Not just data: A method for improving prediction with knowledge

Abstract

Many medical conditions are only indirectly observed through symptoms and tests. Developing predictive models for such conditions is challenging since they can be thought of as 'latent' variables. They are not present in the data and often get confused with measurements. As a result, building a model that fits data well is not the same as making a prediction that is useful for decision makers. In this paper, we present a methodology for developing Bayesian network (BN) models that predict and reason with latent variables, using a combination of expert knowledge and available data. The method is illustrated by a case study into the prediction of acute traumatic coagulopathy (ATC), a disorder of blood clotting that significantly increases the risk of death following traumatic injuries. There are several measurements for ATC and previous models have predicted one of these measurements instead of the state of ATC itself. Our case study illustrates the advantages of models that distinguish between an underlying latent condition and its measurements, and of a continuing dialogue between the modeller and the domain experts as the model is developed using knowledge as well as data.

Clinical evidence framework for Bayesian networks

“…In the ATC BN, we use measurement idioms multiple times to define parts of the BN structure [48]. In order to distinguish the measurement idioms from other BN fragments, we could add a class called 'MeasurementIdiom' as a subclass of the fragment class (see Fig.…”

“…3.2. In the remainder of this section, we use the complete version of the ATC BN (ATC BN 2015; [48]) as a case study to demonstrate a web page generated from the evidence ontology. We call this web page the evidence browser for the ATC BN [6].…”

“…Our framework is composed of two parts: firstly, an ontology that organises the evidence, and secondly, a browser that presents the evidence. We illustrate the evidence framework using a case study of a clinical BN that we have previously developed [48]. This BN accurately predicts a potentially fatal physiological disorder, called acute traumatic coagulopathy (ATC), in early stages of trauma care.…”

“…The structure of the BN was built based on clinical knowledge and published evidence, and its parameters were defined using a patient dataset. More information on development and validation of the ATC BN is described by [48] and the clinical evidence relevant to the model can be browsed online on the model's website [6]. In this paper, we present the entire technical structure of the evidence framework and make a qualitative validation with its potential users.…”

Risk Assessment and Manage- ment of Violence in Patients with Mental Disorders: A Re- view

“…Previous research has relied on a regression model, only considering the association between variables that potentially influence the likelihood of future violence. As causal analysis for simulating potential interventions may be required, particularly for adequate risk management, adopting a Bayesian network model [52] has been recently attempted [53].…”