A new tool for predicting which patients with psoriasis will develop psoriatic arthritis (PsA) is showing promise for such clinical applications as early treatment in those at risk or trials to prevent PsA, according to a summary of progress at the annual meeting of the Canadian Rheumatology Association.
Based on current levels of sensitivity and specificity, psoriasis “can be predicted with reasonable accuracy,” reported Lihi Eder, MD, PhD, director of research in the rheumatology division at the University of Toronto.
Dr. Lihi Eder
The predictive method, called PRESTO (Prediction of Psoriatic Arthritis Tool), is based on variables readily available in clinical practice, according to Dr. Eder. Once values are assigned to the risk factors, the risk of PsA over a 1-year or 5-year time frame can be estimated with a calculator.
She called PRESTO the “first clinical tool for predicting PsA among psoriasis patients.”
The work on this tool began in 2006 when the International Psoriasis and Arthritis Research Team (IPART) initiated a prospectively collected cohort of psoriasis patients. To be enrolled, patients had to be free of signs and symptoms of arthritis upon examination by a rheumatologist. They were then invited to return annually for follow-up that again included screening for joint involvement by a rheumatologist.
At baseline and at follow-up evaluations, 13 predictors were evaluated. These involved psoriasis characteristics, such as nail pitting; symptoms, such as stiffness; comorbidities, such as additional inflammatory diseases; and laboratory values, such as upregulated markers of inflammation.
Symptoms and signs used to predict PsA
Dr. Eder and her colleagues applied regression models to select an optimal combination of variables weighted for predictive value. Variables offering predictive value included higher PASI (Psoriasis Area and Severity Index), greater fatigue score as measured by FACIT (Functional Assessment of Chronic Illness Therapy) score, greater morning stiffness, and greater pain.
When applied to 635 patients in the IPART cohort, in which there were 51 incident PsA cases over 1 year and 75 incident cases over 5 years, the area under the curve (AUC) for PRESTO at the cutoffs studied was 72% for the 1-year time window and 75% for the 5-year time window.
These levels are associated with adequate accuracy, according to Dr. Eder, who explained that “an AUC greater than 70% is considered reasonable” for clinical applicability.
Moreover, the cutoffs can be adjusted for the specific purpose of the predictive tool. For example, to screen patients for risk, lower cutoffs could be employed to increase sensitivity. In order to select patients for a clinical trial to prevent PsA, higher cutoffs could be employed to increase specificity.
But sensitivities and specificities move in opposite directions when cutoffs are adjusted. Showing data from the 5-year prediction model, Dr. Eder reported that specificities climbed from about 58% to 97% as cutoffs were increased. The sensitivities with these adjustments fell from 79% to 14%.
In general, Dr. Eder said there was “excellent calibration” for the cutoffs employed when they compared the predicted and observed rates of PsA according to quintile of predictive probability. The differences were particularly minor over a 1-year time period. Over the 5-year period, observed rates were somewhat higher than predicted in the fourth and fifth quintile, but, again, this discrepancy could be modified for specific applications with cutoff adjustments.