Machines Help With Patient Recruitment for Research Studies

After struggling to pull together a complex research study, wouldn’t it be nice to know if people were going to participate? Machines can help, say researchers from Cincinnati Children’s Hospital Medical Center. The work, which appears in the latest Journal of the American Medical Informatics Association, indicates that the automated algorithm developed by the researchers was significantly better at predicting patient participation response than the program that simulates current recruitment practices. The researchers also found that patients are less likely to participate in randomized studies, multi-center trials, more complex trials, and trials that required follow-up visits.

“Challenges with patient recruitment for clinical trials are a major barrier to timely and efficient translational research, ” said Yizhao Ni, Ph.D., lead author and a researcher in the Division of Biomedical Informatics at Cincinnati Children’s, in the April 27, 2016 news release. “The ultimate goal of our research is to impact patient recruitment strategies to increase participation in clinical trials, and to help ensure that studies can be completed and the data are meaningful.”

As indicated in the news release, “To test their algorithm, researchers collected data from 2010 through 2012 involving clinical trial recruitment in the Emergency Department of Cincinnati Children’s—a busy and challenging environment with 70, 000 annual patient visits. For purposes of the current study, the researchers attempted to collect data on the Emergency Department’s process. For scoring, each Emergency Department patient invited into a clinical trial was counted as an ‘encounter.’ Patients accepting invitations were labeled as ‘+1’; decline responses were labeled ‘-1.’ The researchers counted data that included 3, 345 patient encounters for a diverse set of 18 different clinical trials.”

“At the same time, the researchers collected demographic, socioeconomic data and clinical information from different sources to help build patient profiles. This information was fed into the machine learning algorithm, which processed the data through programs for predictive modeling, comparison, analysis and prediction. Researchers then compared the effectiveness of their algorithm to a ‘random-response-prediction program’ that was developed to simulate the current recruiting method in the medical center’s Emergency Department. Those results were then validated by comparing them to the “acceptance” and “decline” responses recorded from the in-person Emergency Department encounters.”

Dr. Ni told OTW, “To conduct patient recruitment for a clinical trial, the researchers will first chart review patients electronic health records (EHRs) to identify eligible patients. They will then approach the eligible patients for enrollment. Usually chart reviewing a patient’s electronic health record takes a significant amount of time (approximately 30% of their time in patient recruitment). So one of the most frustrating situations is that after spending a lot of time in chart review, the researchers just find that the patient is not willing to particulate due to his/her preference (or other unexpected factors). One of the most frequent questions they ask is that, ‘Is there any magic that can predict patients’ willingness before approaching them?’ This is the motivation of our current study.”