Reliability of basic cancer patient data.

Pooling patient data from multi-institutional medical chart review occurs commonly in the study of cancer and other diseases. To determine the consistency of reporting among institutions, we presented a test set of 25 standardized medical charts to coders at 18 Comprehensive Cancer Centers and compared their resulting codes. This study measures the reproducibility of coding by different persons, but does not assess the accuracy of the underlying medical record. Among 34 data items, we found high disagreement rates in coding stage of disease (14 per cent) and date of diagnosis (8 per cent). Primary site, histologic type and other key items had good reproducibility (disagreement rates less than or equal to 5 per cent). A number of minor disagreements indicated that detailed distinctions could not be reliably coded from medical charts.