Diagnostic performance of essential tremor criteria in electronic health records: a retrospective neurology cohort study

Background: The 2018 Movement Disorder Society criteria introduced stricter diagnostic requirements for essential tremor (ET) compared with the 1998 consensus; however, how electronic health record (EHR)-based ET diagnostic coding aligns with chart-adjudicated criteria in routine practice remains unclear. We evaluated the diagnostic performance of EHR-based ET coding relative to chart-adjudicated criteria in a US neurology practice. Methods: We conducted a retrospective study within the Geisinger Health System in Pennsylvania. Patients with neurologist-assigned ET diagnoses were identified and stratified as ET-only or ET with differential diagnostic coding. A stratified random sample underwent manual chart review. Clinical features were mapped to both criteria after one and two visits. Accuracy metrics-sensitivity, specificity, sample-based positive predictive value (PPV), and sample-based negative predictive value (NPV)-were calculated with 95% confidence intervals. Results: The reviewed sample included 447 ET-only cases and 137 with differential. With one visit, the 1998 criteria yielded sensitivity of 96% (94-98%), specificity of 60% (53-67%), PPV of 82% (78-86%), and NPV of 88% (82-94%). The 2018 criteria demonstrated similar sensitivity (96%, 94-98%) but lower specificity (38%, 32-44%) and PPV of 53% (48-58%), with NPV of 93% (88-98%). With ≥2 visits, specificity increased to 71% (61-81%) and PPV to 85% (80-90%) for the 1998 criteria, and to 51% (43-59%) and 65% (57-71%) for the 2018 criteria. Conclusion: In this cohort, ET diagnostic coding showed consistently high sensitivity and NPV across frameworks. Specificity and PPV differed by criterion and improved with multiple clinical encounters.