POS0311 FLARES IN PATIENTS WITH RECENT-ONSET PSORIATIC ARTHRITIS. PREDICTIVE MODEL BASED ON MACHINE LEARNING

Abstract

BackgroundAn important aspect in the clinical care of patients with PsA is to be able to predict the occurrence of a flare using tools and information that are readily available in daily clinical practice. This information would provide added value in disease management, yet, unfortunately, scarcely any studies provide it.ObjectivesTo identify patient- and disease-related characteristics that make it possible to predict flares in recent-onset PsA.MethodsWe performed a multicenter observational prospective study (2-year follow-up, regular annual visits), promoted by the Spanish Society of Rheumatology [1]. The study population comprised patients aged ≥18 years who fulfilled the CASPAR criteria [2], with less than 2 years since the onset of symptoms. The intention at the baseline visit was to reflect the patient’s situation before disease progress was modified by the treatments prescribed in the rheumatology department.All patients gave their informed consent. The study was approved by the Clinical Research Ethics Committee of the Principality of Asturias.Flares were defined as inflammatory episodes affecting the axial skeleton and/or peripheral joints (joints, digits or entheses) and diagnosed by a rheumatologist between the previous and the current visit.The dataset contained data for the independent variables from the baseline visit and from follow-up visit number 1. These were matched with the outcome measures from follow-up visits 1 and 2, respectively. We trained logistic regression models and a random forest–type machine learning algorithm to analyze the association between the outcome measure and the variables selected in the bivariate analysis (statistical significance was defined as p value <0.05). We used a confusion matrix to visualize the performance of the final model. This matrix shows the real class of the data items, together with the class predicted by the machine learning algorithm, and records the number of hits and misses.ResultsThe sample comprised 158 patients. 14.6% were lost to follow-up. At the first follow-up visit, 37.6% of the patients who attended the clinic had experienced flares since the baseline visit. Of those who attended the second visit, 27.4% had experienced flares since the first visit. Table 1 shows the results of the logistic regression analysis. The variables predicting flares between visits selected in this analysis were age-adjusted Charlson comorbidity index, PsAID score, number of digits with onychopathy, and level of physical activity. The direction of the association was negative for the Charlson index and physical activity and positive for PsAID score and onychopathy.Table 1.Variables associated with flares between visits selected in the logistic regression analysis.VariableRegression coefficient95% CIp value (Wald test)Age-adjusted Charlson comorbidity Index-4.655(-7.021, -2.289)<0.001PsAID score2.212(1.171, 3.254)<0.001No. of digits with onychopathy1.420(0.331, 2.511)0.011Level of physical activity-1.221(-1.87, -0.572)<0.001When the random forest machine learning algorithm was trained with these 4 variables, the order of importance (from more to less) attributed by the model was as follows: PsAID score, number of digits with onychopathy, age-adjusted Charlson comorbidity index, and level of physical activity. The percentage of hits in the confusion matrix was 78.38%.ConclusionPsAID score was the first variable in the predictive hierarchy generated in our model, supporting its importance in the management and follow-up of PsA patients.