Mapping ICD-10 to ICCC-3 for Childhood Cancer Classification Using Cancer Incidence in Five Continents (CI5) Data

Technical Summary: Mapping ICD-10 to ICCC-3 for Childhood Cancer Classification

Problem Statement
Global estimation of the childhood and adolescent cancer burden is hindered by inconsistent diagnostic classification systems. While the International Classification of Childhood Cancer (ICCC-3) serves as the benchmark for pediatric cancer research and epidemiological studies (e.g., the International Incidence of Childhood Cancer project), major global databases such as GLOBOCAN and the Global Burden of Disease (GBD) rely primarily on the International Classification of Diseases, 10th Revision (ICD-10). ICD-10 lacks the morphological granularity required for ICCC-3 classification, and no systematic framework currently exists to map ICD-10 codes back to ICCC-3 categories. This gap prevents accurate cross-database comparison, weakens policy utility, and risks misclassification in international health assessments. Existing conversion rules typically map forward (ICCC-3/ICD-O to ICD-10), leaving the inverse mapping unavailable.

Methodology
The study utilized high-quality, population-based pediatric cancer incidence data (ages 0–19) from the United States national registry within Cancer Incidence in Five Continents (CI5) Volume XII, covering diagnoses from 2013 to 2017. This dataset was selected for its dual coding of cases in both ICD-10 and ICCC-3, alongside ICD-O-3 morphology codes.

The authors developed a three-tier mapping strategy to align ICD-10 topography codes with ICCC-3 diagnostic groups:

1. Direct Mapping: Anatomical and clinical correspondence was used to map ICD-10 codes directly to the 12 main ICCC-3 diagnostic groups. Unmapped or ambiguous codes were assigned to "Other Malignant Epithelial Neoplasms and Malignant Melanoma" (Group XI) but counted toward total cases.
2. ICD-O-3 Mediated Mapping: For lymphoid and hematopoietic neoplasms (ICCC-3 Groups I and II), an intermediary crosswalk was constructed. The authors merged the SEER ICCC Recode table (linking ICD-O-3 to ICCC-3) with CI5 conversion tables (linking ICD-O-3 to ICD-10). This allowed for subgroup-level mapping based on morphology codes, utilizing the 2008 WHO classification alignment.
3. Manual Mapping: For remaining subgroups, diagnostic descriptors within ICD-10 were used to assign cases to the most likely ICCC-3 subgroup where feasible (e.g., mapping C22.2 specifically to hepatoblastoma).

To evaluate alignment, a Consistency Ratio (CR) was calculated as the ratio of cases classified under ICD-10 to those under ICCC-3 for each main group. A CR of 1.00 indicates perfect concordance; CR > 1 indicates over-mapping (ICD-10 counts exceed ICCC-3), and CR < 1 indicates under-mapping.

Key Results
The analysis covered 77,236 cases coded in ICD-10 and 77,152 cases in ICCC-3.

• High Concordance: Near-perfect agreement (CR ≈ 1.00) was observed for retinoblastoma, brain/CNS tumors, renal tumors, and hepatic tumors.
• Over-Mapping (CR > 1): Significant over-mapping occurred for epithelial neoplasms/melanomas (CR = 1.24), bone tumors (CR = 1.11), and leukemias (CR = 1.07). This is largely attributed to ICD-10's anatomical coding capturing metastatic disease (e.g., neuroblastoma metastases to bone) that ICCC-3 excludes from primary bone tumor groups.
• Under-Mapping (CR < 1): Substantial under-mapping was noted for neuroblastomas (CR = 0.67), germ cell tumors (CR = 0.72), soft tissue sarcomas (CR = 0.87), lymphomas (CR = 0.88), and unspecified neoplasms (CR = 0.47). For lymphomas, cases arising in specific anatomical sites (GI tract, CNS) were often classified into site-specific groups in ICD-10 rather than the lymphoma group.
• Subgroup Mapping:
  • Lymphoid/Hematopoietic: The ICD-O-3 mediated strategy enabled precise subgroup mapping for leukemias and lymphomas, resolving ambiguities between leukemia and lymphoma classifications (e.g., T-cell lymphoblastic leukemia/lymphoma).
  • Solid Tumors: Precise one-to-one mapping at the subgroup level was generally not feasible due to ICD-10's lack of histological detail, with the notable exception of hepatoblastoma (C22.2 mapped to ICCC-3 Group VIIa1).

Key Contributions

• First Systematic Mapping: This study provides the first systematic attempt to map ICD-10-coded pediatric cancer cases to ICCC-3 categories, addressing a critical gap in cancer registry data interoperability.
• ICD-O-3 Crosswalk: It establishes a robust, ICD-O-3-mediated crosswalk specifically for lymphoid and hematopoietic neoplasms, allowing for subgroup-level classification that is otherwise impossible with ICD-10 alone.
• Quantification of Discrepancies: The study quantifies the extent of misclassification between the two systems, highlighting where ICD-10 anatomical coding diverges from ICCC-3 histological criteria (e.g., bone tumors and lymphomas).

Significance and Claims
The authors claim that this mapping framework offers a practical tool for standardizing pediatric cancer data across different coding systems. Its primary utility lies in facilitating cross-database comparisons between ICD-10-based global estimates (GBD, GLOBOCAN) and ICCC-3-based research datasets.

The study emphasizes that while the mapping is reliable for specific, highly distinct tumor types (retinoblastoma, CNS, renal, hepatic) and for lymphoid/hematopoietic subgroups (via the ICD-O-3 bridge), it remains limited for other solid tumors. The authors explicitly state that the lack of morphological detail in ICD-10 precludes precise subgroup classification for most solid tumors. Consequently, the framework is presented as a necessary step to enable resource-limited regions to analyze childhood cancer subtypes and to support the development of a global ICCC-3-based surveillance system, rather than as a perfect solution for all tumor types.

Limitations Noted
The authors acknowledge that the study relies on aggregate population-level data, preventing the calculation of individual-level agreement metrics (e.g., Cohen's kappa). The mapping is specific to the ICD-10 (2010), ICD-O-3 (1st edition), and ICCC-3 (WHO 2008) versions used in the CI5 XII dataset, and caution is advised when applying these rules to other coding versions or time periods. Additionally, rare tumor types are more susceptible to proportional errors from misclassification.