Automated eligibility screening using natural language processing and machine learning vastly improved the efficiency of selecting potential patients for clinical trials in a study from Cincinnati Children's Hospital Medical Center.
In the study, published at the Journal of the American Medical Informatics Association, researchers developed the automated system to identify potential patients for 13 disease-specific clinical trials in hopes of finding a more effective method than the time-consuming manual process of the past.
They used data fields including demographics, laboratory data and clinical notes from the electronic health records of 202,795 patients who had visited the hospital's emergency department.
The researchers wanted to determine the effectiveness of natural language processing (NLP), information extraction (IE), and machine learning (ML) techniques for identifying candidates based on real-world clinical data and trials.
They compared the system with decisions made by two board-certified, pediatric emergency medicine physicians doing chart review and a reference standard of historical trial--patient enrollment decisions on a diverse set of clinical trials.
They found the clinical notes in the unstructured data fields particularly important in identifying likely trial participants, with the NLP capabilities building on their previous research.
The automated system reduced the workload by 92 percent with a mean average precision of 62.9 percent. Overall, it increased efficiency by 450 percent.
A study from practices in England and Scotland also found that EHRs can be a boon to identifying likely participants in clinical trials. As trials become more complex, however, interoperability will be more important than ever.
Keith Marsolo, of the biomedical informatics division of Cincinnati Children's Hospital Medical Center, previously called for the redesign of EHR systems to encourage data-sharing between healthcare research and clinical practices. He described the organization's many frustrations with having to develop workarounds to the EHR system's limitations.
To learn more:
- read the JAMIA study