Letter to the Editor: The application of generative artificial intelligence in liver transplantation weight trajectory

To the editor, We would like to comment on Kim et al.1 This study is unique in that it uses generative artificial intelligence (GenAI) to characterize post-liver transplant body weight pathways, with a focus on identifying latent elements that standard statistical models may overlook. However, the research technique has certain shortcomings. To begin, transforming longitudinal data into 8-dimensional vector cross-sectional data may result in a loss of weight-change dynamics over time, such as short-term fluctuations or clinically relevant turning points. Furthermore, the structure and mechanism of the GenAI employed are unknown, limiting the study’s reproducibility, and the bioclinical interpretation of the discovered latent variables may remain confusing. While the analysis reveals that LF1 and LF2 explain up to 99% of body weight variance, this figure may be due to overfitting rather than actual mechanism explanation, particularly without external validation in patient groups from different institutions. Furthermore, the multi-analysis relating latent factors to clinical variables may include uncontrolled confounders such as immunosuppressant use, physical activity levels, post-transplant diet, or long-term problems, all of which have a major impact on body weight. Upon closer examination, LF1, which represents significant weight fluctuations in the first year, may not be simply a “latent factor” but rather reflects the overall healing process after liver transplantation, including recovery from starvation, metabolic changes, and the effects of steroids. LF2, together with its 2 sub-forms, may suggest disparities in health trajectories between individuals who maintain weight management and those who acquire post-transplant obesity, making them a prime target for early preventive measures. Broadly speaking, this research raises crucial questions about the role of AI in generating clinical “new knowledge”: are the derived latent factors merely mathematical models, or do they genuinely relate to biological and behavioral mechanisms? Future studies should integrate data from medicine, nutrition, and health behaviors to test the causal implications of these latent factors and compare the performance of GenAI with other advanced statistical models, such as mixed-effects models or traditional trajectory clustering. Finally, a key question is: how applicable are the AI ​​results to individual prediction and care design for post-liver transplant patients in daily practice?