A Novel Automated Algorithm to Identify Lung Cancer Screening from Free Text of Radiology Orders.

Area of Science:

• Medical Informatics
• Radiology
• Public Health

Background:

• Lung cancer screening (LCS) is recommended for asymptomatic individuals.
• Administrative codes used for LCS may inaccurately include tests for symptomatic patients.
• Accurate identification of screening tests is crucial for unbiased research.

Purpose of the Study:

• To validate an automated algorithm for identifying lung cancer screening (LCS) among asymptomatic patients.
• To differentiate between screening and diagnostic low-dose chest CT scans using electronic health data.
• To improve the accuracy of identifying LCS in large patient populations.

Main Methods:

• Developed and iteratively refined an algorithm using Current Procedural Terminology (CPT) codes and free text from radiology orders.
• Analyzed a national population-based sample of 4503 adults aged 65-80 within the Veterans Health Affairs primary care system.
• Evaluated algorithm performance (sensitivity, specificity, PPV, NPV) against manual chart review as the gold standard.

Main Results:

• The algorithm achieved 97% sensitivity and 79% specificity in identifying LCS-eligible individuals.
• Administrative codes misclassified 31% of LCS scans, while the algorithm correctly identified 95% of screening scans.
• The algorithm demonstrated high positive predictive value (90%) and negative predictive value (93%) in the overall population.

Conclusions:

• An automated algorithm can accurately distinguish screening from diagnostic chest imaging, essential for unbiased non-randomized LCS studies.
• This validated algorithm enhances the reliability of analyzing LCS outcomes.
• Further research should evaluate the accuracy of administrative codes for LCS in diverse healthcare systems.