In 2009, the Health Information Technology for Economic and Clinical Health Act (HITECH) promoted national use of electronic health records (EHR) in the US by giving incentives to providers who adopt ‘meaningful use’ of EHRs. As of 2017, nearly 86% of office-based physicians had adopted EHRs. EHRs have rich information including structured data like diagnosis, medication, patient encounters, laboratory tests, semi-structured data such as problem lists, and unstructured data such as patient notes. The large amount of information in EHRs presents abundant opportunities for clinical research but also challenges. EHRs can facilitate clinical research either alone or combined with other data sources including evaluating drug comparative effectiveness, facilitating patient recruitment for clinical trials, assessing gene-disease associations and many others. In this work, we explored insights that researchers can derive from using EHR data, the challenges it presents along the way, particularly, in comparison to traditional registry data and cohort data, and how to overcome these challenges from a methodology point of view. More specifically, in the second chapter, we focused on the application of EHRs on drug comparative effectiveness. In this chapter, we used structured EHR data to assess second-line type 2 diabetes medication comparative effectiveness on renal disorder to provide real-world evidence for clinical trial findings. In the third chapter, we presented data completeness/accuracy related challenges in EHR data specifically in the context of lupus subtype identification; In the fourth chapter, we proposed a natural language processing-based approach to address some of the challenges mentioned by improving the accuracy of disease phenotypes. In the last chapter, we developed a method to improve cardiovascular disease prediction in traditional cohort studies that could potentially be applied to EHR data in the future. Together, this thesis highlights the insights we can derive from EHR data, the challenges EHR data presents when applied in clinical research and offer ways to overcome some of the challenges from a methodology point of view.

Creator

• Deng, Yu

DOI

• https://doi.org/10.21985/n2-17z8-ek22

Subject

• Bioinformatics

Language

• en

Alternate Identifier

• http://dissertations.umi.com/northwestern:15969
• etdadmin_upload_886066

Keyword

• computational phenotyping
• real world evidence data
• machine learning
• biomedical informatics
• deep learning
• electronic health records

Date created

• 2022-01-01

Resource type

• Dissertation

Rights statement

• In Copyright