The use of artificial intelligence (AI) for the automatic screening of rare diseases such as hereditary transthyretin amyloidosis (hATTR) may be a viable way of diagnosing and treating these conditions early, according to a novel study published in the Journal of the Peripheral Nervous System.

For the purpose of this study, the researchers used the novel natural language processing (NLP) algorithm to scan through the electronic medical records of 1015 patients treated at the neurology department at the University Hospitals Leuven in Belgium. A total of 26,241 records were processed, averaging 25.85 records per patient.

The AI tool was set to detect the predetermined symptoms grouped into 8 red flag categories including family history, early autonomic dysfunction, gastrointestinal complaints, unexplained weight loss, cardiac problems, carpal tunnel syndrome, renal problems, and ophthalmological abnormalities.

A red flag was defined as the presence of multiple symptoms belonging to the same category, while 3 red flags indicated a high risk for hATTR with polyneuropathy and called for further review.

Read more about hATTR testing

According to the results, the NLP algorithm marked 128 patients with 3 or more red flag symptoms out of which 69 patients were eligible for genetic testing after clinical review.

Next, the accuracy of the novel algorithm was assessed through comparison with a manual „gold standard“ review conducted by 2 experienced physicians. The experts researched the multidisciplinary medical records of 300 random patients that were originally part of the total study sample.

According to subsequent analysis, the algorithm obtained an overall recall of 0.93, a precision of 0.92, and an F1 measure of 0.93 on a red flag or category level. Looking at particular symptoms, atrioventricular block, cardiomyopathy, and uremia performed less than the predetermined accuracy (F1 measure <0.85) which can be attributed to a couple of causes.

“Independent on the outcome, the need for these types of models to be put into practice is high, as the average diagnostic time delay in rare diseases is between 6 and 8 years, which is mainly due to the high amount of unstructured information because of its multidisciplinary and heterogeneous character,“ Hens and colleagues noted.

Rare life-threatening diseases, such as hATTR caused by a mutation in the transthyretin gene, are often diagnosed relatively late in the disease course. The screening method using the NLP algorithm led to a 48% increase in genetic testing.

Reference

Hens D, Wyers L, Claeys KG. Validation of an artificial intelligence driven framework to automatically detect red flag symptoms in screening for rare diseases in electronic health records: hereditary transthyretin amyloidosis polyneuropathy as a key example. Published online on December 5, 2022. J Peripher Nerv Syst. doi:10.1111/jns.12523