Natural Language Clinical Pathways for Automated Coding of Topography and Morphology in Cancer Registries: Leveraging
Adele Zanfino1, Carlotta Buzzoni2, Antonio Giampiero Russo1
1Epidemiology Unit, Agency for Health Protection of the Metropolitan Area of Milan (Italy).
Epidemiologia E Prevenzione
|April 23, 2026
Summary
This study developed an AI algorithm to automatically code cancer topography and morphology using clinical data. The AI shows promise for frequent cancer types, reducing manual coding time for cancer registries.
Area of Science:
- Oncology
- Medical Informatics
- Artificial Intelligence
Background:
- Cancer registries (CRs) are vital for cancer surveillance.
- Manual ICD-O-3 coding of topography and morphology is complex and time-consuming.
- Artificial Intelligence (AI) offers automation potential, exceeding current limitations.
Purpose of the Study:
- To develop an AI algorithm for automatic topography-morphology (topo-morpho) coding.
- To process natural language clinical pathways (LN-PDTA) for coding.
- To improve the efficiency of cancer registry operations.
Main Methods:
- Retrospective observational study using linked registry and administrative data.
- LSTM neural network trained on chronological clinical tokens from diagnoses, therapies, and procedures.
- Dataset included 34,168 incident cancer cases (2017-2018), excluding uncertain tumors.
Main Results:
- AI achieved 89% topography, 59% morphology, and 56% combined topo-morpho prediction accuracy on the test set.
- Higher accuracy for frequent sites: breast (73%), colorectal (61%), lung (94% topography), prostate (98% topography).
- Key data sources: pathology reports, mortality data, surgical procedures, and hospital diagnoses.
Conclusions:
- AI-based LN-PDTA approach shows promise for frequent cancer types.
- Enables automatic coding for a significant number of cases.
- Reduces manual coding workload and enhances cancer registry efficiency.
