For familiar diseases, clinicians decide what findings are useful to check based on what seems likely to change the differential diagnosis. Furthermore, clinicians take into account treatability of diseases and cost of tests. However, doing so for rare diseases is difficult because few clinicians have the details about such diseases in their heads.
The software implements the usefulness assessment that clinicians do in their heads by using a computational statistics metric of “Usefulness”. This is possible because of an extensive database of diseases and their treatability, the findings in these diseases and the cost of checking the findings, e.g., the cost of a lab test.
Example to work through
This element is illustrated with the case of an 8-month-old boy with recent onset of macrocephaly (big head) and kyphosis (outward bulge of the back). A ranked list of diseases appears in the differential diagnosis on the left, with blue shading representing probability. All these diseases are rare and unfamiliar to most clinicians. A ranked list of useful findings appears on the right, with green shading representing Usefulness. (Registered users can click the images below to jump into the software; some videos may be helpful in navigating from there)
Consider the useful finding of short stature. A superficial analysis of the literature reveals that in the #1 diagnosis, MPS1-H (Hurler syndrome), all affected individuals are short. This patient, however, was not short, so someone unfamiliar with this rare disease might dismiss the possibility of this diagnosis. However, the database includes the knowledge that children with this disease only become short later on. Adding the absence of short stature actually raises the probability of MPS1-H because it reduces the probability of other diseases in the differential diagnosis in which patients would already be short.
Adding the absence of short stature changes the differential diagnosis and thus changes the Usefulness of other findings; as shown below, the Usefulness is now higher for assessing whether there is corneal clouding. Many clinicians have never observed the unusual finding of corneal clouding, but just-in-time assistance is available by clicking the🔗symbol in the top black Navigation bar, for which the adjacent green number (green is the color for findings) indicates that information about corneal clouding is available.
Adding the presence of corneal clouding (with onset not specified because no one checked previously), dramatically raises the probability of MPS1-H.
The same approach also applies to lab tests on the “Add tests” tab. Clicking the word “Bundle” next to “Limb bone imaging” shows the findings in that bundle, and one could add to the patient phenotype that dysostotic thickening of long bones was present in this patient. This further solidifies the diagnosis of a disease that was previously unfamiliar.
Specialists such as radiologists can focus on tests within their domain by changing the default of “All tests” (top right in the image above) to “Radiology”, showing the most useful radiological findings so that the radiologist can not only comment on pertinent positive findings but also on pertinent negative findings that will be of particular relevance in diagnosis (the video “Filtering for a particular type of lab test” shows an example of such filtering for Gene Tests).
An attending rheumatologist at one of the Harvard hospitals described a case in which by using this approach he “became convinced I should consider Morvan Syndrome in the differential diagnosis. The software suggested the relevant test. I ordered it and the test results were positive. I felt like I had superpowers to diagnose a disease I didn’t know beforehand.”
The cognitive error associated with not using this element is called “Zebra retreat”: not exploring the possibility of a disease because of its unfamiliarity.
This page is part of a series on the Elements of Diagnosis.
Copyright © 2025 SimulConsult