A proof-of-principle study demonstrates how genetic information could expedite the diagnosis of slowly progressing inflammatory diseases that exhibit similar symptoms
A proof-of-principle study demonstrates how genetic information could expedite the diagnosis of slowly progressing inflammatory diseases that exhibit similar symptoms.
The new tool, developed by Assistant Professor Rachel Knevel, a clinical research fellow at Leiden University Medical Centre in the Netherlands, and rheumatologist Professor Soumya Raychaudhuri at the Broad Institute in the US, has been designed for when a patient presents with swollen and inflamed joints.
It delivers a probability for each of the five rheumatic diseases most likely in that situation – rheumatoid arthritis, systemic lupus erythematous, psoriatic arthritis, spondyloarthropathy, and gout – which clinicians can use to make their diagnosis.
The new tool, named G-PROB, combines a patient’s genetic profile with population-based genetic risk scores for variants associated with disease susceptibility, derived from published studies.
Two real-world datasets – with a combined total of nearly 1,500 patients – were used to road-test the ability of G-PROB to correctly discriminate between the five diseases in question. Then it was applied to a pseudo-prospective dataset of roughly 240 patients who had presented with unexplained inflammatory arthritis.
G-PROB could rule out at least one disease in all patients, identified a likely diagnosis for 45% of patients, and showed that 35% of patients were misdiagnosed by their rheumatologist at their first appointment.
“Its performance exceeded our expectations, particularly that genetics can rule out particular diseases with high certainty,” Professor Knevel said.
However, the study didn’t test what clinicians would do with the information G-PROB provided and if its use would indeed lead to less misclassification and quicker diagnoses, as promised.
Professor Knevel said it would be feasible to use G-PROB in practice if about half of patients had genotyped data, which she said could happen within 10 years.
Thomas Hügle, professor of rheumatology at the University of Lausanne in Switzerland, said he doesn’t believe the availability or quality of genetic information will be a bottleneck to using this tool in clinical practice – if health insurance companies start to reimburse genetic testing in rheumatology, as is the case in oncology.
Professor Hügle said G-PROB is a robust algorithm that would help rheumatologists diagnose cases of undifferentiated arthritis. “But this is no self-diagnosis tool,” he said, though it could help avoid more invasive diagnostic procedures used to investigate joint inflammation, such as synovial biopsy.
Imogen Stafford, a PhD candidate at the University of Southampton in the UK who recently reviewed how computational algorithms have been applied to autoimmune diseases, said G-PROB is relatively simple in its design compared to other, more complex artificial intelligence systems. “This makes it more transparent for clinicians,” she said. “Instead of being designed as an algorithm that makes a definitive decision about patient care, it aids the prioritisation of clinical testing.”
Moreover, Stafford said that machine learning models very rarely focus on multiple diseases at once. Trying to predict which disease a patient has from a range of likely diagnoses – especially where early diagnosis is important – is another strength of the tool, she said.
