Leveraging clinical intuition to improve accuracy of phenotype-driven prioritization

Abstract

Purpose:

Clinical intuition is commonly incorporated into the differential diagnosis as an assessment of the likelihood of candidate diagnoses based either on the patient population being seen in a specific clinic or on the signs and symptoms of the initial presentation. Algorithms to support diagnostic sequencing in individuals with a suspected rare genetic disease do not yet incorporate intuition and instead assume that each Mendelian disease has an equal pretest probability.

Methods:

The LIkelihood Ratio Interpretation of Clinical AbnormaLities (LIRICAL) algorithm calculates the likelihood ratio of clinical manifestations represented by Human Phenotype Ontology terms to rank candidate diagnoses. The initial version of LIRICAL assumed an equal pretest probability for each disease in its calculation of the posttest probability (where the test is diagnostic exome or genome sequencing). We introduce Clinical Intuition for Likelihood Ratios (ClintLR), an extension of the LIRICAL algorithm that boosts the pretest probability of groups of related diseases deemed to be more likely.

Results:

The average rank of the correct diagnosis in simulations using ClintLR showed a statistically significant improvement over a range of adjustment factors.

Conclusion:

ClintLR successfully encodes clinical intuition to improve ranking of rare diseases in diagnostic sequencing. ClintLR is freely available at https://github.com/TheJacksonLaboratory/ClintLR.