Machine Learning Approach to Identifying Wrong-Site Surgeries Using Centers for Medicare and Medicaid Services...

Area of Science:

• Medical informatics
• Machine learning in healthcare
• Surgical safety

Background:

• Wrong-site surgery (WSS) is a preventable medical error with severe patient consequences.
• Underreporting and documentation inconsistencies hinder WSS prevention efforts.
• Machine learning (ML) shows promise in identifying medical errors, including medication mistakes.

Purpose of the Study:

• To evaluate the effectiveness and transferability of an ML model for detecting surgical documentation inconsistencies.
• To specifically assess the model's ability to identify laterality discrepancies in surgical procedures and diagnoses.

Main Methods:

• Utilized Centers for Medicare and Medicaid Services Limited Data Set (CMS-LDS) claims data (2017-2020).
• Developed and adapted an Association Outlier Pattern (AOP) ML model to detect unusual procedure-diagnosis combinations, focusing on laterality.
• Trained the model on 2017-2019 data and tested on 2020 orthopedic procedures using ICD-10-PCS codes for laterality verification.

Main Results:

• Identified 2170 claims with significant laterality discrepancies out of 346,382 claims analyzed.
• Clinical review confirmed over 50% of identified discrepancies as actual errors (e.g., left procedure/right diagnosis).
• The AOP model flagged numerous potential errors with over 80% confirmed by clinical review, outperforming rule-based methods.

Conclusions:

• The AOP ML model effectively detects surgical procedure-diagnosis inconsistencies, particularly laterality errors, in claims data.
• The model demonstrates high accuracy and transferability, outperforming traditional rule-based approaches.
• This approach enhances patient safety by improving the identification of surgical errors and aiding clinical decision-making.