Natural Language Processing Framework in Early Detection of Amyloidosis: The ALARM Study

Area of Science:

• Medical Informatics
• Computational Medicine
• Cardiology

Background:

• Early diagnosis of systemic amyloidosis (SA) is crucial for better patient outcomes.
• Current diagnostic challenges stem from SA's varied, nonspecific, and multisystemic presentation.

Purpose of the Study:

• To develop and validate natural language processing (NLP) models for predicting systemic amyloidosis (SA) diagnosis.
• To leverage electronic health record (EHR) data for early identification of SA.

Main Methods:

• Utilized NLP to construct predictive models for SA diagnosis using unstructured EHR data.
• Developed models in a cohort of patients with heart failure (HF) and/or neuropathy.
• Validated models in an independent cohort of HF patients.

Main Results:

• The best NLP model, gradient boosted regression, achieved an AUC of 0.88 in the development cohort and 0.85 in the validation cohort.
• The model demonstrated high specificity (86.0% development, 94.0% validation) in identifying SA.
• Achieved sensitivity of 73.5% (development) and 54.1% (validation), with strong negative predictive values.

Conclusions:

• NLP-based prediction models exhibit excellent diagnostic performance for SA in patients with HF and/or neuropathy.
• An NLP-enabled workflow can facilitate earlier SA diagnosis and treatment initiation.
• Early intervention has the potential to improve patient outcomes in systemic amyloidosis.