A decision tree-based algorithm for structured risk stratification of rare rheumatic diseases in a tertiary referral...

Area of Science:

• Rheumatology
• Medical Diagnostics
• Data Science in Medicine

Background:

• Rare inflammatory rheumatic diseases present with diverse, multisystemic symptoms, complicating early diagnosis.
• Patients referred to tertiary centers for rare diseases often have unclear systemic complaints requiring structured diagnostic approaches.

Purpose of the Study:

• To develop and evaluate a structured, symptom- and laboratory-based decision approach for risk stratification of rheumatologic diagnoses.
• To model a decision tree for prioritizing diagnoses in patients with undifferentiated systemic complaints.

Main Methods:

• Retrospective cross-sectional analysis of 173 patients evaluated at a Center for Rare Diseases (CRD).
• Symptom data (52 symptoms) aggregated into scores; laboratory data summarized into indices.
• Chi-Square Automatic Interaction Detection (CHAID) decision tree model constructed for risk stratification.

Main Results:

• 52.0% of patients received a confirmed rheumatologic diagnosis.
• Fatigue and generalized pain were common; the rheumatologic group showed higher symptom and immunoserological scores (p < 0.05).
• The CHAID model achieved 81.5% classification accuracy (AUC=0.893), outperforming logistic regression (76.3%, AUC=0.823).

Conclusions:

• The decision tree model offers a transparent framework for risk stratification in a specialized referral population.
• Findings are hypothesis-generating, requiring external validation for clinical implementation.
• The approach demonstrates the potential of rule-based symptom aggregation to support clinical reasoning in complex cases.