ASPO Abstracts
Comparison of Electronic Medical Records with Self-report in Measuring Cancer Screening
Category: Early Detection & Risk Prediction
Conference Year: 2021
Abstract Body:
Purpose: This population-based study in Indiana collected information at the patient level about colon,
cervical, and breast cancer screening from both survey self-report (SR) and statewide repository of
electronic medical records (EMRs), enabling comparison of the two data sources when measuring the
same outcome.
Methods: Out of 970 patients who completed the survey, a total of 711 patients provided HIPAA
authorization (73.3%) providing access to their EMRs. Types of cancer screening tests assessed included
colorectal cancer (colonoscopy and fecal immunochemical test), cervical cancer (human papilloma virus
and Pap test), and breast cancer (mammogram). For each cancer screening test, we measured both
receipt of the screening and time since last screening to evaluate the proportion of
agreement/disagreement between measures from survey SR and EMR. Concordance of EMR data with
the survey SR was calculated using Cohen/Conger's Kappa (κ) and the Gwet's agreement coefficient.
Results: Percent agreement of EMR and SR of receipt of screening ranged from 59%-81%, the Kappa
coefficient ranged from 0.17-0.26 and the Gwet's coefficient ranged from 0.24-0.76. For time since receipt
of the last screening test, the Kappa coefficient ranged from 0.13-0.39 and the Gwet's coefficient ranged
from 0.38-0.86. In comparing the proportion of all patients where EMR data indicated screening (but SR
did not) versus SR indicated screening (but EMR did not), the following patterns emerged: colonoscopy
(7% EMR alone vs. 34% SR alone), FIT test (15% vs. 4%), HPV test (24% vs. 12%), Pap test (10% vs.
27%), mammography (8% vs. 20%). EMR data provided relatively more additional information about FIT
and HPV tests, both laboratory tests, and less additional information about colonoscopy, Pap test, or
mammography, all procedures.
Conclusion: The information value added by combining different data sources, about the receipt of cancer
screening, varied by the type of cancer screening. Studies that use a single data-source should consider
the type of cancer screening test to choose the best data collection method. EMR and SR both provided
unique information in measuring cancer screening, and the most robust research approach involves
collecting screening information from both EMR and patient SR.
Keywords: early detection, cancer screening, survey, electronic medical records