A needed (r)evolution to improve colorectal cancer screening: binding patient-centered/shared decision-making medicine with large language models in an emerging knowledge society

Performing procedures in healthy but high risk population is a major challenge for traditional cancer screening methods [1]. One of the most difficult procedures to accept is colonoscopy, which is nevertheless the best way of identifying and eventually removing early colorectal cancer (CRC) lesions. The lack of clarity about why it should be performed is however very worrying, as even first-degree relatives of patients with CRC have reservations about giving their consent (only 15.4% have had a colonoscopy in the previous 5 years) [2]. These types of reservations, which would be greater in patients with no family history of CRC, could result in an inadequate procedure that should be rescheduled or even cancelled with all the associated underlying costs [3]. In addition, the increase in CRC among younger patients requires new, more effective communication strategies. Although phone calls, text messages, and social networks, have been used successfully, the use of large language models (LLMs) has not been adequately evaluated in CRC screening in open populations and health professionals.

“Artificial intelligence systems must learn from the expertise of colonoscopists, and colonoscopists must know the fundamentals of LLMs and how they generate their responses…”

In this scenario, Maida et al. report in this volume of Endoscopy an evaluation of the answers of a LLM (Chat Generative Pretrained Transformer [ChatGPT]) as a possible tool to increase patient awareness of the early detection of CRC [4]. The AI-CORE (Artificial Intelligence COlorectal cancer REsearch) Working Group reported the assessment of fifteen questions posed to ChatGPT about general screening, and endoscopic and therapeutic measures for CRC. The authors then asked 40 physicians (20 experts and 20 nonexperts) to evaluate the accuracy, completeness, and comprehensibility of the answers generated by ChatGPT and 100 consecutive patients to evaluate the completeness, comprehensibility, and trustworthiness of the ChatGPT answers. They found satisfactory performance with high reliability, concluding that this tool has the potential to improve the detection of CRC and should be improved with further scientific evidence and clinical guidelines. Although with minor limitations, the most important advance of this study is the development of objective evaluation tools on the information obtained from ChatGPT responses.

Technological advances are leading us into an ever-increasing spiral that has been intensified by the COVID-19 pandemic. We are facing a paradigm shift that also involves a global change in thinking, which is taking shape as the global knowledge society [5]. Although health-seeking behavior varies according to psychological, social, cultural, geographical, and historical factors, an informed patient is the cornerstone of this new medicine. This type of patients requires not only traditional preventive, curative, or rehabilitative medicine, but also a significant acquisition of the best health information to make decisions with maximum collective responsibility.

We need to change the medical perception that clients/patients should be treated passively to that of being human beings with the same values and needs as our loved ones [6], treating them as we would wish other physicians to treat our loved ones, including the use of LLMs [7]. We cannot expect them to have the ability to find relevant and truthful information in their daily lives, but we can act as educators or guides at the personal and/or community level to provide that knowledge, using a patient-centered model with shared decision-making medicine [8]. The use of LLMs could be a bridge in this crucial step. It is time to evolve towards the mass, and at the same time personalized, dissemination of relevant information to reduce morbidity and mortality from serious diseases such as CRC. To improve interaction with new generations, we need to use LLMs as a technological bridge between patient-centered/shared decision-making medicine and emerging knowledge societies.

Endoscopy, being a relatively new branch of medicine, is very involved in these advances from its foundation to actual practice. We are at a turning point in applying our historic experience and taking the lead to create our own endoscopic global knowledge society through collaboration with expert systems, science communicators, and social networks for the proper use of medical information in the general population and to avoid the incorrect use or abuse of LLMs. Artificial intelligence systems must learn from the expertise of colonoscopists, and colonoscopists must know the fundamentals of LLMs and how they generate their responses ("transformed" type neural networks, for example) because there is a great need to use proper feedback. To achieve this, I suggest realizing the steps proposed in [Fig. 1].

Fig. 1 Proposed cycle to create a broad collaborative network. Technological (big data, deep learning, neural networks, large language models, artificial intelligence, etc.) and medical tools (quality of life assessment, satisfaction with care, etc.) should be used locally and integrated globally.

All of this must be supported by local, national, and international collaborations. In a matter of months, we will generate such a large amount of information (big data) that LLMs will have to consider answering the questions of our patients around the world with our findings. Collaborations with healthcare organizations must optimize LLMs, creating objective guidelines that allow evaluation and subsequent improvements to the responses with the most relevant information for doctors and patients, not only to minimize risk and increase the awareness of patients, but also to enhance engagement and the overall physician-patient experience.

Publikationsverlauf

Artikel online veröffentlicht:
07. November 2024

• References

• 1 Wang X, Li Y-J, Zhou X-Y. et al. Evaluation indicators for access to cancer screening services: a scoping review. Public Health 2024; 235: 194-201 DOI: 10.1016/j.puhe.2024.07.003. (PMID: 39153383)
• 2 Bai Y, Wong CL, He X. et al. Effectiveness of tailored communication intervention in increasing colonoscopy screening rates amongst first-degree relatives of individuals with colorectal cancer: A systematic review and meta-analysis. Int J Nurs Stud 2020; 101: 103397
• 3 Rehman S, Al-Alawi KS, Chan MF. et al. Factors affect knowledge, attitudes, and practices in colorectal cancer screening: A systematic review. J Eval Clin Pract 2024; 2024 DOI: 10.1111/JEP.14086. (PMID: 38993013)
• 4 Maida M, Ramai D, Mori Y. et al. The role of generative language systems in increasing patient awareness of colon cancer screening. Endoscopy 2024; 2024 DOI: 10.1055/A-2388-6084. (PMID: 39142348)
• 5 Fedoroff NV. The global knowledge society. Science 2012; 335: 503-503 DOI: 10.1126/science.1219498. (PMID: 22301283)
• 6 Camacho S. The Incorporating of 3-Dimensional Skills on Surgery Education Would Improve Curricula. J Invest Surg 2021; 34: 205-207 DOI: 10.1080/08941939.2019.1597953.
• 7 Yu K-H, Healey E, Leong T-Y. et al. Medical artificial intelligence and human values. NEJM 2024; 390: 1895-1904 DOI: 10.1056/NEJMra2214183. (PMID: 38810186)
• 8 Marion S, Ghazal L, Roth T. et al. Prioritizing patient-centered care in a world of increasingly advanced technologies and disconnected care. Semin Radiat Oncol 2024; 34: 452-462 DOI: 10.1016/j.semradonc.2024.07.001. (PMID: 39271280)