Will Artificial Intelligence Outperform the Clinical Neurologist in the Near Future? No.

Abstract

Medicine, neurology, and movement disorders are familiar with the implicit tension between evidence and judgment, between technology and experience, and between machines and humans. The outperformance contest posed in this Controversy places Artificial Intelligence and Clinical Judgment on a balance scale, charging my colleague and me to tilt that balance in one direction or the other. Whereas I consider best practice to involve synergy between these two rather than a battled obliteration of one by the other, I am still comfortable arguing in favor of one side of this argument. While I am highly vested in the recognition of Artificial Intelligence as a growing and informative tool, I do not see it as a force to replace or displace the essential role of clinical judgment in the management of neurological patients. In considering my strategy of argumentation, I reflected on the teachings of Jean-Martin Charcot (1825–1893), the first clinical neurologist of international fame and the teacher to many leaders of the next generation. He was largely responsible for the development of the nosology or categorization of neurological diseases, the leader who anchored neurological diagnosis in correlations between clinical signs and pathological lesions, and he contributed fundamentally to the descriptions of numerous movement disorders. Charcot viewed his neurological service as an enterprise composed of divisions and specialty units collecting data that contributed to diagnostic and treatment decisions. He followed his patients longitudinally, studying them from multiple perspectives that incorporated into medical practice the advances offered by new scientific discoveries. In this light, Charcot too saw the implicit pull between the experience and knowledge base of a well-trained clinician and the laboratory discoveries of scientists working beside clinicians or outside the immediate scope of the hospital and clinic. To anchor this Controversy in a historical context and to show how Charcot wrestled with the 19th century equivalent of Artificial Intelligence or “big data” and its place beside clinical medicine, I therefore offer the three-part quotation, known as Charcot's Credo.1 In advocating for the enormous contributions provided by new scientific data, Charcot nonetheless always felt that the clinician was the ultimate arbiter in medicine, a view that embodies my own argument. I firmly believe that in medicine there are areas that belong solely to doctors, that only they can properly cultivate and bring these areas to fruition. These domains are necessarily closed to the scientist who day in and day out is confined to the laboratory and would disdain the teaching methods of the hospital. But I believe also with equal fervor what is by and large accepted by all today that the laboratories must have input into the issues of medical science in order for medicine to progress. To me, the practice of medicine has no real autonomy; it exists by borrowing and making new applications of ideas from other disciplines. Without a constant reinfusion from these other scientific domains, the practice of medicine would soon become an outmoded routine. Here, he rightly argues that a polarized construct, as affirmed in this Controversy, may be convenient but it is inaccurate, and the harmonization of data filtering from the laboratories facilitates medical decision-making with no need to compete with or displace clinical judgment. If you set aside the questions of clinical insight and skill, and the innate artistic gift that a clinician perfects bit by bit through the many years of experience, I would agree that the expert in the laboratory has just as much importance in medicine as the physician. His conclusion, and mine as well, is that clinical skills and insight, perfected over years of experience, cannot be duplicated or displaced. And, hence the balance of the scale must ultimately tilt in favor of Clinical Judgment. I offer this same Charcot-based sequence of considerations with a few examples from my own scientific and medical career. As the first example, I allude to the development of the standard rating scale that our field utilizes in assessing Parkinsonism, the MDS-UPDRS.2 This scale was developed to be a comprehensive inventory of symptoms and signs and to have particular applicability in the early phases of Parkinson's disease.3 Items were selected based on the clinical experience of movement disorder specialists along with patient input and then calibrated using a variety of laboratory-based data on speed, decrement, amplitude and constancy. The descriptive text for items (Fig. 1, blue) is largely based on quantifiable measures, but the final selection must meet the clinical conceptual vocabulary ranging from Normal to Severe involvement for that domain (Fig. 1, red). As such, Artificial Intelligence “big data” techniques can allow a scale development team to calibrate with precision (blue), but the cut-off allocations and final judgment of disease severity that will anchor the final decision requires Clinical Judgment (red). The two must work together, but the Clinical Judgment is purposefully placed before the descriptors because of its greater importance to the final decision. As a second example, I turn to another area of my research: visual hallucinations. Given that the field of movement disorders is a visual specialty, researchers are inherently challenged in trying to study visual hallucinations because we cannot specifically see what we are studying. To this end, like Charcot, I and fellow researchers have turned to ancillary scientific fields, genetics, neuroimaging, neurochemistry, and biomarker research to complement clinical observations.4-7 These strategies yield “big data” that are of high interest and import to understanding the connectomes and pathways involved in hallucinations, to modeling the responses to stimuli, and to identifying machine learning strategies. However, in the final rating of hallucination impact and severity, Clinical Judgment must make the ultimate determination through the clinician's interaction with the patient and family. In returning to rating scales and assessments, the final rating of a patient with hallucinations is left inescapably in every study to an assessment based on Clinical Judgment.2 As the International Parkinson and Movement Disorder Society moves to develop a new MDS Parkinson's Disease Psychosis Scale, the same principles will hold. Finally, I offer an example derived from clinical observation but pivotal to clinical trial design and analysis. All clinicians will agree that there are two well-established clinical archetypes of Parkinson's disease, a tremor predominant and an akinetic-rigid phenotype.8 Even if most patients present with a typical picture encompassing signs from both phenotypes, the isolated archetypes evoke the possibility that two fundamentally different pathophysiological processes underlie Parkinson's disease impairment. In statistical modeling, this conceptual distinction would characterize two clinical constructs, known as thetas for the Motor Examination of the MDS-UPDRS. As such, if a large dataset of MDS-UPDRS cases were examined, items assessing tremor and items measuring non-tremor signs could be tested as potentially two dimensions underlying the construct of parkinsonism rather than one. In fact, in generating this hypothesis, based entirely on Clinical Judgment, such Artificial Intelligence-based modeling, when applied to MDS-UPDRS scores from over 6000 subjects, identified these two entirely separate dimensions.9 Similar to the other examples, the Artificial Intelligence approaches confirmed and strengthened the conclusions, but the core concept and study design were proposed and solidly anchored in Clinical Judgment. Believing in the importance of big data and Artificial Intelligence strategies, I emphasize how much the MDS is investing in the infusion of new programs that rely on these technological advances. The Society is sponsoring the Technology Task Force and supports large data initiatives such as the MDS-Gene, eDiary, and Rating Scales programs. Electronic applications for MDS products, the Educational Road Map, and Social Media programs are ever-advancing in sophistication and outreach (www.movementdisorders.org). Nonetheless, in all cases, the prioritization falls ultimately to clinical medicine with no plan to have Artificial Intelligence products replace or displace the current judgment process. For these reasons and with the confidence of speaking in the halo of the grand luminary, Charcot, I conclude that NO, Artificial Intelligence will not outperform the Clinical Neurologist in the near future. But, at the same time, I am very happy to have Artificial Intelligence as my ally for the better care of patients. For, Charcot emphasized over 100 years ago: “When a patient comes to see you, he is under no obligation to have a simple disease just to please you.”10 (1) Research project: A. Conception, B. Organization, C. Execution; (2) Statistical Analysis: A. Design, B. Execution, C. Review and Critique; (3) Manuscript: A. Writing of the first draft, B. Review and Critique. C.G.: 3A, 3B Ethical Compliance Statement: Neither patient consent nor the approval of an institutional review board was needed for this work. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this work is consistent with those guidelines. Funding Sources and Conflicts of Interest: The author received no specific funding for this work and has no relevant conflicts of interest to declare. Financial Disclosures for Previous 12 Months: Grants/Research: Funding to Rush University Medical Center from NIH, Department of Defense, and Michael J. Fox Foundation for research conducted by Dr. Goetz. Honoraria: Presidential stipend from the International Parkinson and Movement Disorder Society paid to Rush University Medical Center as part of Dr. Goetz's salary. Faculty stipends from the American Academy of Neurology. Guest professorship honorarium provided by University of Chicago and NorthShore University Health System. Royalties: Elsevier Publishers, Wolters Kluwer Publishers. Salary: Rush University Medical Center.