A new diagnostic tool can effectively discriminate different rheumatologic conditions and could potentially aid in the diagnosis of early inflammatory arthritis.
The algorithm — called Genetic Probability tool (G-PROB) — uses genetic information to calculate the probability of certain diseases.
University of Manchester
Dr. John Bowes
“At such an early stage of disease, it’s not always easy to determine what the final outcome will be with respect to final diagnosis,” said John Bowes, PhD, a senior lecturer in the division of musculoskeletal & dermatological sciences at the University of Manchester in the United Kingdom. He was a senior author of the newest study of G-PROB. “What we are hoping for here is that genetics can help [clinicians] with the decision-making process and hopefully accelerate the correct diagnosis and get individuals onto the correct treatment as early as possible.”
Creating the Algorithm
G-PROB was first developed by an international group of scientists with the goal of using genetic risk scores to predict the probabilities of common diagnoses for patients with early signs of arthritis, such as synovitis and joint swelling. According to the study authors, about 80% of these types of patients are eventually diagnosed with the following conditions: Rheumatoid arthritis (RA), systemic lupus erythematosus (SLE), psoriatic arthritis (PsA), ankylosing spondylitis (AS), and gout.
The algorithm combines existing knowledge about single-nucleotide polymorphisms from prior genomic studies to create genetic risk scores — also called polygenic risk score (PRS) — for multiple diseases. Using these scores, the program then calculates the probabilities of certain diagnoses for a patient, based on the assumption that at least one disease was present.
In this first study, researchers trained the tool on simulated data and then tested it in three patient cohorts totaling about 1700 individuals from the Electronic Medical Records and Genomics database and Mass General Brigham Biobank. In the initial study, G-PROB identified a likely diagnosis in 45% of patients, with a positive predictive value (PPV) of 64%. Adding these genetic scores to clinical data improved diagnostic accuracy from 39% to 51%.
Validating G-PROB
But data from these biobanks may not necessarily be representative of early arthritis in patients appearing in outpatient clinics, noted Dr. Bowes. In this new study, researchers sought to independently validate the original study’s findings using data from the Norfolk Arthritis Register, a community-based, long-term observational study on inflammatory polyarthritis. The team applied G-PROB in this cohort and then compared the tool’s probabilities for common rheumatic conditions to the final clinician diagnosis.
The study ultimately included 1047 individuals with early inflammatory arthritis with genotype data. In the cohort, more than 70% (756 individuals) were diagnosed with RA. Of the remaining patients, 104 had PsA, 18 had SLE, 16 had AS, and 12 had gout. The research team also added an “other diseases” category to the algorithm. A total of 141 patients fell into this category and were diagnosed with diseases including chronic pain syndrome (52 individuals), polymyalgia rheumatica (29 individuals), and Sjögren’s syndrome (9 individuals).
G-PROB was best at excluding diagnoses: Probabilities under 5% for a single disease corresponded to a negative predictive value (NPV) of 96%. If probabilities for two diseases were both < 5%, the NPV was 94%.
For patients with a single probability above 50%, the tool had a PPV of 70.3%. In 55.7% of all patients, the disease with the highest probability ended up being the final diagnosis.
Generally, PRSs, as well as tests using biomarkers, were better at excluding diagnoses than affirming them, noted Matthew Brown, MBBS, MD, a professor of medicine at King’s College London, who was not involved with the research. If disease prevalence is low, then a test aimed at diagnosis of that disease would be better at excluding a diagnosis than affirming it, he explained.
Queensland University of Technology
Dr. Matthew Brown
However, he noted that G-PROB’s PPV may have performed better if researchers had started by using established PRS scores to form the algorithm, rather than developing these genetic scores independently using internal datasets.