Emphasizing probabilistic reasoning education: Helping nephrology trainees to cope with uncertainty in the era of AI-assisted clinical practice
Journal
Nephrology (Carlton, Vic.)
Journal Volume
29
Journal Issue
3
Date Issued
2024-03
Author(s)
Abstract
Probabilistic reasoning refers to the construction of the likelihood of conclusions based on one's belief. Contrary to deductive reasoning, which produces either true or false output, probabilistic reasoning requires retrieving prior knowledge from memory and has distinct neurocognitive process.1 Clinical reasoning previously depended on the hypothetico-deductive approach for deriving diagnosis or result interpretation, but uncertainty surrounding clinical scenarios, especially in nephrology ones, may necessitate a probabilistic approach to circumvent errors. Historically, nephrology's development closely intertwines with technological advancements and computer-aided modelling. The interpretation of laboratory data and dialysis prescription heavily rely on process automation and algorithms, fostering a preference for numeric accuracy among nephrologists while instilling apprehension toward clinical ambiguity.3 Artificial intelligence (AI) has transformed medical practice in the contemporary era. A recent article nicely summarizes the utility of AI in dialysis management.2 Emerging studies also examined the applicability of clinical decision support (CDS) systems in aiding drug dosing, acute kidney injury management, allograft rejection prediction, and quality matrix monitoring.3 Despite the gross accuracy observed by researchers, applying CDS output to individual patients often necessitates probability interpretation and a certain degree of ambiguity tolerance. In the forthcoming AI era, nephrologists will be required to make decisions based on probabilities provided by data-driven CDS systems. However, nephrologists often hesitate to communicate prognostic uncertainty to end-stage kidney disease (ESKD) patients. Moreover, maladaptive responses to clinical uncertainty can detrimentally impact the physician-patient relationship and compromise care quality. The complexities and prognostic uncertainties further cause frustrations and contribute to declining interest in nephrology and burnout. Nephrology trainees, lacking a comprehensive background knowledge and emotional preparation, will confront heightened uncertainty in this evolving landscape. The educational gap about uncertainty-coping strategies becomes a lurking concern. To mitigate this challenge, we can reconcile the inherent features of the discipline (difficulty in managing uncertainty) with the inevitable trajectory of AI-assisted clinical practice. We aim to integrate probabilistic reasoning into nephrology training, achieved through approaches such as case-based learning. We propose specific strategies to enhance nephrology trainees' ability to navigate uncertainty (Figure 1). First, we should place emphasis on precisely introducing probabilistic information in undergraduate and postgraduate education. Probability and uncertainty are intrinsic elements of differential diagnosis and clinical decision-making for nephrology patients. The real-time provision of CDS produces probabilities intensifies decision-making urgency and anxiety. Trainees can enhance their probabilistic skills by engaging in repeated exposure to case presentation, including enumeration of indices like pre-test probability, sensitivity/specificity, or predicted risk. Moreover, engaging in peer discussions about uncertainty in risk interpretation can be beneficial. Collaborative efforts can alleviate anxiety, enhance well-being, and reduce uncertainty. Second, many nephrology algorithms are based on assumptions and possess inherent limitations. AI-assisted CDS algorithms are no exception, having their limitations, optimal usage settings, and requiring consistent data input for retraining. Therefore, the focus of training should be on probabilistic reasoning considering both limitations and the applicability of CDS-generated risk stratification and algorithms. Lastly, nephrology trainees should engage in discussions for the best way to refine the algorithm applicability. An example of medical education and training recommendations for probabilistic reasoning can be found elsewhere.4 AI-enabled CDS systems can assist in workflow improvement and optimizing management efficiency in nephrology. The value of CDS to streamline clinical practice is highly augmented by AI technologies. However, how to correctly interpret CDS output by users significantly affects its tremendous clinical potential.5 Without appropriate visualization of results and understanding of the context upon which AI-enabled CDS is built, such system can increase provider dissatisfaction and even resistance to implementation.5 Coping with uncertainty based on probabilistic reasoning can and should be the first step for fully realizing CDS system's potential. In summary, the practice of nephrology is poised to enter a new era with AI assistance, coinciding with escalating clinical uncertainty. We outline three components to enhance trainees' ability of coping with uncertainty and to cultivate their reasoning potential, including the enhancement of probabilistic information understanding, probabilistic reasoning/algorithm training, and case-based practice of applicability. We believe that nephrology education can take proactive steps by highlighting the importance of probabilistic reasoning to aid trainees in effectively grappling with this conundrum. Study design: Chia-Ter Chao. Data analysis: Chia-Ter Chao, Kuan-Yu Hung. Article drafting: Chia-Ter Chao, Kuan-Yu Hung. All authors approved the final version of the manuscript. We are grateful to Ms. Ting-Yu Chen for her kind assistance. Part of the figure content was generated using Microsoft Bing software. The study is financially sponsored by National Taiwan University Hospital (112-N0031 and 112-UN0060) and National Science and Technology Council, Taiwan (NSTC 112-2314-B-002-232-MY3). The authors have no relevant financial or non-financial competing interests to declare in relation to this manuscript. This study did not generate new data or materials.
SDGs
Type
letter