Automated Detection of Periprosthetic Joint Infections and Data Elements Using Natural Language Processing

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Data Analysis

Background:

• Periprosthetic joint infection (PJI) data extraction from electronic health records is labor-intensive, relying on manual chart review.
• Current methods for collecting PJI data are time-consuming and prone to human error, impacting patient care and research.
• Developing automated methods for PJI data collection is crucial for efficient healthcare management.

Purpose of the Study:

• To develop and validate a natural language processing (NLP) algorithm for automated PJI data element extraction.
• To replicate the accuracy of manual chart review for PJI diagnosis using NLP.
• To assess the performance of NLP algorithms in identifying key PJI indicators.

Main Methods:

• A retrospective study was conducted on total joint arthroplasty (TJA) procedures performed between 2000 and 2017.
• Musculoskeletal Infection Society (MSIS) criteria were used as the gold standard for PJI diagnosis.
• NLP algorithms were trained and tested on consultation notes, operative reports, pathology, and microbiology reports from a large cohort of TJA patients.

Main Results:

• The NLP algorithm achieved a high overall f1-score of 0.911 in identifying PJI based on MSIS criteria.
• Specific PJI indicators, including sinus tract, purulence, and microbial growth, were extracted with f1-scores ranging from 0.771 to 0.982.
• The algorithm demonstrated high sensitivity and specificity for extracting individual PJI diagnostic elements.

Conclusions:

• NLP-enabled algorithms can automate the collection of PJI diagnostic data, significantly improving efficiency.
• Automated data collection has the potential to enhance patient care, surveillance, and clinical research in PJI.
• Further validation of these NLP algorithms in diverse healthcare settings is warranted.