Causal probabilistic networks supported in disease diagnosis and therapy based on dependencies of variables and their statistical relationships. The application of clinical causal probability networks has been evolving over the last few decades and has produced promising results in real-time medical decision support for individual patient’s data analysis that even considers different geo-locations or demographically specific conditions.
A probabilistic causal network is basically a statistical model that is built of directed acyclic graph structures (nodes) that are connected through relationships (associations). The strength of each of the relationships is defined through conditional probabilities.
These methods are used to improve diagnosis and other outcomes of interest by computing an outcome and comparing it against the given standard of validity. The results of a probabilistic causal network analysis show the likelihood of a predicted outcome for an individual patient as “percentage correct” or “accuracy”.
Probabilistic network models were used for predicting heart failure using a causal model of cardiovascular hemodynamics (2003) and for modeling cerebral activity (2008).
Another example, where causal probabilistic networks are used, is to determine the course of severe bacterial infections and to increase the efficiency of antibiotic treatment. The model considers the site of infection, the groups of patients and the definite prevalence of infection caused by a given pathogen (2000). Such support system modeling is supported by a recent publication to determine the course of classical swine fever (2010).
A probabilistic causal network for predicting gene expressions that are critical for rheumatoid arthritis was established by considering comprehensive genotyping, whole blood gene expression profiles and clinical measures of the arthritis disease activity (2011).
Finally, an ICU clinical decision support system was developed and implemented as a causal probabilistic network which involved building a physiological model of relevant cardiovascular, respiratory and associated processes. This real-time model integrates the uncertainty of different physiological measurements and predicts parameters like respiratory rate as a probability distribution. The system applies penalty scores to variables indicating the efficacy of a specific therapy in relation to the clinical objectives. This model allows cumulative probability weighted risk for each ICU patient to be calculated and an individualized therapy suggested that minimizes the risks of adverse events and unexpected therapy course (2003).
Probabilistic causal network models that integrate molecular, genetic and clinical data from individuals represent an innovative approach helping to identify disease courses and therapy.