Endocrine Abstracts

Background: Cranial diabetes insipidus (DI) is characterised by the inability to produce vasopressin leading to uncontrolled diuresis. Management includes administering synthetic vasopressin analogue desmopressin (DDAVP). Recently, there have been several national reports of DDAVP omission causing serious patient harm. This study aims to evaluate the feasibility of an automated alert system using Natural Language Processing (NLP) in electronic health records (EHR) to detect DI cases in a large tertiary hospital.

Methods: Retrospective analysis of data (February 1st-28th, 2022) of an automated search using Cogstack NLP for the following words “DDAVP”, “desmopressin” and “insipidus” in patients’ EHR, to generate daily alerts sent to a dedicated email inbox. We included all adult inpatients (≥18 years).

Results: 97 alerts were detected corresponding to 41 patients. On average, 2.7 alerts where generated each day. 16 of them (8 patients) met the inclusion criteria. No patients experienced adverse outcomes secondary to inappropriate DI management. The endocrinology team was aware or involved in 6 of the 8 cases (75%). In 43 alerts, DDAVP was used for other indications. 34 were paediatric patients. We were unable to identify what word triggered 4 alerts.

Conclusions: This preliminary study shows promising potential for the use of NLP to help identify DI inpatients in clinical practice. If used as a real-time alert system, it would alert the inpatient team to the 25% of DI inpatients currently unidentified. Given the high risk of deterioration if these patients are inappropriately managed, it is crucial for them to be referred to the specialist team early in the admission. Further work is required to refine the alert system to facilitate its implementation in a clinical setting.