Abstract
The purpose of this thesis is the implementation of Bayesian methods for the diagnosis andprediction of asthma progression, as well as the modeling of a complex neurophysiologicalsystem such as the muscle spindle.In the case of childhood asthma, a very important issue is the early detection of the individuals who are at risk of persistence of the disease after the age of five. In medicine, the prediction accuracy is very significant, as an accurate prediction can lead to a better outcome in the future. For asthma persistence prediction in children, a Bayesian logistic regression model was developed, in conjunction with the principal component analysis, in order to deal with the strong correlations between the prognostic factors. An important issue in applying Bayesian logistic regression is the absence of prior knowledge for the distribution of the regression coefficients. For this reason, a non-informative uniform prior and a weakly informative Cauchy prior are used. In this way a tes ...
The purpose of this thesis is the implementation of Bayesian methods for the diagnosis andprediction of asthma progression, as well as the modeling of a complex neurophysiologicalsystem such as the muscle spindle.In the case of childhood asthma, a very important issue is the early detection of the individuals who are at risk of persistence of the disease after the age of five. In medicine, the prediction accuracy is very significant, as an accurate prediction can lead to a better outcome in the future. For asthma persistence prediction in children, a Bayesian logistic regression model was developed, in conjunction with the principal component analysis, in order to deal with the strong correlations between the prognostic factors. An important issue in applying Bayesian logistic regression is the absence of prior knowledge for the distribution of the regression coefficients. For this reason, a non-informative uniform prior and a weakly informative Cauchy prior are used. In this way a test for the robustness of the model is carried out in the case of a prior distribution change. The resulting models from the application of this method predict asthma with high accuracy (> 85%) and also give steadier results between patients who are at risk of asthma persistence and those who are not. In addition they provide important information for the significance of the prognostic factors in different credible intervals. A significant advantage of Bayesian methods is that the posterior distribution which is obtained for estimated parameters can be used, when new data become available, as informative prior distribution in order to derive better results. One such implementation can be carried out for modeling the muscle spindle system. The objective in this case is a more accurate modeling of this system with the use of a two-step approach. This approach is based on Bayesian logistic regression with a sequential application of a weakly informative Cauchy and an informative prior distribution which are applied to a large dataset with more than 15800 observations. The Cauchy prior is used in the first 5000 observations so that a posterior distribution can be derived, which is going to be used as an informative prior to the rest of the data. The parameters of the system are the threshold function, the recovery function, the summation function and the carry-over effect function. The results of this approach are almost identical to the ones obtained by the maximum likelihood method, but this two-step Bayesian approach leads to results with smaller errors and subsequently smaller confidence intervals. This means that the derived model is more representative. Generally, the achievement of the control of the asthma disease is an issue of great importance. As the control of the disease is sustained the medication should be gradually reduced and then stopped. However, after the cessation of medication, there is always a possibility of loss of disease control which will lead to an asthma exacerbation. The implementation of another Bayesian method like Bayesian classifiers for asthma exacerbation prediction in data with repeated measurements consists a very important application. Bayesian classifiers are graphical models with the capability of displaying probabilistic relationships between predicting factors clearly. This is a great advantage in contrast with other classifiers as the identification of risk factors for asthma exacerbations remains a research task which is not completed in the international bibliography. For the implementation several algorithms were applied. The Bayesian classifier with the use of backward sequential elimination and joining algorithm is able to predict if a patient will have a disease exacerbation after his last assessment with high accuracy (> 90%). Moreover, the resulting structure and conditional probability tables of the classifier give a clear view of the factors contributing to asthma exacerbation, which can help clinical doctors a lot. These factors are found to be important in other research works in international bibliography as well.
show more
