Principles Of developing predictive models in medicine and Their application in liver transplantation

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Prognostic models have become an integral part of clinical practice in liver transplantation (LT), supporting decision-making throughout the patient’s clinical pathway – from assessment of the severity of underlying disease to postoperative monitoring of graft function. This review aims to systematize the principles of development, validation, and quality assessment of clinical prognostic models, with a particular focus on the scales used in LT. The review is based on an analysis of key publications on the methodology of prognostic modeling, international quality assessment standards (PROGRESS, TRIPOD+AI, and CHARMS), as well as original studies describing prognostic scoring systems in hepatology and transplant medicine. The main statistical approaches, including logistic regression, Cox proportional hazards regression, and machine learning algorithms, are discussed. A multilevel framework of prognostic models used in LT is presented, including preoperative tools for assessing recipient disease severity (CTP, MELD, MELD 3.0, and PELD), donor-related risk (DRI and BAR score), and postoperative criteria (EAD, SOFT, L-GrAFT, MEAF, and EASE). Of the entire array of scales, only three are applicable to pediatric practice: PELD, Pedi-SOFT, and ASPELT. The shortage of specialized pediatric prognostic tools persists and underscores the need for targeted multicenter studies.