ThinkRare: A search algorithm to identify patients with undiagnosed rare genetic disease in an electronic medical...

Undiagnosed rare genetic diseases (RGD) can go unrecognized by health care providers, delaying appropriate genetic testing. This proof-of-concept study aimed to address this barrier through the development of a rule-based search algorithm called "ThinkRare."

The algorithm used structured electronic medical record data and clinical criteria to identify patients who may have a complex undiagnosed RGD, are eligible for clinical exome sequencing, and are not yet referred to genetics (true positives). Iterative testing on gold standard and test data sets (input) informed algorithm design and optimization. Medical record reviews were conducted to verify whether the algorithm identified patients (output) were true positives or false positives, and these outcomes informed algorithm modifications. Physicians of identified patients were notified with the option to refer to genetics.

The search algorithm was applied retrospectively to 262,296 patients (test data set), excluding 99.9% of patients and identifying 30 patients eligible for exome sequencing. The algorithm's estimated recall (sensitivity) was 60% and precision (positive predictive value) was 15%. This process resulted in the diagnosis of 50% of patients (4/8) referred and evaluated in genetics.

The search algorithm effectively identified patients retrospectively with an RGD and prospective deployment will ultimately help physicians "think rare."