Rapid Learning Health Systems: Speed Is Good, With Caution

NPC’s Chief Science Officer, Dr. Robert Dubois, and Merck & Co.’s Executive Director, Dr. Newell McElwee, recently authored a column in the Journal of Comparative Effectiveness Research. The article, “Enthusiasm for Rapid Learning Health Systems Exceeds the Current Standards for Conducting It,” explores the potential of rapid learning (RL) health systems to leverage the growing data infrastructure to generate comparative effectiveness research (CER) evidence, and the challenges of putting RL into practice.

Citing the backdrop of concerns from critics that CER will move too slowly and prove too costly to improve health care delivery, Dubois and McElwee discuss how RL could allow researchers to answer CER questions more quickly and cost effectively by analyzing data within existing electronic health records (EHRs). They estimate that, due to government efforts to foster EHR adoption, almost every U.S. health care organization will have EHRs in place within a few years, setting the stage for RL that provides greater capacity to generate CER evidence than traditional CER approaches.

The process for RL involves collecting and analyzing data, putting findings into practice, gauging impact, developing a new hypothesis and the re-starting the process. This real-time, iterative approach has already been used in some settings. Several cases of RL in action are garnering notice and discussed in the article. 

However, Dubois and McElwee sound a few notes of caution. Enthusiasm for generating evidence with RL could exceed the scientific rigor with which the research is conducted. If traditional research processes are not applied, including careful study design and subsequent peer review, rapid learning health systems could introduce negative changes into clinical practice and coverage decisions. In particular, because many health organizations are building large data repositories, these CER studies could be conducted within organizations and applied to decision-making without any external scrutiny.

To maximize the potential for positive impact, Dubois and McElwee recommend a starting point for discussion, including steps such as developing consensus standards for RL, implementing an independent study review process that would emulate peer review, evaluating the real (rather than hypothesized) effects of all changes, and sharing analyses publicly to promote transparency and broader system-wide improvements.