Automatic ICD-10 classification of cancers from free-text death certificates

Area of Science:

• Medical Informatics
• Public Health Surveillance
• Computational Pathology

Background:

• Death certificates are crucial for cancer mortality statistics but contain unstructured natural language data.
• Extracting accurate, quantitative cancer data from death certificates is challenging due to data volume and variability.
• Automated systems are needed to efficiently process and analyze free-text death certificate information.

Purpose of the Study:

• To propose and evaluate an automatic classification system for identifying cancer-related causes of death from death certificates.
• To enable accurate and timely monitoring and reporting of cancer mortality by organizations like Cancer Registries.
• To demonstrate the general applicability of the developed methods to other medical text analysis tasks.

Main Methods:

• Utilized Support Vector Machine (SVM) classifiers trained on features extracted from 447,336 death certificates.
• Employed a cascaded architecture: first, binary cancer/no-cancer identification, then cancer type classification using ICD-10 codes.
• Features included terms, n-grams, and SNOMED CT concepts; performance evaluated using precision, recall, and F-measure on a held-out test set.

Main Results:

• The system achieved high effectiveness in identifying cancer as the underlying cause of death (F-measure 0.94).
• Accurate classification of common cancer types was achieved (F-measure 0.7), though rare cancers were challenging (F-measure 0.12).
• Feature analysis highlighted the importance of SNOMED CT concepts and oncology-specific morphology features for classification.

Conclusions:

• The developed system automates the identification and characterization of cancers from free-text death certificates.
• This facilitates timely and accurate cancer mortality monitoring and reporting for public health organizations.
• The methodology is adaptable for analyzing other medical texts and for classifying different types of diseases.