Using Probability in Medical Diagnosis - A Headache Example

Experienced clinicians begin the process of making aThe physical examination provides another source of
diagnosis upon first laying eyes on a patient, andfacts to distinguish among still-viable possibilities. If my
probability is one of the main tools they use in thispatient has migraine or medication-overuse headaches,
process. A glimpse "behind the scenes" from the pointshe might have tender muscles in her scalp and neck
of view of a diagnosing physician might help to explainbut should not have a blind spot in her visual fields,
an otherwise mysterious process.slurring of her speech or clumsiness on just one side
The diagnostic process can begin even before layingof her body. These findings, if present, would cause
eyes on the patient. As an exercise (and to prove athe probabilities of migraine and medication overuse
point) I ask medical students who are with me in theheadaches to be revised downward. By contrast, the
office to diagnose the patient we haven't seen yetprobability of a brain disease - like a tumor, for
who is still in the waiting room. Of course, they look atexample - that started with a low anchor probability
me like I'm crazy. But I tell them that we already knowwould get revised upwards.
a lot about the patient and can make some educatedIf a blood test or a scan is ordered, it is again with the
guesses. For example, we might already know thatidea that the test has been individualized to discriminate
the patient is a 34-year-old woman referred by abetween competing diagnoses and re-adjust their
family doctor because of headaches.relative probabilities.
So what have other women in their thirties referred toThere is an important principal in medical diagnosis
me for headaches ended up having as their diagnosis?called Bayes' theorem. In a nutshell, Bayes' theorem
In my neurology practice, as well as in those of moststates that the probability of a diagnosis after a new
other headache specialists, about a third (33%) havefact is added depends on what its probability was
migraine, another third have medication-overusebefore the new fact was added. Another way of
headaches (in which the treatment has become thesaying this is that the same "yes" answer on
problem instead of its solution), and the remaining thirdhistory-gathering, reflex result on physical exam or
fall into an "everything else" category that includesdark spot on an MRI scan has different implications in
tension-type headaches, arthritis of the neck ordifferent people. The meaning of each depends on its
jaw-joints, sinus disease, tumors, etc. So before seeingcontext. Yet another implication of Bayes' theorem is
the patient I'm already able to identify the two mostthat one can't skip past the history and examination by
likely diagnoses and assign an initial probability for each.ordering a test in isolation and expect it to make an
These starting-point likelihoods are called "anchor"accurate diagnosis. A test is an answer to a question.
probabilities. During the subsequent history, examinationIf there was no question, how could the test be an
and supplemental testing (if necessary) the anchoranswer?
probabilities will undergo a series of upward andLet's say that at a particular point in time we have
downward adjustments according to what the patientcompleted the diagnostic process for a patient. Then
has to say and what does or does not turn up on herwhat? By the end of the diagnostic process the
physical examination and testing. The physiciandoctor might have a diagnosis that is nearly 100% likely,
individualizes the questions asked and items examinedbut in other cases, the working diagnosis (number one
so that the outcome of each query forces onechoice) might still be just 70% or 80% probable, with a
diagnosis to be more likely and another to be less likely.number two choice less likely, but still on the radar
Thus, diagnosis is a dynamic and sequential process.screen. It might make some patients uncomfortable to
We invite the woman into the examining room andrealize that the diagnostic process does not lead to
listen to her story. In the headache example given, one100% certainty in every case, but a doctor wouldn't be
key piece of data is how many days per month shedoing a patient any favors by pushing the analysis past
takes an as-needed medication - for example, aspirin,the outcome that the available information leads to.
acetaminophen or a prescription drug. If she takesWhen a diagnosis is not 100% likely at the time of initial
as-needed medicine more days than not and has beenevaluation, the patient's course of symptoms over time
doing so for a matter of months, then the initial 33%provides yet another form of data that can lead to
anchor probability of medication-overuse headachesrevision of diagnostic probabilities. Fortunately, in cases
gets adjusted upward and the initial anchor probabilityinvolving uncertainty, even just narrowing down the list
of uncomplicated migraine moves downward. This, ofof diagnoses to a small number of concrete
course, is just a single distinguishing feature, and cannotalternatives allows the doctor and patient to discuss
be relied upon to tell the whole story. The physicianreasonable options and make sensible choices.
gathers many such data points to refine the diagnosis.