Using procedural codes to supplement risk adjustment: a nonparametric learning approach.

Area of Science:

• Surgical Quality Improvement
• Artificial Intelligence in Healthcare
• Machine Learning for Risk Prediction

Background:

• The American College of Surgeons National Surgical Quality Improvement Program (NSQIP) utilizes work relative value unit (RVU) and Current Procedural Terminology (CPT) codes.
• Existing methods for assessing perioperative risk using CPT codes have limitations.

Purpose of the Study:

• To develop and evaluate a fully automated nonparametric learning approach to map individual CPT codes to perioperative risk.
• To assess the performance of this approach against traditional methods like work RVU and standard CPT categories.

Main Methods:

• Support vector machines (SVMs) were developed using NSQIP participant use file data (2005-2006) to learn the relationship between CPT codes and 30-day mortality/morbidity.
• SVMs were validated on data from 2007-2008.
• Performance was evaluated using areas under the receiver operating characteristic curve (AUROCs) and compared to work RVU and standard CPT categories.

Main Results:

• SVM operation scores achieved higher AUROCs for mortality (0.798-0.822) and morbidity (0.745-0.758) compared to work RVU and standard CPT categories (p < 0.001).
• Multivariable models incorporating SVM scores showed improved AUROCs for 30-day mortality and morbidity.
• The improvement in AUROCs was statistically significant for morbidity, though not for mortality.

Conclusions:

• Nonparametric artificial intelligence methods can effectively translate CPT codes for perioperative risk assessment.
• This automated approach complements existing risk adjustment models, such as those used in NSQIP.
• The findings support the integration of AI-driven CPT code analysis for enhanced surgical risk prediction.