Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

Achieving Personalized Medicine Using Machine Learning: Clinical Data Mining Studies on Coronary Heart Disease, Substance Use Disorder, and Alzheimer’s Disease

Hu, Ziheng (2019) Achieving Personalized Medicine Using Machine Learning: Clinical Data Mining Studies on Coronary Heart Disease, Substance Use Disorder, and Alzheimer’s Disease. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

[img] PDF
Restricted to University of Pittsburgh users only until 29 August 2024.

Download (3MB) | Request a Copy


A prerequisite for personalized medicine is to identify patient characteristics that alter treatment response or the disease risks, and to make personalized predictions. However, variable selection using conventional statistical methods is difficult when large number of variables and high-order curvilinear features exist. Machine learning (ML) methods provide a powerful approach for data-driven hypothesis generation as well as fitting complex predictive models. Therefore, this dissertation hypothesizes that clinical data mining using ML methods can inform the individualized selection of interventions. Here we present three studies in which the application of ML has informed the selection of interventions. In the first case, we compared the effect of two anti-platelet drugs – ticagrelor vs. clopidogrel – in Chinese coronary heart disease patients who have undergone percutaneous coronary intervention. Although there was no overall difference between the efficacies of these two drugs, further analysis using the Chi-Square Automatic Interaction Detection decision tree algorithm discovered that ticagrelor might have better efficacy than clopidogrel in single-vessel disease patients. This was confirmed using stratified survival analysis. In the second study, we sought to predict adolescent substance use severity (SUS) using childhood psychological traits. We first derived a SUS scale as a prodrome of substance use disorder based on the harm of consumed substances, and clustered the subjects in to high and low severity groups based on their SUS trajectory. Using the random forest algorithm, we identified thirty childhood psychological traits that predicted SUS trajectory during adolescence, and showed that ML prediction models based on these traits could identify high-risk subjects for selective intervention programs before substance use initiation. In the third study, we searched for the optimal treatment combination for probable Alzheimer’s disease (AD) patients with comorbid hypertension. Using mixed-effect regression models, we identified a combination of anti-hypertensive (aHTN) medications that was associated with the slowest rate of cognitive decline compared to other aHTN therapies. In addition, we found that the protective effect of aHTN medications against cognitive decline was only significant in patients who were also using cholinesterase inhibitors (ChEIs), suggesting a potential synergistic effect. The molecular mechanism underlying these clinical findings were studied using computational systems pharmacology methods. In the fourth study, we investigated the long-term cognitive benefits of ChEIs and memantine in probable AD patients. Using the Really Fast Causal Inference (RFCI) algorithm, we obtained a causal diagram depicting the relationships among baseline characteristics, treatment and the 2-year cognitive outcome. We then performed do-calculus causal inference to estimate treatment effects on individual patients. Finally, we developed a clinical decision support system for treatment selection. Together, these cases demonstrate the utility of ML methods for promoting personalized medicine in neuropsychiatric and cardiovascular disorders.


Social Networking:
Share |


Item Type: University of Pittsburgh ETD
Status: Unpublished
CreatorsEmailPitt UsernameORCID
Hu, Zihengzih2@pitt.eduzih2
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Thesis AdvisorXie,
Committee MemberCooper, Gregory
Committee MemberLopez, Oscar
Committee MemberWang,
Committee MemberWang,
Date: 29 August 2019
Date Type: Publication
Defense Date: 29 July 2019
Approval Date: 29 August 2019
Submission Date: 28 August 2019
Access Restriction: 5 year -- Restrict access to University of Pittsburgh for a period of 5 years.
Number of Pages: 217
Institution: University of Pittsburgh
Schools and Programs: School of Pharmacy > Pharmaceutical Sciences
Degree: PhD - Doctor of Philosophy
Thesis Type: Doctoral Dissertation
Refereed: Yes
Uncontrolled Keywords: personalized medicine, machine learning, coronary heart disease, substance use disorder, Alzheimer’s disease
Date Deposited: 29 Aug 2019 15:22
Last Modified: 29 Aug 2019 15:22


Monthly Views for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item