Natural Language Processing CAM Algorithm Improves Delirium Detection Compared With Conventional Methods.

Area of Science:

• Medical Informatics
• Geriatrics
• Critical Care Medicine

Background:

• Delirium is frequently underdiagnosed and underdocumented in clinical settings.
• Current detection relies on clinician diagnosis or nursing notes, which may be insufficient.
• Electronic health records (EHRs) offer a potential data source for improved detection.

Purpose of the Study:

• To evaluate the effectiveness of a natural language processing-Confusion Assessment Method (NLP-CAM) algorithm for delirium detection.
• To compare NLP-CAM performance against traditional methods like clinician diagnosis and nursing documentation.
• To assess factors influencing delirium detection across different modalities.

Main Methods:

• A multicenter retrospective study involving 4351 hospitalized COVID-19 patient records.
• Utilized three delirium detection methods: clinician diagnosis, nursing documentation, and the NLP-CAM algorithm.
• Defined delirium occurrence as detection by any of the three methods for comparison.

Main Results:

• NLP-CAM algorithm detected 80% of delirium cases, significantly higher than clinician diagnosis (55%) and nursing documentation (43%).
• Increased odds of delirium detection were associated with older age, higher Charlson Comorbidity Score, and longer hospitalization.
• The NLP-CAM algorithm demonstrated superior performance in identifying delirium from EHR data.

Conclusions:

• The AI-based NLP-CAM algorithm substantially improves delirium detection compared to conventional methods.
• NLP-CAM shows significant promise as a diagnostic tool for delirium in electronic health records.
• Optimizing delirium detection through advanced algorithms is crucial for patient care.