How big data can hurt healthcare

Big data must be coupled with rigorous observational methods to prevent grave errors in assumptions, according to an article published at the American Journal of Managed Care.

Austin B. Frakt, Ph.D., health economist with the Department of Veterans Affairs, and Steven D. Pizer, Ph.D., director of healthcare financing and economics at the VA Boston Healthcare System, note that for every 5 million packages of x-ray contrast media distributed to healthcare facilities, about six people die from adverse affects.

With big data, those deaths can be found to be highly correlated with things like electrical engineering doctorates awarded and per-capita mozzarella cheese consumption. 

"[B]ecause we cannot conceive of a causal mechanism, it is obvious that these variables play no causal role in x-ray contrast media deaths. That such high correlations can be easily mined from big data is concerning nonetheless, because it is not always trivial to assess whether they are telling us something useful.

"The way forward requires careful selection of observational research designs coupled with rigorous testing for violations of key assumptions on which causal inference relies," they say.

A falsification, which aims to pinpoint false assumptions about a suspected causal variable and outcomes, is one way to do this.

One example Frakt and Pizer use is a study that took Veterans Health Administration and Medicare data for more than 80,000 patients to see mortality and hospitalization rates among Type 2 diabetes patients prescribed two types of drugs; sulfonylureas (SUs) and thiazolidinediones (TZDs). The researchers used falsification tests to analyze whether the prescribing patterns were random in the way a clinical trial randomization would be. The researchers found ways that falsification tests can help boost the confidence that outcomes are truly causal.

With the push to value-based reimbursement, healthcare organizations are rethinking their strategies for embedding analytics in clinical care workflows and trying to figure out how to reduce the "time to value'" of their data, according to John Showalter, M.D., University of Mississippi Medical Center's chief health information officer.

Penn Medicine, for one, uses an algorithm to flag when a patient is slipping into severe sepsis, based on vital signs and lab values.

To learn more:
- here's the article