Machine learning-based delta check method for detecting misidentification errors in tumor marker tests.

Area of Science:

• Clinical Chemistry
• Bioinformatics
• Medical Diagnostics

Background:

• Tumor marker test misidentification errors pose significant risks to patient diagnosis and treatment.
• Conventional delta check methods have limitations in accurately detecting these errors.

Purpose of the Study:

• To develop and evaluate a machine learning (ML)-based delta check method for detecting sample misidentification in tumor marker tests.
• To compare the efficacy of ML models against traditional delta check approaches.

Main Methods:

• Analysis of 246,261 records for five tumor markers: AFP, CA19-9, CA125, CEA, and PSA.
• Development of ML models using random forest (RF) and deep neural network (DNN) algorithms.
• Comparison of ML models with delta percent change (DPC), absolute DPC (absDPC), and reference change values (RCV) using in silico simulations.

Main Results:

• The DNN model demonstrated superior balanced accuracies across all tested tumor markers (0.792-0.842).
• The DNN model outperformed RF, DPC, absDPC, and RCV in detecting misidentification errors.
• The RF model showed mixed performance, generally better than DPC and absDPC but comparable or lower than RCV.

Conclusions:

• ML-based delta check methods are more effective than conventional approaches for detecting tumor marker sample misidentification.
• The DNN model offers a robust and stable solution for improving the accuracy of tumor marker testing.