Causal discovery is the key aspect of science. Inferring causality can be achieved in various ways. Typically, you start with your hypothesis (based on what you know so far) and based on the data you collect, you update your hypothesis. In a nutshell, causality can be inferred via your background knowledge and empirical data. Causal Both Bayesian networks (BN) and Structural equation model (SEM) are graphical models that are able to model causality both from background knowledge and empirical data. SEM heavily relies on your background knowledge and use data to justify your knowledge. On the other hand, BN can combine your background knowledge with a causal model that gives the maximum likelihood based on data. If the relationships turn out to be statistically significant, then expert's knowledge is considered statistically valid and can be used to provide guidelines in practice using the model based on expert's knowledge. Functional changes of the bladder are commonly seen in patients with benign prostatic hyperplasia (BPH). We investigated the predictive factors for the alteration in the bladder storage function in patients with BPH using Bayesian networks (BN) and Structure Equation Model (SEM). We analyzed database of consecutive 1,352 patients with BPH who underwent urodynamic studies from Oct 2004 to Oct 2011 in a single institution. We show that combining BN and SEM enables us to build a data driven prediction model with latent constructs. The BN showed that (1) predicting outcome variables, when TZVol is known TPVol and PSA do not add value in prediction; (2) if we know StoragePhaseDetrusor, then all input variables do not help more in predicting Bladder Compliace; (3) when BladderSensation is known, BOOI plays important role in predicting BladderCapacity and StoragePhaseDetrusor; (4) if we know TPVol then TZVo and PSA are independent. User Self Reported Condition and Eurodynamic Study Results did not receive significant latent score and all the variables are discrete, BN was a natural pick to model latent constructs. Volume Capability of Patient's Bladder receive significant latent score and all the variables are continuous, SEM was used to model latent. The combined data driven model reveals that bladder outlet obstruction (BOO) increases the risk of secondary bladder storage dysfunction in patients with BPH. This suggests that more aggressive treatment for BOO might be recommended.
[1]
David Maxwell Chickering,et al.
Learning Bayesian Networks: The Combination of Knowledge and Statistical Data
,
1994,
Machine Learning.
[2]
P. Spirtes,et al.
Causation, prediction, and search
,
1993
.
[3]
Donald Geman,et al.
Stochastic Relaxation, Gibbs Distributions, and the Bayesian Restoration of Images
,
1984,
IEEE Transactions on Pattern Analysis and Machine Intelligence.
[4]
Sumeet Gupta,et al.
Linking structural equation modeling to Bayesian networks: Decision support for customer retention in virtual communities
,
2008,
Eur. J. Oper. Res..
[5]
Nir Friedman,et al.
Learning Belief Networks in the Presence of Missing Values and Hidden Variables
,
1997,
ICML.
[6]
Gregory F. Cooper,et al.
A Bayesian method for the induction of probabilistic networks from data
,
1992,
Machine Learning.