Developing Recommended Practices For Analyzing EHR Databases

November 03, 2021 - As electronic health records continue to grow in healthcare, scientists must have a uniform set of recommendations to conduct research and analyze data.

In a collaborative effort, the University of California, Irvine School of Pharmacy & Pharmaceutical Sciences researchers looked to conduct a study using the All of Us research database provided by the National Institutes of Health to characterize adverse drug events.

Once characterized, researchers wanted to create a clinical prediction model using machine learning to see if they could identify patients at higher risk of experiencing adverse drug events on chronic medications.

However, as the research team began to develop their study proposal, they discovered significant challenges.

“We recognized that there are a lot of limitations with doing these types of studies with large scale electronic health records databases. We had a need for figuring out how to define an adverse drug event and identify those cases in these large databases, and how to define a drug exposure in these types of databases,” Christine Cadiz, PharmD, health sciences assistant clinical professor and corresponding author told HealthITAnalytics.

To expand their understanding and define terms, the team got a group of undergraduate students together and submitted a proposal to conduct a systematic review to determine current recommendations and practices for adverse drug events using large electronic health record (EHR) databases.

According to the lead author and student in UCI’s PhD in pharmacological sciences program, Quinton Ng, there is a lot to consider when using EHRs to conduct research. An important factor to examine is the concept of defragmentation in healthcare. Ng explained that a patient could go to multiple healthcare institutions for treatment, making data management and organization difficult.

However, through UCI’s systematic review, researchers are working to create conformity in EHRs to easily analyze digital records for adverse drug events.

Undergraduate researchers reviewed thousands of articles and records to standardize measures and definitions for adverse drug events.

“We developed inclusion, exclusion criteria for the articles. We trained the students to look through the title and abstracts to see if the article met the inclusion criteria for the system to be included in the systematic review. They were the ones that did the initial screening and then double-checked by Quinton and the rest of the team,” Cadiz continued.

“Then they also did the full-text screening where they read the articles. If some were flagged, then the other members of the team would check on those.”

The data was then extracted and recorded.

“After the data is extracted, I step in to analyze the data, just grouping them into different terms, different keywords and trying to change the groups here and there,” Ng said.

According to Ng, this study will significantly improve research and guide future EHRs and adverse drug events analyses.

“Even though EHR data is commonly used in practices, there are always new researchers coming in to use the data. This is a very straightforward and easy-to-understand approach,” Ng stated.

Cadiz added, “With the availability of large-scale databases expanding. It’s a good opportunity for researchers to find answers to some important research questions that they may have, but there are so many limitations with using EHR data. This might be a way to encourage more people to utilize EHR data, to look at potential adverse drug events, but have a better starting point.”

According to Cadiz, the research team is now collecting descriptive data on adverse drug events for several chronic disease state medications. The researchers are also studying antidepressant drugs to develop clinical prediction models that would incorporate risk factors for bleeding associated with selective serotonin reuptake inhibitors (SSRIs).