Foundations of AI for Early Detection of Lung Cancer: Equity and Fairness, Physician Dynamics, and Screening Efficiency
Lucem Health and Medial EarlySign recently held a roundtable discussion titled “Foundations of AI for Early Detection of Lung Cancer,” aimed at sharing valuable insights from a diverse group of experts, including physicians, data scientists, and clinical researchers. These experts examined the essential components involved in creating AI models aimed at the early detection of lung cancer.
In parts one and two of this series, the panel shared insights on data used to train models, what it takes to validate models in the real-world, the clinical and financial impact of augmenting standard clinical screening with AI, and more.
The final installment of this roundtable blog series addresses equity and fairness within the context of screening, if AI will really be replacing physicians (No!), and the efficiencies gained when implementing AI-driven screening.
As a reminder, our panelists for the roundtable included:
- Eran Choman, Chief Clinical Research and Compliance Officer, Medial EarlySign
- Coby Metzger, MD, Data Scientist, Medial EarlySign
- Ben Glicksberg, PhD, Chief Data Science Advisor, Lucem Health
- Andrei Petrus, Senior Manager of AI Solutions, Lucem Health
- Moderator: Marilyn Agbeko, Product, Lucem Health
You can access the recording of the discussion, here.
Promoting Fairness in AI-Driven Screening
Achieving fairness in AI-driven healthcare screening is a multifaceted challenge that requires careful consideration during the development and deployment of AI models. The goal is to ensure that these tools assist in making healthcare more equitable rather than exacerbating existing disparities. Andrei Petrus discusses the importance of striving for equity not only in the performance of the AI model itself but also in the broader context of healthcare delivery. This dual focus helps to identify and mitigate biases that might favor certain groups over others, such as insured versus non-insured or one ethnicity over another.
Equity in Model Performance
The pursuit of equity in model performance involves rigorous testing across diverse populations to ensure that the AI’s predictive capabilities do not vary significantly among different demographic groups. This process is crucial for making the tools universally applicable and helpful across all sections of the society. As Petrus mentions, “So we measure parity amongst different groups as they interact with the healthcare system and as they receive treatment for the condition.” This approach helps in identifying any biases towards screening particular groups more than others and facilitating adjustments to the models to correct these imbalances.
Equity in Healthcare Delivery
Beyond model performance, ensuring equity in healthcare delivery involves scrutinizing how different populations access and interact with healthcare services and how these interactions are influenced by AI tools. The aim is to use AI to identify disparities in care and screening rates among various demographic groups, including those based on insurance status or ethnicity. By shining a light on these disparities, AI can help in tuning healthcare delivery processes to increase parity. Petrus notes, “So we can tune the models, and we generally try to increase parity through AI rather than promote bias and disparities of care.” This approach embodies a proactive use of AI to address and ameliorate healthcare inequalities.
The Intersection of AI and Physicians
Dr. Coby Metzger offers insightful perspectives that serve to demystify the operational synergy between AI and clinical practice. As he articulately points out, “Actually, the more pressing part of AI, which is the larger language models and the generative models are not in play right now. The models we are discussing here as clinical decision support tools do not replace the physician and they are absolutely not doing the [core work of a physician]”
Dr. Metzger’s commentary sheds light on a critical aspect of AI in healthcare – its role as a supportive tool rather than a replacement for human expertise. The infusion of AI into healthcare settings has sparked a spectrum of reactions, ranging from fear of job displacement among clinical professionals to unrealistic expectations of AI’s capabilities. By focusing on real-world applicability, Metzger bridges these contrasting views, proposing a balanced perspective on AI as an enhancement to, rather than a substitute for, the irreplaceable human elements of medical care.
Diving deeper into utility, Dr. Metzger emphasizes the supportive essence of AI technologies. He points out that AI, particularly when it comes to larger language models and generative models, should be primarily designed to augment and support the capabilities of healthcare professionals. This augmentation is particularly crucial in a field as dynamic and complex as healthcare, where the volume and intricacies of data can overwhelm even the most experienced professionals. Through its computational power, AI has the potential to distill vast data sets into actionable insights, thereby enabling clinicians to make more informed decisions.
Drawing from Dr. Metzger’s statements, it becomes evident that the envisioned landscape of healthcare is one where AI and clinicians work in concert. The overarching goal is not to construct an ecosystem where AI usurps the role of medical professionals but to create an environment where AI tools empower these professionals to deliver more precise, efficient, and personalized care.
“And the beauty of the AI system is that it does exactly that. But at a larger scale with more computational power. An experienced oncologist would probably be able to do the exact same thing and look at the clinical variables: the blood, the labs, and make a similar decision to the AI, but oncologists are not involved in screening for lung cancer. They are involved in diagnosing it after screening and treatment…So, it’s the exact same job which previously was prioritized by smoking status, age, and a few other variables. Now, it can be prioritized by more variables with a larger computational power behind it. So, effectively the job hasn’t changed. No one has been replaced, we’ve just added an additional substrate of knowledge onto the delivery, the screening, to patients at the highest risk.”
Such a collaborative approach not only harnesses the strengths of both AI and human judgment but also aligns with the ethical considerations of ensuring that the deployment of AI in healthcare enhances patient outcomes without compromising the personal touch that is fundamental to clinical care.
Streamlining the Healthcare Process
AI emerges as a transformative force, not merely through its capability to dissect and understand complex data but also in its potential to redefine efficiency within healthcare delivery systems. Ben Glickberg describes the multifaceted advantages that AI brings to the healthcare industry. He points out, “one added benefit is in addition to learning things that humans may have trouble learning from the complex data themselves is just streamlining, right? A lot of it is making the healthcare process more efficient and more streamlined. Which in and of itself can make it the overall benefit of the healthcare delivery and remove certain repetitive tasks from the doctor’s plate.” This statement captures the essence of AI’s role in healthcare—not as a replacement of human intellect but as a powerful adjunct aimed at amplifying efficiency and streamlining operations.
The essence of AI’s impact lies in its dual capacity to both unearth insights from data complexities that are impossible in human cognition and to usher in a wave of operational efficiencies across healthcare channels. This twin capability of AI is a cornerstone for its value proposition in healthcare. It’s not just about making sense of the vast sea of data but about repackaging the healthcare delivery model into something more agile, more responsive, and ultimately more effective in addressing patient needs.
With AI models handling the heavy lifting of data analysis and operational logistics, healthcare professionals can dedicate more time and resources to patient interaction and care provision. This shift not only enhances the quality of healthcare services but also elevates the patient experience, making healthcare more accessible and personalized. AI’s role, therefore, transcends technological innovation, touching the core of healthcare’s humanitarian mission—improving lives.
Navigating Resource Challenges
The integration of AI into healthcare practices, especially in critical areas like lung cancer screening, represents a frontier of immense promise and potential. Yet, as with any revolutionary shift, it is beset with its own set of challenges, principal among them being the allocation and management of resources. Eran Choman succinctly highlights this pressing issue, stating, “In a sense, the main challenge across the healthcare ecosystem is the lack of resources.” This simple yet impactful statement opens up a broader dialogue on the multifaceted nature of resource constraints in healthcare’s ongoing journey to embrace AI technologies.
The notion of ‘resources’ in this context is broad, encompassing not just financial investment but also the human capital required to develop, implement, and maintain AI systems, and the data infrastructure essential for AI models to function optimally. Despite competing claims, only a select few organizations have developed the purpose-built ML infrastructure and the talent to deliver models that achieve high levels of performance in both accuracy and fairness.
Financial resources are often the most immediate hurdle for many healthcare institutions, as developing and deploying AI technologies requires upfront investment in both technology and training. This includes the costs associated with acquiring state-of-the-art computational hardware, software licensing, and securing access to large, annotated datasets necessary for training AI models effectively.
Beyond the financial aspects, human capital represents another critical resource. There is a need for skilled professionals who not only understand AI technology but can also navigate the complex ethical, legal, and social implications of its application in healthcare. This includes data scientists, AI ethicists, legal experts, and healthcare professionals with training in data literacy. Moreover, the successful deployment of AI in healthcare hinges on interdisciplinary collaboration, underscoring the need for concerted efforts in education and training to prepare the next generation of healthcare workers.
The challenges extend into the technological domain, particularly in the development and maintenance of robust data infrastructures. As we addressed in part one of this series, AI models require extensive and diverse datasets to learn effectively, demanding advanced data collection, processing, and storage capabilities. Moreover, the quality of AI’s outputs is directly tied to the quality of input data, highlighting the critical need for organized and comprehensive datasets. This, in turn, adds additional challenges to address patient privacy, data security, and the ethical use of health data, further complicating the resource equation.
Choman’s earlier remark encapsulates these challenges succinctly, drawing attention to the critical need for strategic planning, investment, and collaboration in overcoming the resource constraints facing the integration of AI in healthcare. Addressing these challenges is not merely a technical or financial endeavor but a holistic one that requires addressing societal, ethical, and infrastructural issues. As healthcare continues to evolve with technological advancements, navigating these resource challenges will be paramount in realizing the full potential of AI to change patient care, diagnosis, and treatment outcomes, particularly in areas as critical as lung cancer screening.
Wrapping Up
We hope you have found this series valuable and encourage you to watch the entire roundtable discussion on our YouTube channel. If you have questions about early detection of lung cancer, model development, or any of the topics covered during the roundtable, please don’t hesitate to get in touch with Lucem Health or Medial EarlySign!
