Artifical intelligence use in orthopedics: an ethical point of view

Introduction

The age of artificial intelligence (AI) began during the second half of the 20th century. The term ‘artificial intelligence’ was coined in 1955 by John McCarthy, a computer and cognitive scientist, and his associates (1). They proposed that if certain aspects of human intelligence, such as learning, abstraction, language, and problem-solving could be described in an accurate and detailed way, it would be possible for a machine to simulate these human capabilities. Today, the evolution in machine learning (ML) has reached a level of sophistication that enables machines to gather data, learn, analyze a process, define a problem, and come up with a solution. These capabilities have resulted in various degrees of defining AI and its associated terms (see Table 1). In recent years, AI has infiltrated every aspect of life, including medicine. In the medical arena, AI typically occurs in the form of either algorithms or robotic technologies. The ability to gather, analyze, and store big data has resulted in the creation of algorithms for calculating specific risks, expected patient satisfaction, and even the expected success rate for procedures or medications. Additionally, in orthopedics and other disciplines, evolving robotic technology has reached a state that could allow a surgeon, at least hypothetically, to remotely operate on patients. With the ability to assist in decision-making processes, increase surgical precision, and minimize the time spent on certain aspects of surgery, such as the use of endoscopic knot-tying robots, for example, AI and robotic technology have earned a permanent seat in the operating theater.

Ethical concerns have arisen, however, alongside the marked improvements and widespread utilization of AI in orthopedics. In 2019, the European Federation of National Associations of Orthopaedics and Traumatology (EFORT) published a guideline on aspects of AI applications in the field of orthopedics and traumatology that warrant ethical consideration (2). Examples include the protection of patient privacy while gathering and utilizing patient data, algorithm biases created based on the data collected, and cybersecurity concerns or accountability for undesirable outcomes (2, 3). The aim of this paper is to discuss the pros and cons of AI applications in orthopedics and traumatology from an ethical perspective.

Orthopedic applications of artificial intelligence

AI has gained increasing popularity in orthopedic practice (9, 10), including clinical, preoperative, intraoperative, and postoperative settings. AI also has a role in the research setting. Federer et al. published a scoping review on AI in orthopedics in 2021, which emphasized the noticeable increase in the number of AI-related published articles during the previous 3 years (11). This body of research was primarily focused on image interpretation and clinical decision-making. The most frequently reported AI-associated body region was the spine (43%), followed by the knee and hip, respectively (11).

Ethical considerations

As an integral part of medical practice, AI could be revolutionary and groundbreaking. It is no longer in the realm of ‘science fiction’ that AI will become an indispensable component of surgical practice, with continuing developments in the field of data science, computer science, and robotic technologies. These advances will be beneficial for patients, as AI offers the potential to improve patient care, estimate the risk of morbidities or complications, aid diagnosis, make preoperative and intraoperative assessments, assist in surgical procedures, provide feedback to the surgeon, and monitor postoperative patient care (9, 20). However, rapid progress that challenges regulatory mechanisms may also involve dangers that lead to ethical dilemmas. As exciting and groundbreaking as the pros of AI applications are, the cons also need to be thoroughly assessed. To address these concerns, some countries have already developed specific regulatory standards for the use of AI in clinical practice (21).

From a medical ethical standpoint, an innovation should be in harmony with the Declaration of Geneva from the World Medical Association, with the intention that AI in medicine should not be used to evoke discrimination with respect to age, gender, race, religion, disability, or other social or cultural criteria (2). At first glance, one could assume that this principle would be followed in good faith. However, that is not always the case. A major problem is often posed by unconscious discrimination. AI relies on computing big data, finding patterns through the data, and constituting algorithms on that basis. Big data may contain big biases as it is not always clear whether the data represent an entire population or a specific subset. If certain minorities are not well represented and evaluated, for example, the training data will be flawed, and the algorithm based on that set will be faulty. To avoid misrepresentations and establish data safety, data sets must be carefully monitored (20). The quality of medical data should also be regulated and monitored. Otherwise, low-quality medical records will lead to faulty guidance and misrepresentations in data sets. In addition, changes in the population and medical practices should be recorded regularly to update AI processes. Such a barrier could be overcome by the continuous monitoring of data sets and state-of-the-art medical knowledge and practice, as well as rechecking and adjusting AI algorithms, accordingly (3).

Patient privacy, consent, and confidentiality should also be taken into account when data is collected for AI applications in orthopedics (9). The corporations developing AI technology often gain access to the stored data of a particular patient population. This type of access could violate both patient privacy and confidentiality, in addition to violating data protection regulations. Patients must have the right to control their personal data, including how their data will be used, for how long, and for which purposes. Once data has been accessed, it should be clear that patients do not lose their rights to withdraw their consent; yet, in practical terms this right cannot be guaranteed because the relevant data has already been processed. For purposes of data protection, patient data may be anonymized. However, such anonymity also means that patients lose control over their data once it has been accessed (14).

Informing patients about AI-associated procedures could potentially pose an ethical problem. A patient can give consent after he/she has been fully informed about the procedure to be performed, other possible alternatives, the rationale leading to the treatment choice, potential risks, and associated complications. If the AI procedures run into the ‘black box phenomenon’ such that neither the doctor nor the patients understand the underlying procedures, the concept of informed consent becomes worthless. Consent may have been provided, but if the information is not properly understood, such consent may not be informed (3). One can assume that informing the patient with a statement such as ‘our algorithm has chosen this procedure for you, and we think it is the right choice for you, but, we cannot explain why’ cannot be acceptable or sufficient for obtaining consent.

Furthermore, the automation of procedures, such as a robot planning the bony incisions, directing the incisions, and informing the surgeon about the accuracy of his or her instrumentation during a TKA, may lead to a phenomenon called ‘deskilling’ (22, 23). This phenomenon reflects the concern that surgeons could lose their high standards of knowledge, skills, and experience if technological aids are excessively used. Under normal circumstances, this may not be an important issue; however, in the event of a system collapse due to malware, a cyberattack, or any other possible technical issue, the surgeon who has lost the ability to manage the situation could cause serious damage to a patient.

At the present time, one might assume that surgeons can manage every type of issue that can also be managed automatically through AI. In the event of a malfunction, therefore, the physician in charge would realize this and act upon it. If a surgeon does not utilize conventional procedures on a daily basis, or if these are omitted from the medical education program altogether and handed over to AI, conventional surgical skills may be gradually lost. Until AI procedures include a better solution for preventing this scenario, medical professionals should continue to learn and practice conventional methods as well.

Continued practice of conventional techniques is also important if AI contradicts the physician with regard to an indication or a diagnosis. In this situation, a surgeon may question whether he/she should pursue his or her own decision or proceed with what AI has suggested. In this scenario, classical surgical knowledge and experience are important. The primary concern is that the physician who lacks background knowledge and experience may lose his or her self-confidence based on these factors and will have no choice but to be overly reliant on AI algorithms. In the event of faulty algorithms, for whatever reason, over-reliance on AI may ultimately cause harm to patients (9).

Cybersecurity is also an important factor. Data security and patient privacy could be easily violated if data are stolen. Moreover, software systems can be attacked at any time, leaving them vulnerable to the total collapse of the systems relying on them. Malware and threats may cause system collapse or incorrect guidance, both of which would have detrimental consequences. Threats will always exist that need to be neutralized. As noted by Finlayson, securing a system is more difficult than finding a back door for breaking the system (3, 24).

Finally, accountability is a serious concern in the use of AI technologies (20). Imagine the following scenario: A corporation develops a specific automated medical technology. The regulatory parties either approve or disapprove it. Upon approval, the product appears on the market. Hospitals acquire the technology, and specially trained medical practitioners adopt it in their regular practice. We should consider AI as a supplementary tool that makes some decisions and provides answers instead of, or to support, a physician, a situation that could lessen the burden, adjust workload, prevent burnout, and create extra time for surgeons to engage with patients, etc. In the case of surgical procedures, utilization of these technologies may yield greater surgical precision, increased accuracy of implant positioning, and fewer human errors. We could continue to enumerate the upsides; however, if we look at the downsides, such as an unwanted event occurring with this technology, the critical question becomes: who will be held accountable? The computer scientist, the corporation, the regulatory agencies, the hospital, or the doctor? Certainly not the computer. The issue of accountability remains a tough and challenging question, and legal systems may require regular updates, or more information, in order to answer it.

Conclusion

At some point in the future, computer technology may reach a level where a human presence is no longer necessary in medical practice. For the time being, such a scenario appears unrealistic. The degree of technical development in AI could be regarded as a utopia come to life; however, such a utopia also presents a dystopic challenge. As mentioned previously, machines require supervision. Data sets also need to be monitored and discrepancies checked. Human beings are still required for these tasks. Machines do not get tired and can work continuously. Such functions may eliminate human error, but what about computer error? It is well known that algorithms can get confused by simple changes to expressions or images, resulting in a different conclusion when confronted with the same situation, an error that humans are not likely to make. With these scenarios in mind, AI is currently supported by augmented intelligence, in which a medical practitioner is in charge of caring for the patient, while supported by assistance from AI systems.

In conclusion, existing applications of AI in orthopedics are gaining enormous ground. AI and robots play important roles in making clinical decisions, analyzing images, establishing diagnoses, and estimating risks, morbidities, and complications. In addition, AI assists in performing some of the steps in specific orthopedic procedures and during postoperative patient monitoring. New developments are accompanied by new ethical challenges and AI in orthopedics is not devoid of these concerns. Patient privacy, consent, data protection, cybersecurity, the safety and monitoring of data, bias, and accountability are some of the ethical issues associated with AI. Currently, differing points of view on these topics are being debated, but facing the current stage of development is only the beginning. As AI technology improves, new and unprecedented ethical dilemmas may become increasingly evident.

ICMJE conflict of interest statement

The authors declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the research reported.

Funding statement

This study did not receive any specific grant from any funding agency in thepublic, commercial, or not-for-profit sector.