Electronic health records are a great way to identify patients who satisfy predefined criteria for clinical trials, survival analysis research and other uses, but the processes used to cull this information is difficult and needs refining, according to a new study in the Journal of the American Informatics Association.
The researchers, from Ohio State University and elsewhere, noted that patient phenotype information often is buried within EHR data; abbreviations, misspellings and the use of local dialects in clinical notes make identification even harder. The authors reviewed 97 different articles on patient cohort identification using EHRs, and examined the different approaches used for phenotyping, including natural language processing, statistical analyses, rules-based algorithms and hybrid approaches.
They found that the current approaches to phenotyping were "often inadequate" to the task.
"Although phenotyping has developed at a steady pace in the past few years, there is a lot of room for improvement," the researchers warned.
They suggested, among other things, that the data be more standardized using terminologies, and that open-source phenotyping tools be created for users. They also recommended that there be a focus on developing systems that make "holistic use of the EHR" in characterizing a patient for phenotyping.
Other studies have pointed out that while the data in EHRs holds great promise for research and other uses, the information needs to be modified to be more effective.
To learn more:
- read the article