013  Validation of an algorithm identifying epilepsy related hospital admissions from routinely collected administrative data. Issue 6 (27th May 2022)

• Record Type: Journal Article
• Authors: Mitchell, James; Zammit, Louise; Kallis, Constantinos; Dixon, Pete; Grainger, Ruth; Powell, Graham; Marson, Tony
• Is Part Of: Journal of neurology, neurosurgery and psychiatry. Volume 93:Issue 6(2022)
• Page Start: A105
• Page End: A105
• Publication Date: 2022-05-27
• Subjects: Neurology -- Periodicals; Nervous system -- Surgery -- Periodicals; Psychiatry -- Periodicals; 616.8
• DOI: 10.1136/jnnp-2022-ABN.338
• Languages: English
• ISSNs: 0022-3050

Abstract

Abstract : Routinely collected administrative datasets can be used to identify healthcare resource use and inform strategic planning and policy development. We evaluate a new comprehensive epilepsy admission algorithm using ICD-10 coding data from these administrative datasets against a gold standard and existing admission identification algorithms. All consecutive admissions to a large UK general hospital meeting predefined criteria were identified over a one-month period. Case note review was performed by independent clinicians and admission classi- fied as epilepsy or not epilepsy related. The Liverpool algorithm and a widely used comparator algorithm were applied to the ICD-10 administrative data and compared to clinician classification. In total 205 admissions were identified. Case-note review (gold standard) identified that 35 cases were epilepsy related admissions, and 156 were not epilepsy related. It was not possible to classify 14 admis- sions due to missing or insufficient data. The Liverpool algorithm incorrectly classified only 17 out of 191 admissions compared to 30 incorrectly classified admissions from a comparator algorithm. The Liverpool algorithm demonstrates sensitivity of 85.7% and specificity of 92.3%. Whilst no perfect algorithm is likely to exist, the Liverpool algorithm identifies hospital admissions due to epileptic seizures using administrative data with improved sensitivity and specificity compared to previous methods. [e-mail]