Knowledge-Enhanced Prognostic Prediction Models for Depressive Disorders and Suicidal Behavior

Abstract

Prognostic prediction models (PPMs) estimate a patient’s risk of disease development based on various predictors. Predictors may include patient demographics (such as age and sex), family history, lifestyle factors (such as smoking status or physical activity level), prior medical conditions, laboratory test results, radiographic imaging, or genetic markers. In turn, health care practitioners and patients can make decisions informed by disease risk. For example, a patient found to be at high risk of lung cancer may be advised by their health care provider to quit smoking. The development of PPMs can be summarized in three phases: regression-based models and rule-based models; neural network-based models, and knowledge-enhanced deep learning models. Extensive models ranging from rule based and regression based scoring systems to advanced deep learning models have been proposed and validated in previous studies. In this thesis, we first introduced the general concept of PPMs, then presented one application of PPMs and three newly developed PPMs to demonstrate the designation, implementation, and validation of different types of PPMs in case of the prediction of depressive disorders and suicidal behavior. Specifically, we a) applied a multitask deep learning model to predict the development of depressive disorders in the elderly. The proposed multitask deep learning model exhibits superior performance as compared with baseline regression-based models; b) proposed a novel knowledge-aware deep learning framework for the prediction of suicidal attempts using patient’s historical medical records; c) proposed a novel knowledge-enhanced natural language processing framework to detect individuals at crisis risk level of suicide through mining their social text messages; and d) proposed a knowledge-enhanced explainable learning framework for the prediction of attempted suicidal behavior, and the interpretation of the predicted results. These newly developed models contributed to the prognostic prediction task from different angles and provided better decision support for clinicians and health care professionals.

Date of Award	27 Aug 2020
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Qingpeng ZHANG (Supervisor)