Identifying mislabelled samples: Machine learning models exceed human performance.

Area of Science:

• Clinical laboratory science
• Medical informatics
• Artificial intelligence in healthcare

Background:

• Identifying mislabelled patient samples is a critical challenge in clinical laboratories.
• Current methods rely on delta checks and manual review, which can be error-prone.
• Machine learning (ML) has shown potential to improve error detection over traditional methods.

Purpose of the Study:

• To compare the performance of various machine learning models against human laboratory staff in identifying mislabelled patient samples.
• To evaluate the efficacy of ML in detecting sample mislabelling errors.

Main Methods:

• Eight ML models (artificial neural network, gradient boosting, SVM, random forest, logistic regression, k-NN, decision trees) were trained on 127,256 electrolyte, urea, and creatinine result sets.
• A separate test set of 14,140 results was used for evaluation.
• The performance of ML models was compared to 500 manual reviews by laboratory staff volunteers.

Main Results:

• The artificial neural network achieved the highest accuracy at 92.1%.
• The simplest decision tree model had the lowest accuracy at 86.5%.
• Laboratory staff achieved an accuracy of 77.8% in identifying mislabelled samples.

Conclusions:

• Even basic machine learning models can surpass human performance in detecting mislabelled samples.
• Machine learning offers a viable and superior alternative to manual review for sample mislabelling.
• Implementation of ML techniques in clinical labs is recommended to enhance error identification and patient safety.