Defining a Patient Population With Cirrhosis: An Automated Algorithm With Natural Language Processing.

Area of Science:

• Medical Informatics
• Clinical Data Analysis
• Natural Language Processing

Background:

• Manual identification of cirrhosis patients is time-consuming and ICD-9 codes can be inaccurate.
• Natural Language Processing (NLP) is a novel method for analyzing electronic health records.
• NLP has successfully identified other gastrointestinal conditions but not yet cirrhosis.

Purpose of the Study:

• To evaluate NLP as a supplement to ICD-9 codes and laboratory values.
• To develop an automated algorithm for defining and risk-stratifying cirrhosis patients.
• To improve the accuracy and efficiency of cirrhosis patient identification.

Main Methods:

• Retrospective cohort study at an academic medical center.
• Developed an algorithm using NLP of radiology reports, ICD-9 codes, and lab data.
• Validated the algorithm against a manual review of 168 patient charts.

Main Results:

• The combined algorithm achieved high performance metrics: PPV 91.78%, NPV 96.84%, sensitivity 95.71%, specificity 93.88%.
• The NLP component demonstrated superior performance: PPV 98.44%, NPV 93.27%, sensitivity 90.00%, specificity 98.98%.
• NLP was identified as the most critical element within the algorithm.

Conclusions:

• NLP is a valuable tool for identifying patients with cirrhosis.
• Combining NLP with administrative and laboratory data enhances patient identification accuracy.
• This approach offers a powerful method for population-level cirrhosis patient identification and risk stratification.