Academic

PhD Oral Presentation

Statistical learning methods are widely used in medical literature for the purpose of diagnosis or risk prediction. Conventionally, the accuracy of these models is assessed using metrics of diagnostic performances such as sensitivity, specificity, and ROC curves which fail to account for clinical utility of a specific model. Decision curve analysis (DCA) is a novel complement to address the utility. Using DCA, a clinical judgment of the relative value of benefits (treating a true positive case) and harms (treating a false positive case) associated with prediction models is made by using a decision analytic measure called net benefit. The preferences of patients or policy-makers are accounted by using the threshold probability. In this study, estimation procedure and inference methodology are proposed for DCA. We illustrate our proposals with simulation study and the real datasets. Medical diagnostics with modern machine learning methods reduce human efforts and improve our understanding of disease propagation. When the data is unstructured, shallow learning methods may not be feasible. Deep learning neural networks like multilayer perceptron (MLP) and convolutional neural network (CNN), have been incorporated in medical diagnosis and prognosis for better health care practice. For a binary outcome, these learning methods output predicted probabilities for patient’s health condition. Investigators need to consider appropriate decision threshold to split the predicted probabilities into positive and negative regions. We review shallow and deep learning methods to select the optimal decision threshold based on optimization of the ROC curve criteria and also the utility-based method – DCA. This study will help medical decision makers to understand the diagnostic procedures with various learning methods, assess them with appropriate decision threshold and derive the inference.