Accuracy of basic cancer patient data: results from an extensive recoding survey

Area of Science:

• Medical Informatics
• Oncology Data Management
• Clinical Data Accuracy

Background:

• Accurate coding of patient data from medical records is essential for reliable cancer research.
• Previous assessments of data coding accuracy in multi-center studies are limited.
• Variability in coding practices can impact the validity of research findings.

Purpose of the Study:

• To assess the accuracy of coded data from medical records across multiple U.S. cancer centers.
• To identify specific data items with high disagreement rates.
• To evaluate the causes of coding disagreements and propose improvements.

Main Methods:

• Recoding of 29 data items (demographics, diagnosis, therapy) from 985 cancer patient records.
• Comparison of original codes with recoded data to classify disagreements as major or minor.
• Analysis of disagreement rates across different data items, sites, and centers.

Main Results:

• The highest major disagreement rates were observed for stage of disease (23%), histology (10%), and site (7%).
• Most disagreements (over 97%) resulted from errors in record use, not interpretation.
• Significant variability in major disagreement rates was noted across sites (10-fold), centers (4-fold), and disease stage (2-fold).
• Overuse of the 'unknown' code contributed to disagreements.

Conclusions:

• Coding accuracy in cancer patient records varies significantly, with substantial errors impacting key variables like disease stage.
• The findings underscore the need for standardized coding procedures and targeted training to enhance data quality in cancer research.
• A new procedure for analyzing disagreement rates is presented to guide efforts for improving coding accuracy.