The pitfalls and potential of precision health, big data, and evidence-based medicine
With good humor, an easy, conversational style, and an intimate knowledge of statistical meta-analysis, John P.A. Ioannidis, M.D., D.Sc., delivered a convincing argument of precision health as an impossible undertaking.
Famous for his 2005 journal article “Why Most Published Research Findings Are False,” Ioannidis— professor of Medicine, Health Research and Policy, Biomedical Data Science, and Statistics at Stanford University—spoke April 19 as part of the IHPI Research Seminar series. To a pessimist, his lecture, titled “Precision Health, Big Data and Evidence-Based Medicine–Contradictions or Companions?,” perforated with impressive swiftness precision health’s foundations of evidence-based medicine and big data. To an optimist (and, despite his iconoclasm, Ioannidis seems to be one), his talk deftly pointed out the potential pitfalls of the precision health philosophy, lessons learned from previous analyses and meta-analyses, and how to approach the undertaking with awareness and scientific rigor for sound results.
He began with a brief history of “evidence-based medicine” (EBM): a term first used by Dr. David Eddy in the ’80s, and given its best-known definition by Dr. David Sackett in 1996: “the conscientious, explicit and judicious use of current best evidence in making decisions about the care of the individual patient. It means integrating individual clinical expertise with the best available external clinical evidence from systematic research.” The emphasis on the “individual” shows how EBM was a precursor to precision health, which Ioannidis describes as “a medical model that proposes the customization of health care” to the individual.
The “E” in EBM encompasses clinical, observational, mechanistic, and other forms of evidence, but, due to allegiance biases and conflicts of interest—financial and otherwise—such evidence is typically “less than optimal,” Ioannidis said, and ultimately, researchers lack the type of evidence they need.
Part of the problem is that the tools typically used to measure quality of evidence—such as scientific papers and grant proposals—are easily skewed. He pointed to reports about cancer prognostic markers published in 2005, an unrealistic 96% of which claimed to have a significant result. “Almost any result can be obtained,” said Ioannidis, depending on how a study is structured and which questions are asked. He added, “Many treatment effects seem to be large, especially in small, early trials, but few survive scrutiny.” The average small study contains 12 participants, and any significant effect observed in such a study tends to “evaporate” with larger groups.
Many treatment effects seem to be large, especially in small, early trials, but few survive scrutiny. John P.A. Ioannidis, professor of Medicine, Health Research and Policy, Biomedical Data Science, and Statistics, Stanford University
Also problematic in EBM is that the evidence used is drawn largely from published research. Most research does not make it to publication, particularly if the results are negative and unfavorable to the trial sponsor, and trial designs are often chosen precisely because they will optimize positive results. Re-analyses of published studies, furthermore, show results completely opposite to the original studies 35% of the time, said Ioannidis. The high incidence of contradictory trial results demonstrates that the evidence in published studies is far from infallible. At the same time, however, the 65% of trials that do stand up to re-analysis often lack the validation of publication, because the same scientists who performed the original studies performed the re-analyses, and it is more likely that contradictory findings would be chosen for publication over findings that show a second instance of the same result.
Does the inclusion of big data in precision health have the potential to clarify or multiply the questions raised with EBM? And what exactly is “big data”?
Ioannidis defines big data as carrying “the least possible information content per unit.” Therefore, “the more insignificant the content of information, the bigger the data” needs to be to make the content useful. Ioannidis used a quote from Greek Nobel-Prize-winning author Odysseas Elytis to explain the importance of big data: “You’ll come to learn a great deal if you study the Insignificant in depth.”
The key, though, is knowing which data are worth studying in depth. In a 2018 article in Human Molecular Genetics, Ioannidis states, “It is even possible that people with greater access to big data eventually have worse outcomes. Some big data may confuse, overwhelm and mobilize people in vain to change their lives or seek treatment adventures for no good reason. A reliable framework is needed to decide which big data should be used.” The foundation of this framework is clinical utility.
In a 2016 essay in PLOS Medicine, Ioannidis established eight criteria for useful clinical research, and the first is to identify a real problem that needs to be solved, as opposed to applying big data to a problem that researchers have manufactured, or adding additional—but not necessarily valuable—data to a topic that has already been researched extensively. In the Human Molecular Genetics article, Ioannidis states that use of big data should be pragmatic—patient-centered, cost-effective, and feasible—as well as transparent, concluding, “The combination of EBM and [big data] may allow reaping maximum benefits.”
Ensuring Precision in Precision Health
After identifying the shortcomings in both EBM and big data approaches, Ioannidis described how the two, taken together, can meet the goals of precision health, the strengths of each compensating for the other’s weaknesses. Randomized controlled trials, a mainstay of the EBM approach, can ground the use of big data in the context of clinical utility, while big data’s emphasis on data-sharing can alleviate some of the biases inherent in EBM and “generate more openness, more transparency,” to make clinical research better. These “ways to optimize transparency and magnitude of evidence” make for a successful precision health approach, which Ioannidis says should target “important problems” and “replace wasteful interventions with something that would work.”
Ioannidis sees the potential of precision health to revitalize patient care, and that joining forces is the way to get there. “To get the most utility” out of a precision health approach, said Ioannidis, “you need to get collaborations across disciplines, joining forces with other teams that do parallel work.”