Predictive modeling can flag potential candidates for clinical trials

Predictive modeling can be effective in assessing the eligibility of patients for clinical trials, according to research published at BMC Medical Informatics and Decision Making.

Translating eligibility criteria from free text in EHRs into decision rules has been a problem in automating the process of identifying good candidates for clinical trials, the authors of the German research write.

They took a different approach: Developing prediction models based on characteristics of existing trial subjects to find new ones. They looked at data such as age, gender, diagnosis and procedure codes for clinical trial participants in setting up the clinical trial recruitment support system (CTRSS). Once an adequate number of acceptable participants has been collected, the system can from there find other potential participants based on the same criteria.

Once they got algorithms working, they set out to find the smallest number required in the "learning" pool for the system to effectively flag other potential participants. Of three statistical models--random forests, decision trees and logistic regression-–random forests were most effective, with the ability to learn from just three percent of patients screened.

The advantage of this approach, they said, is that it does not require the trial's eligibility criteria to be translated into a computable form. That means it can be independent of a specific hospital and can be applied across institutions.

Community hospitals aren't capitalizing on clinical research to generate revenue, recruit and retain physicians or enhance their impact on the community, a recent GuideStar Clinical Trials Management survey found. There were multiple reasons: lack of resources, internal clinical research program awareness and financial support, as well as insufficient staffing.

Better tools and improved access through the cloud, however, are giving scientists and researchers more democratic access to the reams of information the U.S. government collects, federal officials say.

Researchers from Cleveland, meanwhile, have found that natural language processing can be effecting in scrubbing personal health information from records to be used for analysis. It can be inexpensive and faster than humans, they found.

To learn more:
- find the research

Suggested Articles

Nearly 10,000 patients involved in research studies were impacted by a third-party privacy breach that may have exposed their medical diagnoses, test results…

Veterans Health Administration medical facilities currently have a paper medical record backlog that if stacked up would be 5.15 miles high, according to the…

The Department of Health and Human Services announced proposed changes to privacy restrictions on patients' substance use treatment records.