Decision Tree Analysis as a Preliminary Evidence-Based Tool for Identifying the Syndrome of Undifferentiated...

To develop evidence-based criteria to classify patients with syndrome of undifferentiated recurrent fevers (SURF).

One hundred twelve patients with SURF observed in a single tertiary referral center were analyzed. Patients with genetically confirmed hereditary recurrent fever (HRF) or with periodic fever, aphthosis, pharyngitis, and adenitis (PFAPA) syndrome already analyzed for the Eurofever classification criteria were used as disease controls. A decision tree approach was tested by randomly splitting the available data in a training set and in an internal test set. An alternative model using a classical regression model was also analyzed. An external validation for both approaches was performed on 123 patients recruited from four other centers.

The decision tree model integrating clinical and genetic data identified 91% of patients with SURF. A decision tree model based solely on clinical variables identified up to 88% of patients with SURF. The logistic regression model including genetic tests exhibited an overall accuracy of 89.2% (95% confidence interval [CI] 81.1-94.7). In contrast, the logistic regression model exclusively based on clinical manifestations displayed an overall accuracy of 66.7% (95% CI 56.1-76.1). When the classification criteria including genetic tests were applied to the external validation cohort, the model demonstrated a strong discriminative power, with areas under the receiver operating characteristic curve of 96.3% using the decision tree model and 88.0% with the logistic regression model.

The study shows the possibility of achieving evidence-based criteria that can classify SURF at least with respect to the main HRF and PFAPA syndrome and may be considered as a preliminary tool for the enrollment of more homogeneous cohorts of patients in future studies.