Integrating NLP to Enhance Algorithmic Identification of Metastatic and Castration-Resistant Prostate Cancer in Large...

Area of Science:

• Oncology
• Medical Informatics
• Health Services Research

Background:

• Accurate classification of prostate cancer (PC) disease states, specifically metastasis and castration resistance (CRPC), is crucial for population-based research.
• Current chart review methods are not feasible for large-scale studies, necessitating the development of accurate automated classification methods.

Purpose of the Study:

• To evaluate the accuracy of automated algorithms for identifying CRPC and metastatic PC in a large patient cohort.
• To determine the effectiveness of integrating natural language processing (NLP) and treatment patterns with ICD codes for improved classification.

Main Methods:

• A retrospective study was conducted using Veterans Affairs Health Care System data.
• Algorithms for CRPC and metastatic PC were evaluated against manual chart review as the gold standard.
• Methods included ICD codes, NLP, and a novel algorithm combining these with treatment patterns.

Main Results:

• The CRPC algorithm achieved 85.1% sensitivity and 96.1% specificity.
• For metastatic disease, the combined ICD codes, treatment patterns, and NLP algorithm showed the highest sensitivity (94.4%) and specificity (93.0%).
• NLP integration enhanced metastatic classification sensitivity with minimal impact on specificity.

Conclusions:

• Developed CRPC and metastasis algorithms are effective automated tools for identifying advanced PC states in large populations.
• Integrating NLP into classification algorithms significantly improves sensitivity for metastatic disease detection.
• A multifaceted approach combining data sources like ICD codes, treatment patterns, and NLP is valuable for large-scale PC research.