I'm using my "5th Tuesday" commentary option this month to mention not an original study, but rather an accompanying editorial for an article on hypothermia treatment for neonatal hypoxic ischemic encephalopathy. I hope the original article itself appears in an upcoming AAP Grand Rounds series, maybe I'll include it in a subsequent commentary if it does.
Quintana M, Lewis RJ. Bayesian analysis: using prior information to interpret the results of clinical trials. JAMA 2017; 318:1605-6. doi:10.1001/jama.2017.15574.
Thomas Bayes, in addition to being a Presbyterian minister, also was an accomplished statistician. I suspect he never expected how his eponymous Theorem would impact the world today.
The editorial article cited above is part of a series in this journal, the JAMA Guide to Statistics and Methods. Unfortunately, I haven't found a foolproof method to search this collection separately, so I can't give you a decent link to it. I'd suggest googling. The editorial primarily discusses how Bayesian statistical analyses are helpful in clinical situations where large quantities of patient data are not available, allowing for synthesis of data from multiple sources. Here, though, I'd like to focus on the importance of Bayesian reasoning in clinical medicine. I consider it an important road forward in evidence-based clinical practice.
For starters, let me tell you about 2 examples from my prior EBM teaching to highlight the importance of pretest probability in clinical decision making.
First, take yourself out of your pediatric mindset and imagine a 64 year old gentleman seeking your care for a 3-week history of early morning vomiting. You decide to order a urine pregnancy test, and the result is positive. What is the probability that this patient is pregnant? I hope you'll come to the conclusion that the answer is zero, given both sex and age of the patient, and that you shouldn't have ordered the test in the first place. This is a situation where the pretest probability is zero, so the test result doesn't matter.
Another example I've used is ordering a stool rotavirus assay in a 4 month old child with acute diarrhea. If this child lives in a more northern climate, where rotavirus is highly seasonal and occurs almost exclusively in the winter months, a positive result in the summer approximates the male pregnancy absurdity above (unless the child has recently traveled to the southern hemisphere). Furthermore, even if you ordered it in the winter, in our era of rotavirus immunization a positive result might more likely reflect the fact that the infant has been immunized, rather than being the diagnostic answer. Again, pretest probabilities greatly impact how we (should) use clinical information.
To add a bit more numbers to the question, here's an example widely studied and used:
One percent of 40 year old women who have routine mammography have breast cancer. Eighty percent of women with breast cancer have positive mammograms. Women without breast cancer have positive mammograms 9.6% of the time. A 40 year old woman seeking your care has a positive mammogram in a routine screening. What is the probability that she actually has breast cancer?
Take a couple minutes to think about your answer. Note that studies have shown only 15% of physicians get the correct answer. You can find a good discussion of this problem online. It will show you how Bayesian reasoning, whereby the clinician considers pretest probabilities prior to deciding on tests or treatments, can improve outcomes.
If you want a great explanation of Bayesian reasoning, with a choice of how you want to view the explanations, try out this new site; I recommend it highly.
If we all used Bayesian reasoning in our everyday clinical practice, we'd like save a lot of money, and our patients would have better outcomes. Think about it.
Oh, and the correct answer to the screening mammography problem is that 7.8% of the women with positive screening mammographies will actually have breast cancer.
