Optimising insulinoma detection: Algorithm development and validation in a hospital discharge abstract database in China

Background: The validity of using hospital discharge abstract database (DAD) diagnostic codes to identify insulinoma remains unverified. Objective: The study aimed to develop case-finding algorithms to identify insulinomas using ICD-10 codes from hospital DAD, evaluate their performance and investigate causes of misidentification. Method: This study utilised a 12-year retrospective dataset from a large medical centre in China to test four ICD algorithms for identifying insulinoma patients with hospital DAD. Algorithm performance was evaluated against electronic medical records, using sensitivity, specificity and predictive values as metrics. Results: This study involved 4929 patients with pancreatic tumours, including 610 insulinoma patients, resulting in 5760 hospitalisations. Algorithm variant 1, which uses the code M8151, achieved a sensitivity of 69% and a specificity of 99.7%. The most comprehensive algorithm, variant 4, demonstrated a sensitivity of 84.8% and a specificity of 99.5%. Both algorithm variants 1 and 4 showed higher sensitivity in patients admitted to the endocrinology department and in those lacking pathological evidence (p < 0.001). The use of “pancreatic neuroendocrine tumour” instead of “insulinoma” in pathology reports and inadequate documentation of hypoglycaemia were the main contributors to coding inaccuracies. Conclusion: Owing to the complexity and variability of insulinoma documentation, a single morphology code was not ideal for identifying insulinoma in the hospital DAD. Enhanced, comprehensive algorithms are expected to better recognise confirmed cases. Implications for health information management practice: A multidisciplinary approach with improved pathology reporting, hypoglycaemia statements and precise tumour site-specific coding is crucial for improving the sensitivity of coded data in the hospital DAD for the identification of insulinoma.