveries, which they presented in an article. They created an app system that can be easily used to develop a tumor sample-based mutation discovery system. This technology is particularly useful for diagnosis and treatment selection. When paired with various devices, such as a smartwatch, this technology can identify a wide range of conditions.
For instance, using a smartphone, the cardiovascular condition of atrial fibrillation can be detected, as published in a journal in 2019. This irregular pulse can be processed, diagnosed, and validated by EKG. Recently, a study was published showing similar device used to diagnose heart failure, a life-threatening condition. This is an amazing development and with this kind of device combined with remote systems, the potential applications of AI are vast.
Moreover, AI has the potential to transform the healthcare system. During the pandemic, we have witnessed the use of AI in global healthcare. Despite its power and widespread use, we are also facing many challenges, as AI systems heavily depend on training data, which requires large volumes of existing data. Therefore, data quantity and data sharing are crucial to the usefulness of these systems.
Additionally, it is essential to ensure that these tools get into the hands of the right practitioners, as previously emphasized. This is related to value systems, not just cultural but human value systems, how to handle and use these systems is a big policy-making process for a global picture. Furthermore, many regional areas lack the regulatory capacity to oversee and manage these technologies, which is a significant challenge for all of us. In 2020, we published a paper calling attention to this issue.
Next, I would like to delve deeper into how AI can improve clinical trials. As you may be aware, our journal publishes a majority of articles related to clinical trials. Traditionally, the approach for most clinical trials, particularly for randomized clinical trials (RCTs), has been a linear and sequential pathway. This process begins with early discoveries and moves through clinical development, which includes several phases such as phase one, phase two, phase three, and phase four. Each phase is both expensive and time-consuming, and the success rate is often low, less than 10%. Even after approval by the FDA, EMA, or CDE system, there are still long and arduous processes such as manufacturing, supply chain, launch, commercialization, and market surveillance that must be completed.
To improve this system, AI can be useful in several ways. Firstly, in the design of clinical trials, which is the most crucial step for the success of the entire process. This design must be science-based, determining the control arm and testing arm based on the design. Additionally, AI can be used to enhance patient enrichment, recruitment and enrollment by taking advantage of large databases to make more accurate judgments.
Another important use of AI in clinical trials is identifying high-functioning productive investigators and selecting suitable sites. This is especially important during pandemics or emergency developments, such as with vaccines and medicines. While data on investigators and sites may exist, it may be inaccessible to individual physicians. AI systems can assist in this identification process. Additionally, AI can improve patient monitoring, medication adherence, and retention.
Finally, using operational data to drive AI-enabled clinical trial analytical systems is also crucial. This kind of analysis and reporting is important for clinical trial design. AI can assess the flexibility of the protocol design for patient recruitment and use real-world data. AI can also be used to assess site performance through real-time monitoring, and can help analyze and interpret unstructured and structured data from previous trials and scientific publications.
After the design stage, AI systems can also empower the processes of patient enrichment, recruitment, and enrollment. The system can verify biomarker benefits in order to reduce variability and increase study power. Selecting patients is the most important step in this process and AI systems can assist in this by using measurable clinical endpoints based on biomarkers and phenotypes. For predictive enrichment, the system can identify patients who are more likely to respond to a specific treatment.
The workflow for this process includes using AI algorithms to optimize the data capture and digitalized standard clinical assessments, as well as sharing data across systems in combination with wearable technologies. These devices can be used for continuous patient monitoring and providing real-time insights for safety and effectiveness. This kind of continuous feedback is crucial and also to help predict the risk of dropouts, thus enhancing retention.
However, it is important to note that while AI systems are powerful and widely used in clinical practice, particularly in clinical trials, there are also several challenges to be addressed. AI systems require large amounts of high-quality training data that is representative of target populations, and if this kind of data is not available, the system will not be reliable. Additionally, it can be difficult to interpret the results generated by deep learning systems and such systems require a lot of professional learning and training. Methodologies for high-deep neural networks to generate diagnostic and treatment selection are also less clear and not well-established at the moment.
Another challenge in using AI in healthcare and medicine is the computing environment for collecting, storing, and sharing data. This challenge exists both nationally and internationally, and even across different hospitals. The use of technology can lead to bias in datasets, which can be a potential problem. For example, during the pandemic, a system for sepsis alerting was set up at Michigan University Hospital, but the dataset was generated by a single center, leading to bias in the system. Fortunately, the system was stopped in a timely manner. However, it is important to note that such bias can occur and efforts must be made to avoid it.
Despite these challenges, the outlook for the development and application of AI in healthcare and medicine is bright. We look forward to the continued evolution of AI systems to accommodate rapid advances in molecular and genomic science. It is also important to note that physicians must adapt to new roles as information integrators, interpreters, and patient supporters. Medical education systems also need to be reformed to provide clinicians with the necessary tools and methods to incorporate AI into healthcare.
In China, AI is projected to disrupt multiple sectors and generate up to 600 billion dollars to its economy. Specifically, in healthcare and life sciences, AI can generate about 25 billion US dollars. The identification of new drug targets and the design of new molecules is expected to contribute about 10 billion US dollars, while AI-enabled clinical trials are projected to contribute another 10 billion. Efficiency and productivity improvements are expected to contribute an additional five billion dollars. This highlights the exciting future ahead for the use of AI in healthcare and medicine. We look forward to the continued advancements and developments in this field. Thank you for your attention.
