Author ORCID Identifier

https://orcid.org/0000-0002-9869-1836

Date Available

10-21-2021

Year of Publication

2021

Degree Name

Doctor of Philosophy (PhD)

Document Type

Doctoral Dissertation

College

Engineering

Department/School/Program

Computer Science

First Advisor

Dr. Jin Chen

Abstract

Longitudinal healthcare data encompasses all tasks where patients information are collected at multiple follow-up times. Analyzing this data is critical in addressing many real world problems in healthcare such as disease prediction and prevention. In this thesis, technical challenges in analyzing longitudinal administrative claims data are addressed and novel deep learning based models are proposed for multi-stream data analysis and disease prediction tasks. These algorithms and frameworks are assessed mainly on substance use disorders prediction tasks and specifically designed to tackled these disorders. Substance use disorder is a public health crisis costing the US an estimated $740 billion annually in healthcare, lost workplace productivity, and crime. Early identification and engagement of individuals at risk of developing a substance use disorder is a critical unmet need in healthcare which can be achieved by producing automatic artificial intelligence based tools trained using big healthcare data. In fact, healthcare data can be harnessed together with artificial intelligence and machine learning to advance our understanding of factors that increase the propensity for developing different diseases as well as those that aid in the treatment of these disorders.

Here in, a disease prediction framework is first proposed based on recurrent neural networks. This framework includes three components: 1) data pre-processing, 2) disease prediction using long short term memory models, and 3) hypothesis exploration by varying the models and the inputs. This framework is assessed using two use cases: substance use disorder prediction and mild cognitive impairment prediction. Experimental results show that this proposed model can efficiently analyze patients' data and creates efficient disease prediction tools. Second, the limitationsof current deep learning models including long short term memory models in claimsdata analysis are detected and addressed, and a novel model based on the transformer models is proposed. In fact, leveraging the real-world longitudinal claims data, a novel multi-stream transformer model is proposed for predicting opioid use disorder as an important case of substance use disorders. This model is designed to simultaneously analyze multiple types of data streams, such as medications, diagnoses, procedures and demographics, by attending to segments within and across these data streams. The proposed model tested on the IBM MarketScan data showed significantly better performance than the traditional models and recently developed deep learning models.

Digital Object Identifier (DOI)

https://doi.org/10.13023/etd.2021.395

Share

COinS