Reply: The application of generative artificial intelligence in liver transplantation weight trajectory

To the editor, Weight change following liver transplantation (LT) is a complex phenomenon consisting of metabolic risk, immunosuppression, and lifestyle changes (diet and physical activity) following transplant surgery. The conventional approach, which we think is reductive, is to present summary statistics (mean, median) to evaluate post-LT weight change. Thus, the Generative Artificial Intelligence (GenAI) approach utilized in the present is a powerful analytical technique, enabling a more robust approach to complex datasets and relationships, as noted above. While there are inherent limitations within GenAI, it is a powerful statistical tool that should be incorporated into clinical research to answer questions that were previously more difficult to address. In the manuscript, we provided a detailed description of the methodology to add to the rigor of the study to allow for future interrogation of our findings.1 The statistical approach was supplemented further with weight assessment every 3 months. While theoretically, weight measurements can be done more frequently to identify “short-term fluctuations” or “clinically relevant turning points,” measuring weights more frequently than every 3 months is clinically difficult and not practical in transplant clinics. Moreover, our approach and the presented data demonstrate that several inflection points were identified that can be targeted for interventional therapy to implement a risk-mitigating strategy. The latent factors (LFs; LF1 and LF2) are purely statistical constructs that are used to better understand the changes in weight trajectory after LT. To better understand the biological relevance of the study, we constructed a regression analysis using clinically relevant variables. A key message of these regression analyses is that the majority of the weight change is not explained by the variables that are readily measured in clinical practice, such as immunosuppressants, gender, ethnicity, etc. These findings highlight the importance of other unmeasurable factors, such as metabolic flexibility, impaired fuel utilization, and organ-specific insulin resistance.2 The larger issue is not whether these factors should be measured, but rather whether they can be measured in routine and busy clinical practice without being disruptive. While we agree that factors such as diet and exercise can lead to changes in weight, data in LT recipients demonstrate that these rarely lead to positive changes.3 Thus, a key message of the manuscript is to highlight the importance of additional investigation into these “unmeasurable” factors to better understand the nature of weight change in LT recipients. Finally, we want to re-emphasize that our study highlights the importance of incorporating evolving statistical approaches to better understand complex medical issues in LT recipients with the overarching goal of improving clinical outcomes. The study findings require further validation in larger cohorts, international cohorts, and even other solid organ types, as these types of studies will increase confidence in utilizing powerful technologies. In addition, the advanced statistical tools also need to be integrated with the rapidly evolving multi-omics approach to better understand the biological nature of these statistical signals. We hope that our findings provide the foundation for future investigations into the metabolic health of LT recipients using complex and integrated technology.