Temporal and dynamic Bayesian networks for prognosis and diagnosis in clinical settings: A scoping review

Temporal and dynamic Bayesian networks offer a promising approach for modeling temporal relationships within health data, allowing researchers to analyze how clinical variables evolve over time.

A scoping review was conducted to map the existing literature on the application of Bayesian networks for temporal modeling in a clinical context. A comprehensive search of five electronic databases (PubMed, Web of Science, Scopus, EBSCOhost, and DBLP) was conducted to identify studies utilizing temporal or dynamic Bayesian networks to model temporal relationships within a medical or clinical setting.

Our search yielded 47 studies (0.44 %) from an initial pool of 10631 publications. Prognostic applications dominated the studies (74.5 %, n = 35), while diagnostic applications comprised 25.5 % (n = 12). Publication dates ranged from 2000 to 2023, with a recent surge from 2021 onwards (n = 18). Twelve medical specialties were represented, with oncology (n = 10), intensive care (n = 9), and cardiology (n = 7) being the most frequent. Standard Dynamic Bayesian Network (DBN) models were used in 72.3 % (n = 34) of the studies, with the remaining studies employing other Bayesian network types (27.7 %, n = 13).

DBNs demonstrate potential for temporal modeling of clinical variables across a wide range of medical specialties, with applications in oncology, intensive care, and cardiology. Their use has been predominantly focused on life-threatening conditions, such as cancer and organ failure. However, given their ability to model temporal events, DBNs are well-suited for managing long-term patient care, particularly in chronic diseases. Expanding their application to underrepresented fields such as pneumology, mental health, and geriatrics could further enhance their impact. A surge in publications in recent years reflects a growing interest in the clinical utility for DBNs, reinforcing their role in supporting decision-making in dynamic health contexts.

PROSPERO CRD42023431306.