Crozet Annals of Medicine: Shades of Gray


Medicine is a science of uncertainty and an art of probability. 
–William Osler, MD

Medicine is a science based on probabilities and thus error is built into the science. 
–Robert Reiser, MD

 By Dr. Robert C. Reiser

There are very few yes or no answers in medicine. This is hard for patients and families to grasp. It is also hard to teach to new physicians. Our faith in high technology blinds us to the true nature of medical decision-making, which is always predicated on calculating a range of probabilities. I was reminded of this on a recent morning when I was supervising a resident taking care of a sweet little 100-year-old lady who had fallen down for no apparent reason. Well, I suppose the fact that she was 100 might have been the reason, but remember there are very few yes or no answers in medicine.

She had some neck pain after her fall so we ordered an X-ray of her neck to make sure she did not have a fracture. When you order an X-ray you are required to provide your colleague the radiologist some history and a general idea of what you are looking for, what question you want answered. In medical speak it might go something like this: 100 y/o f, s/p GLF, r/o c-spine fx. Translation: A 100-year-old female status post ground level fall (she fell from standing earlier today) please rule out (make sure she doesn’t have) a broken neck.

Then the chess match begins. The radiologist, cooped up in a dark room all day staring at black and white pictures sees only shades of gray, endless probabilities and no yes or no answers. She sends us her report, answering our question with a question of her own.

“Age indeterminate degenerative changes of the bones of the neck, likely chronic in nature but cannot rule out fracture in the appropriate clinical setting. Suggest clinical correlation.”

Well jeeze, that is not an answer. If we could tell clinically if she had a broken neck we wouldn’t need an X-ray. Of course these degenerative changes could be chronic; she is 100 years old.

So I called the radiologist for some clarification and she gave me the probabilities, most likely not fractured but impossible for her to say no for sure. She suggested a CT of the neck, which we ordered. The report came back.

“The previously noted age indeterminate changes favor congenital abnormalities versus fracture, which cannot definitively be excluded. If the clinical question persists MRI may be helpful to further delineate these changes.”

After 100 years we may have discovered a congenital (present from birth) abnormality? Wow. But still the question of fracture had not been ruled out.

By now the sun was setting on my 100-year-old lady. She was getting tired and confused and couldn’t remember if her neck still hurt. Time to fish or cut bait. I ran the probabilities by the resident on the case. The CT scan was 95 percent sensitive in picking up fractures, meaning it would pick up 95 of 100 fractures. Pretty good, but it would miss 5 of 100 fractures. The patient’s pre-test probability of having a fracture based on her age and history was estimated at 10 percent, pretty low. I asked the resident what the likelihood was after the CT that the patient had a fracture that we had missed. This is called the negative predictive value; the likelihood that a negative test means the patient doesn’t have the condition. This is really what we want to know.

The resident confidently stated that since there was a five percent chance of missing a fracture the negative predictive value was only 95 percent. He did not feel comfortable with only a 95 percent certainty that the patient did not have a broken neck and thought we should pursue an MRI despite the patient’s increasing agitation and confusion brought on by her prolonged stay in the unfamiliar setting of the ER.

I redid the math for the resident. A 95 percent sensitive test applied to a population with a 10 percent incidence of the disease (similar to our patient’s 10 percent risk) will pick up 9.5 of the 10 cases, missing 0.5 cases, meaning a negative predictive value of 99.5 percent.

I was 99.5 percent confident that my patient did not have a fracture. There is no test that will ever reach 100 percent. This is where the error is necessarily built into the decision-making. It was time to stop testing. Time to put granny to bed.

I summed it all up for the resident. Risk versus benefit. There are no yes or no answers in medicine, young man. He nodded agreeably at my wisdom and then slyly pointed out the radiology request form I had filled out for the CAT scan for our 100-year-old lady. As on all radiology request forms, there was a standard question, was the patient pregnant?

I had circled no.