U.S. Healthcare: Is AI a Threat or a Solution?

Much has been written recently about AI and healthcare. Is it a latent threat or a solution to the existing healthcare crisis? Does the answer to this question change if you translate AI as ‘augmented intelligence’ rather than ‘artificial intelligence’? For this narrative, let’s consider AI to mean the former.

Note: This is the final part of this blog series on US Healthcare. If you haven't read the previous parts, we recommend you to read it from our official website or from the links provided at the end of this blog.

AI = Augmented Intelligence

While the term ‘AI’ conjures up images of exotic and mysterious technology, the reality is far more mundane—much like the Wizard-of-Oz (who turns out to be nothing more than a man behind a curtain) or ‘the cloud’ (which is just another way of saying ‘somebody else’s computer’). AI is not a single technology; it’s an umbrella term that covers a range of technologies, each with different capabilities and levels of complexity. These technologies are not in of themselves mysterious and can be explained relatively easily. Deep-learning neural networks, the computational bedrock of AI algorithms, consist of digitized inputs (e.g., images, text, numbers) put through multiple layers of connected ‘neurons’ that progressively detect features (patterns) and ultimately provide an output. The output must then be validated in independent cohorts to determine its accuracy and generalizability. As expected, the output quality is directly related to the quality of the data inputs. While there is a threshold requirement for the size of the derivation cohort, it is not critical—larger derivation datasets do not necessarily provide more accurate and reliable outputs. Additionally, AI systems are autonomous and dynamic, learning and adapting as more data become available.

Healthcare applications

Computers can run thousands of tasks at once, don’t get tired, don’t forget what they have learned, and—once appropriately trained—can recognize patterns in data more quickly than most humans. AI is best seen as a tool to enhance human competence; an assortment of technologies to create greater synergy between the human and ‘AI clinician.’ AI can be a trusted and capable assistant, augmenting human capabilities across the value chain. When AI works alongside human decision-makers and when it’s data inputs are accurate, unbiased, and representative, it can significantly improve speed, efficiency, cost, and capacity for a wide range of processes in healthcare. It is already being used widely across the industry, especially in the fields of radiology, pathology, dermatology, and cardiology, to variable effect.[1-3] To date, the biggest barrier to the AI healthcare revolution has been implementation.[4] To fully harness the potential of these formidable technologies, providers will need to be well-versed and proficient in how to incorporate these algorithms into their clinical workflow.

 "Our intelligence is what makes us human, and AI is an extension of that quality.” -- Prof. Yann Lecun, NYU --

Disruptive potential of AI

Within the healthcare space, AI has the potential to be disruptive across three major domains:

1. Operational efficiency. Healthcare organizations are already using AI to improve the efficiency of both back-office administrative tasks (e.g., documentation, scheduling, workflow, coding/billing, gathering and sharing information) and patient care (e.g., access, drug safety, 24/7 virtual/chatbot ‘nurse assistants’). Shifting administrative tasks from providers to computers frees up time for clinicians to engage in more direct patient care, where human judgment and interactions matter most. It also has the potential to engage patients, enabling them to access and process their own data to promote health.
   

2. Diagnosis and clinical decision-making. AI solutions can be used to help providers analyze large datasets to assist in clinical decision-making. Potential benefits include, among others, increased efficiency in healthcare diagnoses, reduced misdiagnosis rates, fewer medical errors, and the ability to incorporate data from patient fitness monitors and wearables. According to Harvard’s School of Public Health, it is estimated that “if implemented correctly, AI could improve health outcomes by up to 40% and reduce treatment costs up to 50% by improving diagnosis, increasing access to care and enabling precision medicine.”[5]
   

With AI as an ally, precision medicine can finally transition from generic treatment models to a more patient-centric approach, with personalized treatment plans designed to enhance both patient care and resource allocation. With time, healthcare professionals will learn to leverage AI in augmenting the care they deliver, allowing them to provide safer, more evidence-based, and more effective care. For example, clinicians could use an AI ‘digital consult’ to examine ‘digital twin’ models of their patients, allowing them to ‘test’ the efficacy and safety of an intervention (such as a cancer drug) in the digital environment before delivering the intervention to the patient in the real world.[3]

A word of caution. Validation of an algorithm’s accuracy is not the same as demonstrating clinical efficacy—a phenomenon often referred to as the ‘AI chasm.’ An algorithm with an AUC of 0.99 is not worth very much if it does not improve clinical outcomes.[6]

3. Predictive models for risk stratification. The increasing availability of multi-dimensional (sociodemographic, clinical/phenotypic, genomic, metabolomic, economic, social determinants) and multi-modal datasets coupled with technologic innovations in computing power and data security offers the opportunity to fundamentally transform models of healthcare delivery. AI/machine-learning models can be developed to predict individuals and populations at risk of a variety of outcomes, such as particular diseases (cancer, heart disease, renal failure) or accidents or to predict hospital readmissions. On an individual level, the ability to leverage such precise prognostics can assist providers in creating personalized care plans to mitigate future risks. On a population level, insights from such predictive models can be used to promote proactive preventative care interventions, optimize resource allocation, and address drivers of health disparities.

Risks

It is not possible to entirely de-risk new technologies. AI is no exception. A common concern is that AI will lead to automation of jobs and substantial displacement of the workforce. Although there are many instances in which AI can perform healthcare tasks as well as or better than humans—such as recognizing skin cancer,[7] and providing more rapid and accurate interpretation of radiologic images[8] and pathology specimens[9]—implementation challenges and ethical concerns (discussed below) will almost certainly prevent large-scale automation of healthcare professional jobs for a considerable period.[10] 

Like humans, AI will not produce perfect results every time. This is because data inputs are never perfect, especially population-level data that are subject to inaccuracies, bias, and data gaps. To avoid system errors and patient harm, existing limitations in AI technology will need to be mitigated, including algorithmic bias, concerns about privacy/security, and lack of transparency. Prof. Eric Topol offers some sage advice: “Over time, marked improvements in accuracy, productivity, and workflow will likely be actualized, but whether that will be used to improve the patient-doctor relationship or facilitate its erosion remains to be seen.”[2]

Ethics and AI implementation

As for all technological advances—the internet, social media, drones, CRISPER-Cas9 gene editing, and digital cryptocurrency (bitcoin) to name just a few—AI can be used for good or nefarious purposes. The introduction of AI tools to automate and support clinical decision-making brings with it a host of serious ethical questions. Of particular concern are issues related to algorithmic bias and the potential for discrimination, the delegation of agency and authority from humans to AI, and the complexity and lack of transparency in algorithmic systems (the so-called “black box problem”) that makes it difficult to assign accountability.[11] In June 2021, after 18 months of deliberation with leading experts in ethics, digital technology, law and human rights, and public health, the W.H.O. produced a report entitled Ethics & Governance of Artificial Intelligence for Health. It summarizes the ethical challenges and risks to using AI in healthcare, and outlines six principles to ensure that AI is used responsibly in the best interests of public health: (i) Protecting autonomy, (ii) Promoting human safety and well-being, (iii) Ensuring transparency, (iv) Fostering accountability, (v) Ensuring equity, and (vi) Promoting tools that are responsive and sustainable.[12] AI is becoming increasingly important in healthcare. It is not going away. “AI is neither good nor evil,” said Oren Etzioni, founding CEO of the Allen Institute for Artificial Intelligence (AI2), “It’s a tool; it’s a technology for us to use. To paraphrase spiderman: with great technology comes great responsibility. The choice about how it gets deployed is ours.”[13]

The opportunity

U.S. healthcare needs bold solutions. AI is arguably the most exciting of these prospects, and has the potential to be disruptive across the full breadth of the industry.[1,3,6] AI technology is advancing rapidly—becoming increasingly sophisticated at doing what humans do, but doing it more efficiently, more quickly, and at lower cost. FDA approval of AI applications is accelerating, and organizations ranging from NIH to industry and academic institutions are getting into the AI business. Healthcare organizations that fail to maximize the technology’s promise and exploit the disruptive potential of AI—or fail to do so in a timely fashion—will be left behind.

Previous parts of this blog series: US Healthcare: Misaligned Incentives, Why is US Healthcare so expensive?, US Healthcare: Lack of Engagment, Tracking the satisfaction of end-user, the patient, and Innovation in Healthcare

Authors: This Blog is Co-authored by Dr. Errol Norwtiz and Venkata N. Peri

References:

1. What doctor? Why AI and robotics will define New Health. PWC, 2017. www.pwc.co.uk/intelligent-digital/id-healthcare/PwC-Why-AI-and-robotics-will-define-New%20Health.pdf

2. Topol EJ. Deep Medicine: How artificial intelligence can make healthcare human again, New York, NY: Basic Books; 2019.

3. Bajwa J, Munir AU, Nori A, Williams B. Artificial intelligence in healthcare: transforming the practice of medicine. Future Healthc J 2021; 8:e188-94. doi: 10.7861/fhj.2021-0095

4. Aristidou A, Jena R, Topol EJ. Bridging the chasm between AI and clinical implementation. Digit Med 2022: 399(10325):P620. doi: 10.1016/S0140-6736(22)00235-5.

5. https://www.hsph.harvard.edu/event/exec-ed-online-applied-artificial-intelligence-for-health-care-2/

6. Topol EJ. High-performance medicine: the convergence of human and artificial intelligence. Nat Med 2019; 25:44-56. doi: 10.1038/s41591-018-0300-7

7. Haenssle HA, Fink C, Schneiderbauer R, et al. Man against machine: diagnostic performance of a deep learning convolutional neural network for dermoscopic melanoma recognition in comparison to 58 dermatologists. Ann Oncol 2018; 29:1836-42. doi: 10.1093/annonc/mdy166

8. Park EK, Kwak SY, Lee W, et al. Impact of AI for digital breast tomosynthesis on breast cancer detection and interpretation time. Radiol Artif Intell 2024; e230318. doi: 10.1148/ryai.230318

9. Li S, Ye X, Tian H, et al. An artificial intelligence model based on transrectal ultrasound images of biopsy needle tract tissues to differentiate prostate cancer. Postgrad Med J 2024; 100:228-236. doi: 10.1093/postmj/qgad127

10. Davenport T, Kalakota R. The potential for artificial intelligence in healthcare. Future Healthc J 2019; 6:94-98. doi: 10.7861/futurehosp.6-2-94.

11. Gjødsbøl IM, Ringgaard AK, Holm PC, et al. The robot butler: How and why should we study predictive algorithms and artificial intelligence in healthcare? Digit Health 2024: 10:1-13. doi: 10.1177/20552076241241674

12. www.who.int/publications/i/item/9789240029200

13. www.youtube.com/watch?v=H9Ddd1H9uLE