Network-Based Analysis of Cancer Mortality Misclassification Reveals Signatures of Metastatic Pathways: Insights into Breast-To-Lung and Other Pairs of Anatomical Sites

Background: Accurate cancer Cause-of-Death (CoD) data are fundamental for public health, but misclassification can bias mortality profiles, and existing correction methods fail to capture patient-level diagnostic trajectories. We propose a network-based framework to characterize CoD misclassification linked to metastatic spread. Methods: We conducted a retrospective population-based study in Ceará, Brazil (2014-2019), linking mortality and cancer registry records. After filtering, 11,226 histologically confirmed cancer deaths were analyzed. We constructed a directed weighted Inconsistency Network (IN) between Last Diagnosis (LD) and CoD. Misclassification probabilities were compared with autopsy-derived metastasis probabilities (1,922 records), and effective mortality rates accounted for misclassification flows. Findings: Overall, 22\% of deaths showed no agreement between LD and CoD. Misclassification was not random: lung cancer was a major category into which deaths were incorrectly assigned in both sexes, whereas esophageal cancer showed this pattern only in males. The INs indicate reduced lung cancer mortality and increased stomach cancer mortality. In males, esophageal cancer mortality should also be reduced. In females, reduced lung cancer mortality is largely driven by breast-to-lung inconsistencies. This is compatible with autopsy evidence, suggesting that many deaths recorded as lung cancer should have been attributed to metastatic breast cancer. Interpretation: Cancer mortality misclassification appears to follow plausible metastatic pathways. Pulmonary metastasis recorded as primary tumours may partly inflate lung cancer mortality while underestimating deaths from other primary tumours, notably breast cancer. This may bias surveillance, resource allocation, and cancer control policies. The framework offers a scalable approach to improve mortality estimates.