Exploring the Role of Artificial Intelligence in Rheumatology

I looked at the joints. They spoke back to me—”I need more humanism,” they whispered.

To longtime readers, those two sentences may sound both familiar and alien, perhaps even a little humorous. That’s because those sentences were generated entirely by a computer using artificial intelligence (AI). It was simple, too: I just copied the text of 120 previous Rheuminations columns and entered them into a freely accessible, online AI software program (GPT-3).¹ Nine lines of code and two clicks later, the computer “wrote” an entirely new, fantastical 1,067-word article.

Rheuminations columns are only the beginning of the AI revolution. Artificial intelligence, to those who may be unaware, is “the capacity of a computer to perform operations and tasks analogous to learning and decision making in humans.”² AI is doing things, such as reading X-rays and diagnosing skin cancers, that we thought previously impossible for unsupervised machines to do.^3,4

As AI becomes more precise and reliable, there is no question that AI will have profound effects on the field of rheumatology, from direct clinical service to education, research and beyond. Should we be excited—or worried? Let’s rheuminate.

Reality Check

It’s not just the intelligence that is artificial.

Before we explore how rheumatology professionals see AI, it may be worthwhile to investigate how AI views rheumatology professionals. To do so, let’s visit an AI engine called DALL-E mini.⁵

DALL-E mini is AI software that creates totally new pictures based on what it has learned from a gigantic database of images. When I typed in “rheumatologist,” I was excited to see that DALL-E mini had generated nine pictures of white-coat-clad rheumatologists palpating various joints. But this excitement turned to worry as I realized that each blurry-faced virtual rheumatologist was, as far as I could tell, white and male. Even after resubmitting five times, not a single non-white or non-male rheumatologist showed up in the 45 images.

It’s a great example of how AI can mimic not only our intelligence but the foolishness of our societal biases.⁶ Indeed, without strict scrutiny, implementation of AI can entrench societal divides and injustices. Rheumatologists, along with other clinicians and patients, need to engage with computer scientists to ensure that AI software not only mimics our collective thought patterns, but also upholds our collective values, such as diversity, equity and inclusion.

Error 404

Patient not found

This quick example with DALL-E demonstrates that AI may amplify known, existing biases. Yet AI can also generate new biases that we could not have even imagined before. Rheumatology is a field whose practice demands a sense of humility and creativity, something that cannot, at this time, be easily coded into machines. Our classification criteria are not meant to be unfailing algorithms for machines to faithfully implement; as such, a degree of misclassification is inevitable.⁷ Moreover, we still don’t quite understand the immunopathogenesis of many diseases and make several assumptions to fill in those gaps. Therefore, asking AI to engage in diagnostic decision making based on these assumptions without stringent external validation may lead to unforeseen harms.⁸

Some intrepid researchers are starting to engage in this painstaking validation process. For certain purposes, such as predicting response to methotrexate in rheumatoid arthritis patients, AI has shown great promise.⁹ But when tasked with other high-stakes clinical decisions, such as predicting the diagnosis of ankylosing spondylitis, the results show a need for more refinement and validation.¹⁰

The precision of such technology will continue to advance, but what degree of imprecision will we be able to tolerate? After all,these are patients whose lives we are placing into the responsibility of computers with algorithms so convoluted their own programmers don’t know how they work.¹¹ When a patient doesn’t quite fit into an algorithm, how will AI cope? And what are the legal and ethical ramifications of outsourcing our clinical decision making to a computer?¹²

These questions will need to be explored further as AI encroaches more and more upon the duties and tasks originally intended purely for humans.

The Robot Will See You Now

Even then, even if we reach the point that AI can reliably make clinical decisions, we will need to ponder the ramifications on our healthcare workforce. Per the ACR’s 2015 workforce assessment, there will be a deficit of more than 4,800 rheumatology providers by 2030.¹³ It is very well possible that sophisticated AI-based algorithms can help address this crisis.

Existing rheumatology clinicians may be better able to use AI to automate burdensome and tedious tasks so direct patient care can be prioritized. Similarly, AI can help to support diagnostic decision making to facilitate care so that greater numbers of patients can be seen promptly. At its most ambitious, AI may support primary care providers identify those at risk for diagnostic delays or undertreatment, obviating even the need for a rheumatologist.¹⁴

But we’ve heard these sorts of promises before, with electronic health records (EHRs). And although EHRs have been quite helpful, the burden of documentation and administrative work has been a major driver of burnout, a contributor in and of itself to the workforce crisis.¹⁵ If clinicians and patients are not in the driver’s seat in programming and implementing AI for real-world clinical settings, I anticipate more administrative tasks, clicks and overall waste, furthering our burnout. And this doesn’t even get into the very real potential of clinicians having to complete prior authorizations and peer-to-peer requests through an AI-powered insurance robot.¹⁶

The Joints Were Right

We need more humanism

I confess: When I initially read how the joints whispered to the clinician that they needed more humanism, I thought “What clichéd nonsense is this?” But the more I thought about it, the more I realized that the machine actually proposed the only path for rheumatology professionals to balance the risks and benefits of AI: We have to wholeheartedly embrace our own humanism. At this point in time, we need to ensure the greater efficiency provided by AI will afford us more agency to be humanistic to those whispering joints—and the humans that use them.

Altogether, this means we must prioritize our engagement with AI. Important first steps include increasing funding for research on informatics, artificial intelligence and machine learning, bolstering information technology support for clinical divisions, training fellows to familiarize themselves with AI and scrutinizing quality improvement work to ensure it embraces the principles of human-centered design. Moreover, we have to do this promptly, or events may swiftly overtake us. This is exemplified by the last, seemingly ominous, words of that AI-generated Rheuminations column:

The joints were happy and satisfied. Humanism prevailed, at least for the day.

Guest columnist Bharat Kumar, MD, MME, FACP, FAAAAI, RhMSUS, is the associate program director of the rheumatology fellowship training program at the University of Iowa, Iowa City. He will be assuming the reins of physician editor of The Rheumatologist from Philip Seo, MD, MHS, with the January 2023 issue. Follow him on Twitter @BharatKumarMD.

References

1. OpenAI API (n.d.). https://openai.com/api/.
2. Artificial intelligence. Dictionary.com (n.d.). https://www.dictionary.com/browse/artificial-intelligence.
3. Gupta A, Sheth P, Xie P. Neural architecture search for pneumonia diagnosis from chest X-rays. Sci Rep. 2022 Jul 4;12(1):11309.
4. Vaiyapuri T, Balaji P, S S, Alaskar H, Sbai Z. Computational intelligence-based melanoma detection and classification using dermoscopic images. Comput Intell Neurosci. 2022 May 31;2022:2370190.
5. DALL·E mini – by Craiyon.com. https://huggingface.co/spaces/dalle-mini/dalle-mini. Accessed July 8, 2022.
6. Kusters R, Misevic D, Berry H, et al. Interdisciplinary research in artificial intelligence: challenges and opportunities. Front Big Data. 2020 Nov 23;3:577974.
7. Aggarwal R, Ringold S, Khanna D, et al. Distinctions between diagnostic and classification criteria? Arthritis Care Res (Hoboken). 2015 Jul;67(7):891–897.
8. De Cock D, Myasoedova E, Aletaha D, et al. Big data analyses and individual health profiling in the arena of rheumatic and musculoskeletal diseases (RMDs). Ther Adv Musculoskelet Dis. 2022 Jun 30;14:1759720X221105978.
9. Duong SQ, Crowson CS, Athreya A, et al. Clinical predictors of response to methotrexate in patients with rheumatoid arthritis: A machine learning approach using clinical trial data. Arthritis Res Ther. 2022 Jul 1;24(1):162.
10. Deodhar A, Rozycki M, Garges C, et al. Use of machine learning techniques in the development and refinement of a predictive model for early diagnosis of ankylosing spondylitis. Clin Rheumatol. 2020 Apr;39(4):975–982.
11. Beltramin D, Lamas E, Bousquet C. Ethical issues in the utilization of black boxes for artificial intelligence in medicine. Stud Health Technol Inform. 2022 Jun 29;295:249–252.
12. Gerke S, Minssen T, Cohen G. Ethical and legal challenges of artificial intelligence-driven healthcare. Artificial Intelligence in Healthcare. 2020:295–336.
13. Battafarano DF, Ditmyer M, Bolster MB, et al. 2015 American College of Rheumatology workforce study: Supply and demand projections of adult rheumatology workforce, 2015–2030. Arthritis Care Res (Hoboken). 2018 Apr;70(4):617–626.
14. Meskó B, Hetényi G, Győrffy Z. Will artificial intelligence solve the human resource crisis in healthcare? BMC Health Serv Res. 2018 Jul 13;18(1):545.
15. Tajirian T, Stergiopoulos V, Strudwick G, et al. The influence of electronic health record use on physician burnout: Cross-sectional survey. J Med Internet Res. 2020 Jul 15;22(7):e19274.
16. Hut N. Revamping prior authorization: How AI and automation could boost care and revenue. Healthcare Financial Management Association. 2021 Oct 5. https://www.hfma.org/topics/technology/article/revamping-prior-authorization-how-ai-and-automation-could-boost-care-and-revenue.html.