Fusing Randomized Trials With Big Data – The Key to Self-learning Health Care Systems?

Randomized clinical trials (RCTs) have revolutionized medicine by providing evidence on the efficacy and safety of drugs, devices, and procedures. Today, more than 40 000 RCTs are reported annually, their quality continues to increase, and oversight mechanisms ensure adequate protection of participants. However, RCTs have at least 4 related problems: (1) they are too expensive and difficult; (2) their findings are too broad (average treatment effect not representative of benefit for any given individual) and too narrow (trial population and setting not representative of general practice); (3) randomizing patients can make patients and physicians uncomfortable, especially when comparing different types of existing care; and (4) there are often long delays before RCT results diffuse into practice.

The new alternative is “big data.” Because medical care is increasingly digitized in electronic health record (EHR) data sets and linked biological and genetic data banks, proponents suggest that health care systems are at the dawn of an era in which a patient’s prognosis and optimal therapy will be generated from rapid analysis of these data sets using sophisticated machine-learning strategies. The information is relatively inexpensive, generated as a by-product of patient care (overcoming the cost problem), and both specific to individuals (ie, adequately narrow) and, en masse, descriptive of the entire delivery system (ie, adequately broad). No individuals are randomized, so the ethical issues appear less complex. The richness and immediacy of these new data could allow tailored treatment decisions in real time, overcoming delays in knowledge translation. As such, although the RCT remains the gold standard for evaluation of experimental therapies, big data is proposed as a better approach for the broad swath of comparative effectiveness questions that arise in clinical practice. Indeed, the Institute of Medicine envisions big data as the engine for so-called learning health care systems.1

Please click here to access the report abstract.

This site is registered on Toolset.com as a development site.