Assessing data availability and quality within an electronic health record system through external validation against...

Area of Science:

• Health Informatics
• Clinical Data Mining
• Public Health

Background:

• Smoking causes 20% of US deaths annually, but EHR data capture is insufficient for improving outcomes.
• Existing algorithms for extracting smoking data from clinical notes lack external validation.
• Current electronic health record (EHR) practices do not effectively translate smoking risk information into improved patient health outcomes.

Purpose of the Study:

• To evaluate the clinical implementation performance of an informatics pipeline designed to extract smoking behaviors from EHR notes.
• To assess the accuracy of the pipeline in identifying smoking status, pack year history, and lung cancer screening eligibility.

Main Methods:

• Developed and implemented an informatics pipeline to extract smoking status, pack year history, and cessation date from clinical notes.
• Validated the pipeline's performance using 1,504 clinical notes matched against an external questionnaire.
• Assessed lung cancer screening eligibility based on extracted smoking data.

Main Results:

• 73% of notes lacked smoking behavior information.
• Weighted Cohen's kappa for smoking status was 0.62 (95% CI 0.56-0.69).
• Correlation for pack years was 0.39; lung cancer screening eligibility had a positive predictive value of 85.4% and specificity of 94.7%.

Conclusions:

• The informatics pipeline reliably extracts smoking behaviors from EHR notes when present.
• Extracted smoking data is sufficiently reliable for identifying patients eligible for smoking-related services and interventions.
• Prioritizing smoking information capture in clinical encounters is crucial for public health.