Date of Award

August 2015

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Department

Management Science

First Advisor

Hemant Jain

Second Advisor

Atish Sinha

Committee Members

Huimin Zhao, Rashmi Prasad, Carmelo Gaudioso

Keywords

Clinicla Trail, Natural Language Processing, Text Mining

Abstract

Patient recruitment and enrollment are critical factors for a successful clinical trial; however, recruitment tends to be the most common problem in most clinical trials. The success of a clinical trial depends on efficiently recruiting suitable patients to conduct the trial. Every clinical trial research has a protocol, which describes what will be done in the study and how it will be conducted. Also, the protocol ensures the safety of the trial subjects and the integrity of the data collected. The eligibility criteria section of clinical trial protocols is important because it specifies the necessary conditions that participants have to satisfy.

Since clinical trial eligibility criteria are usually written in free text form, they are not computer interpretable. To automate the analysis of the eligibility criteria, it is therefore necessary to transform those criteria into a computer-interpretable format. Unstructured format of eligibility criteria additionally create search efficiency issues. Thus, searching and selecting appropriate clinical trials for a patient from relatively large number of available trials is a complex task.

A few attempts have been made to automate the matching process between patients and clinical trials. However, those attempts have not fully integrated the entire matching process and have not exploited the state-of-the-art Natural Language Processing (NLP) techniques that may improve the matching performance. Given the importance of patient recruitment in clinical trial research, the objective of this research is to automate the matching process using NLP and text mining techniques and, thereby, improve the efficiency and effectiveness of the recruitment process.

This dissertation research, which comprises three essays, investigates the issues of clinical trial subject recruitment using state-of-the-art NLP and text mining techniques.

Essay 1: Building a Domain-Specific Lexicon for Clinical Trial Subject Eligibility Analysis

Essay 2: Clustering Clinical Trials Using Semantic-Based Feature Expansion

Essay 3: An Automatic Matching Process of Clinical Trial Subject Recruitment

In essay1, I develop a domain-specific lexicon for n-gram Named Entity Recognition (NER) in the breast cancer domain. The domain-specific dictionary is used for selection and reduction of n-gram features in clustering in eassy2. The domain-specific dictionary was evaluated by comparing it with Systematized Nomenclature of Medicine--Clinical Terms (SNOMED CT). The results showed that it add significant number of new terms which is very useful in effective natural language processing In essay 2, I explore the clustering of similar clinical trials using the domain-specific lexicon and term expansion using synonym from the Unified Medical Language System (UMLS). I generate word n-gram features and modify the features with the domain-specific dictionary matching process. In order to resolve semantic ambiguity, a semantic-based feature expansion technique using UMLS is applied. A hierarchical agglomerative clustering algorithm is used to generate clinical trial clusters. The focus is on summarization of clinical trial information in order to enhance trial search efficiency. Finally, in essay 3, I investigate an automatic matching process of clinical trial clusters and patient medical records. The patient records collected from a prior study were used to test our approach. The patient records were pre-processed by tokenization and lemmatization. The pre-processed patient information were then further enhanced by matching with breast cancer custom dictionary described in essay 1 and semantic feature expansion using UMLS Metathesaurus. Finally, I matched the patient record with clinical trial clusters to select the best matched cluster(s) and then with trials within the clusters. The matching results were evaluated by internal expert as well as external medical expert.

Download

Share

COinS