Assessing the accuracy of the International Classification of Disease (ICD) frameworkin the identification of patients with Chronic Subdural Haematoma from hospital records.

Abstract

Abstract Background Chronic subdural hematoma (CSDH) is one of the commonest neurosurgical pathologies with an increasing incidence. Observational studies of routine care have demonstrated high perioperative morbidity and approximately 10% mortality at one year. 20 The development, implementation, and evaluation of a potential care framework relies on an accurate and reproducible method of case identification and case ascertainment. With this manuscript, we report on the accuracy of diagnostic ICD codes for identifying patients with CSDH from retrospective electronic data and explore whether basic demographic data could improve the identification of CSDH. Methods: Data were collected retrospectively from the hospital administrative system between 2014–2018 of all patients coded with either S065 or I620. Analysis of the ICD codes in identifying patients with CSDH diagnosis was calculated using the caretR package in RStudioR .and stepwise logistic regression analysis was performed to evaluate the best predictive model for CSDH. Results: A total of 1861 patients were identified. Of these, 189 (10.2%) had a diagnosis of non-traumatic SDH (I620) and 1672 (89.8%) traumatic subdural haematomas (S065). Variables that identified CSDH as a diagnosis on univariate logistic regression included male sex (Odds Ratios (OR) − 1.606(1.197–2.161), elderly age (OR) − 1.023(1.015–1.032) per year for age (p < 0.001) and shorter length of hospital stay. Using stepwise regression against AIC the best model to predict CSDH included male sex, older age, and shorter LOS. The calculated sensitivity for identifying CSDH with the model is 88.4% with a specificity of 84.5% and PPV of 87.9%. Conclusion: CSDH is a common neurosurgical pathology with increasing incidence and ongoing unmet clinical need. We demonstrate that case ascertainment for research purposes can be improved with the incorporation of additional demographic data but at the expense of significant case exclusion.