Performance of a Natural Language Processing Method to Extract Stone Composition From the Electronic Health Record

Area of Science:

• Nephrology
• Medical Informatics
• Computational Linguistics

Background:

• Kidney stone composition is crucial for understanding disease etiology and guiding treatment.
• Extracting this information from large electronic health records (EHRs) is challenging due to unstructured clinical notes.

Purpose of the Study:

• To develop and validate a natural language processing (NLP) algorithm for mining kidney stone composition from a large EHR repository.
• To demonstrate the utility of this NLP tool for epidemiological research in kidney stone disease.

Main Methods:

• Developed StoneX, a pattern-matching NLP algorithm trained on annotated kidney stone types (calcium oxalate monohydrate, calcium oxalate dihydrate, hydroxyapatite, brushite, uric acid, struvite).
• Applied StoneX to over 125 million clinical notes to identify patients with kidney stone composition data.
• Validated the algorithm's performance and used extracted data for analyses including stone type conversion, survival analysis, and disease associations.

Main Results:

• The NLP algorithm identified 11,585 patients with kidney stone composition data from >125 million notes.
• Achieved high positive predictive values (>90%) for most stone types, with 87.5% for uric acid stones.
• Identified statistically significant differences in survival from a second stone surgery among stone types (P=.03) and revealed several known and novel disease associations.

Conclusions:

• NLP extraction of kidney stone composition from large-scale EHRs is feasible and highly precise.
• This approach enables high-throughput epidemiological studies, advancing kidney stone disease research.
• StoneX provides a valuable tool for high-fidelity research using EHR data.