March 12, 2020 – Experts on data, including three DCRI faculty, recently published a report on using patient-reported health data in pragmatic trials that stemmed from an NIH Collaboratory roundtable.
A report from an NIH Health Care Systems Research Collaboratory roundtable recently published in Journal of the American Medical Informatics Association discusses the use of patient-reported health data in pragmatic clinical trials. Since 2012, the DCRI has served as one of the coordinating centers for the NIH Collaboratory, which supports the design and conduct of pragmatic clinical trials embedded in health care systems. DCRI faculty Frank Rockhold, PhD; Keith Marsolo, PhD; and Emily O’Brien, PhD, contributed to the report.
Because pragmatic clinical trials are embedded in health care systems and mirror routine care as closely as possible, these studies often leverage electronic health records (EHRs) and claims data. However, special considerations need to accompany use of these types of data. EHRs often have missing data or data that is in the wrong format for research. In these instances, patient-reported health (PRH) data can sometimes be used to fill in gaps where data is incomplete, especially to define study endpoints. The report draws on the experience of the NIH Collaboratory to detail considerations for using PRH data in these types of trials. As a case study, the report examines ADAPTABLE, a DCRI-led pragmatic trial. Outcomes are drawn primarily from the EHR, but this information is supported by an online portal that collects PRH data.
Although PRH data can be helpful in increasing the completeness of data in certain scenarios, such as when EHR data is not collected or is collected unreliably, teams working on pragmatic trials should understand that PRH data comes with its own limitations. In some scenarios, PRH data can be verified with hospital bills or clinician notes, but in other cases, the PRH data is all that is available. The most reliable method, the paper’s authors note, will often involve considering both kinds of data. “A combination of EHR data and the patient’s memory may represent the best triangulation of truth for questions regarding whether the patient has ever had a given diagnosis or procedure,” the authors write.
In some scenarios, events reported as part of PRH data may be in conflict with EHR data. Further studies are needed to determine how integrating PRH data into a pragmatic trial affects the performance of event classification algorithms.
Other Duke faculty contributors include Jessica Tenenbaum, PhD, and Rachel Richesson, PhD, MPH.