Predictive Capability of Historical Data for Diagnosis of Dizziness

Abstract

Objective: The study examines categorical responses to questions on a comprehensive dizziness questionnaire, to find the overall predictive power of the questionnaire, and to identify which question(s) are most predictive of each diagnosis. Study Design: Retrospective chart review. Setting: Specialized dizziness and balance center at a tertiary care hospital. Patients: A total of 619 patients (aged 19-89 yr, of whom 60% are women and 40% are men) diagnosed with 1 of 23 types of dizziness or postural instability. Intervention: All patients were administered a standard 163-item dizziness questionnaire (including 77 review of systems items). Outcome Measures: Predicted diagnoses from the questionnaire, as determined by binary and multinomial logistic regressions, are compared with an ultimate clinical diagnosis made by an expert neurotologist based on full interview, examination, and clinical tests. Results: Significant question groupings exist for each of the main diagnoses. A subset of 47 questions under multinomial logistic regression gave high predictive accuracies for migraine (92%), benign paroxysmal positional vertigo (90%) and Ménière's disease (86%), and fair predictive power for vestibular neuritis (63%), contributing to an overall predictive accuracy of 84%. A smaller subset of 32 questions gave an overall predictive accuracy of 71%. Conclusion: The capability of historical data to accurately predict the ultimate diagnosis for dizziness emphasizes the importance of a structured questionnaire in the evaluation of such patients. Future developments include the formulation of a computer-based program to generate a differential diagnosis for the practitioner to consider.