Uncertainty, probability and causality in hepatology
TL;DR
Bayesian networks are discussed as a framework that unites prognostic modelling with causal decision making, supporting interpretable, adaptive and data-informed care and can move beyond prediction toward decision-relevant understanding.
Abstract
Clinical decision-making in medicine and, by extension, hepatology is shaped by uncertainty. Guidelines and diagnostic algorithms provide structure for physicians and clinicians, yet clinical assessment still relies on probabilistic and contextual reasoning at the level of the individual patient. To make this reasoning explicit, we draw on concepts from information theory, statistics and causal inference to understand how uncertainty can be described, quantified and acted upon in hepatology. Using hepatology-relevant examples, we examine clinical reasoning under uncertainty. Shannon entropy is introduced as a formal measure of diagnostic uncertainty, illustrating how uncertainty decreases as additional clinical evidence becomes available. Subsequently, frequentist and Bayesian interpretations of probability are contrasted, highlighting how Bayesian updating makes belief revision transparent at the level of individual patients. Building on these concepts, the ladder of causation is introduced and applied to a hepatology example to demonstrate how probabilistic models can progress from recognising associations to reasoning about interventions and counterfactuals. Finally, Bayesian networks are discussed as a framework that unites prognostic modelling with causal decision making, supporting interpretable, adaptive and data-informed care. By formalising uncertainty and causality within a single probabilistic language, hepatology can move beyond prediction toward decision-relevant understanding.