Diagnosis through differentiation: a pilot study on improving the diagnostic efficiency of primary headaches in ICHD3

What is the minimum number of questions necessary to diagnose any given primary headache disorder? The answer to this question may provide clinicians with a more efficient approach to history taking and lead to a more concise form of ICHD3. In this project, we attempt to address this problem mathematically.

We defined a headache phenotype as a collection of characteristics, variables that can take on a Boolean (true/false) value and correspond to elements of the diagnostic criteria for each headache in ICHD3. There may be multiple phenotypes that fit a given diagnosis in the ICHD3. We extracted all characteristics used to describe primary headaches up to two levels deep in the hierarchy in the ICHD3. Episodic syndromes associated with migraine-recurrent gastrointestinal disturbance, benign paroxysmal vertigo, benign paroxysmal torticollis-were excluded because these require a primary diagnosis of migraine. For each headache diagnosis, we determined its "necessary true" (NT) characteristics. We also generated a list of "necessary false" (NF) characteristics by identifying characteristics that logically contradict the NT for each given diagnosis. We then sought to algorithmically identify the smallest set of NT and NF needed to differentiate between all primary headache diagnoses. As a result, any primary ICHD3 headache phenotype can be diagnosed once both the NT and NF criteria are satisfied. We verified this by translating all the possible conditions described by the ICHD3 criteria to our phenotype encoding schema.

We were able to minimize the NT and NF criteria to a set of 22 and 6 characteristics, respectively, with 5 overlaps between the groups. This results in a final set of 23 unique characteristics. Though an even smaller NT set may be possible, we are limited by computational power. These characteristics can be queried by using the following questions to generate a headache phenotype: (1) duration, (2) frequency, (3) sudden/rapid onset, (4) laterality, (5) clearly remembered onset, (6) sharp contour, (7) severity, (8) relationship to sleep/awakening, (9) reversibility of aura, (10) stabbing quality, and whether the headache can be triggered by (11) sex, (12) compression, (13) traction, (14) cold, or (15) exercise.

Fifteen questions are necessary to differentiate the primary headache disorders in ICHD3. A smaller set may be possible, but we cannot prove its existence. Using this reduced set of questions, clinicians may be able to more efficiently arrive at ICHD3 diagnoses, but further research is required.