Dr. Randi E. Foraker, Dr. Philip R.O. Payne, and their colleagues at Washington University in St. Louis published the results of their analysis confirming the validity of synthetic data in clinical research. In their pivotal analysis, published this week in the journal JAMIA Open, the researchers confirmed the robustness of MDClone’s Synthetic Data Engine by producing synthetic data sets for three disparate research scenarios and then statistically comparing the resulting synthetic data with the original patient data.
They found in all three cases, the results of the analyses were sufficiently similar between the synthetic derivative and the real data to draw the same conclusions, demonstrating that MDClone and the synthetic data it creates empower researchers to glean scientifically valid results, quickly and with no compromise of patient privacy. “It is expected that the analysis of synthetic data will accelerate the conduct of data-driven research studies,” the authors wrote. “We anticipate that generating synthetic data derivatives for research will reduce barriers to data sharing, which have traditionally included concerns about data privacy and ownership.” The authors described “significant advantages” the synthetic data process has over data deidentification.
The three clinical research scenarios validated in the study included pediatric trauma (observational data), sepsis prediction (cohort data), and community prevalence of chlamydia (population-level data). The real data and the MDClone-generated synthetic data were compared statistically, graphically, and geospatially, producing statistically similar or identical results in all three scenarios. Further, the synthetic data from the sepsis case were successfully used to train a machine-learning algorithm, suggesting that “synthetic data can be used to construct prediction models for use on real patients,” the authors concluded. This finding was deemed particularly important for rare diseases, were “one site may not have sufficient data to train a predictive model, yet a consortium of institutions contributing synthetic data might have sufficient numbers of patients to make meaningful inferences.”
The paper’s lead author, Dr. Randi Foraker, is Associate Professor at the Institute for Informatics (I2) and Director of the Center for Population Health Informatics at I2, Washington University in St. Louis, School of Medicine. “Data synthesis platforms like MDClone are expected to dramatically enhance the research community’s ability to use clinical data for faster insights and improved data sharing in support of precision healthcare,” she wrote.
“We are thrilled that the team from Washington University used our technology to conduct this landmark study,” said Dr. Noa Zamstein, Senior Data Scientist at MDClone and co-author of the paper. “Their findings will allow other healthcare organizations around the world to have confidence in the validity of synthetic data for clinical research and analytics, opening new avenues for collaboration and discovery to the benefit of patients everywhere.”
Founded in 2016 in Israel, MDClone works with major health systems, payers and life science companies in the U.S., Canada and Israel.
Foraker RE, Yu SC, Gupta A, Michaelson AP, Pineda Soto JA, Colvin R, Loh F, Kollef MH, Maddox T, Evanoff B, Dror H, Zamstein N, Lai AM, Payne PRO. Spot the difference: comparing results of analyses from real patient data and synthetic derivatives. JAMIA Open.
Published online Dec. 14, 2020. doi:10.1093/jamiaopen/ooaa060
