Shoulders of Giants: Rules of Evidence
Updated: Sep 2, 2021
In this series we have reviewed some (mostly) famous examples of research in medical science that illustrate paths to future success: elucidation, courage, belief in one’s ideas and observations, persistence and, if necessary, the ability to go it alone.
This review has also led to exploration of the various methods used to design clinical research for medical science. We have seen how the best designs offer researchers and clinicians the best evidence and guidance for treating their patients. The scientific method must be addressed as the primary structure of all research. Even when unconventional methods are employed, hypotheses should focus on one isolated variable which can be tested. Sample error is very difficult to avoid; utilizing randomness and prospective studies help mitigate against this.
Prospective studies, either as controlled experiments, observational studies or registries, offer the more error free method for obtaining results over retrospective studies (which run the greater risk of sample error, as true randomness is difficult to achieve retrospectively). Randomness emerges as the key design element in clinical research. Volunteers must be randomly assigned to treatment or nontreatment groups. Researchers, as well as the participants, must be blind to the group assignment.
Randomized, placebo controlled, double blind prospective studies (RCT) therefore are the gold standard for clinical research studies. Though expensive and time consuming, this design best avoids confounding (missing or confusing real causes), cognitive bias and sampling error, all of which make it less likely a patient will respond to treatment in the same way the study participants did. When RCT data is available, it should be ranked as the highest evidence, provided the study was designed without flaws and powered appropriately (meaning that enough participants were included in each study cell: placebo group, treatment group, etc.).
There are times, however, such as during our recent pandemic, when answers are needed and there is not enough time to complete these rigorous studies. What to do then? This series also identified some study designs that, while falling short of the validity provided by RCT, still have some data to offer.
At any given time, clinicians may be called upon to make treatment decisions based on insufficient data. The Achilles’ heel of evidence-based medicine, is, after all, that there is often not enough evidence. Practioners must then weigh what evidence is available and make their best choice, as they endeavor to “do no harm.”
If RCT are underpowered or flawed, or unavailable altogether, then observational studies may help fill the gap. We have examined two such studies, one well done and one flawed (see Shoulders of Giants: Can Seeing Be Believed? 21.3.10). All studies should be evaluated by the nature of their design and by how well they conform to their original design. When good RCT are not available or feasible, naturalistic studies and registries, preferably prospective, may provide some information that, while not ideal, is still useful.
Scholarship is an essential foundation in medicine. The prepared mind will be able to sift through evidence, rank it according to the methods through which it was obtained and, in the face of necessity, utilize it based on quality. As the task is almost Herculean, we can be grateful we have giant shoulders to stand upon.
This Shoulders of Giants series will now take a break so that other important topics can be explored in future posts. Coming up: social media and mental health.
 Barton S. Which clinical studies provide the best evidence? The best RCT still trumps the best observational study. BMJ. 2000;321(7256):255-256. doi:10.1136/bmj.321.7256.255  Frieden TR. Evidence for Health Decision Making – Beyond Randomized, Controlled Trials. N Engl J Med 2017; 377:465-475. DOI: 10.1056/NEJMra1614394