Highly accurate and explainable detection of specimen mix-up using a machine learning model.

Area of Science:

• Clinical Chemistry
• Medical Diagnostics
• Machine Learning in Healthcare

Background:

• Delta check is a common method for identifying specimen mix-ups in laboratory testing.
• Current delta check methods suffer from low positive predictive value (PPV) due to low incidence rates and high false alert rates.
• This necessitates labor-intensive manual review to confirm true errors.

Purpose of the Study:

• To develop and evaluate an accurate machine learning model for detecting specimen mix-ups.
• To overcome the limitations of traditional delta check methods, specifically low PPV and high false alert rates.
• To improve patient safety through more effective mix-up detection.

Main Methods:

• A gradient-boosting-decision-tree (GBDT) model was developed using historical laboratory data.
• The model utilized partial time-series data of 15 common complete blood cell count and biochemical test items with a sliding window approach.
• Artificial mix-up cases were generated by shuffling data, and the GBDT model was trained to identify these artificial results.

Main Results:

• The GBDT model achieved a highly accurate performance, with an area under the receiver operating characteristic curve (ROC AUC) of 0.9983.
• The model demonstrated superior effectiveness in distinguishing true mix-ups from false alerts compared to conventional methods.
• The performance was validated on a separate dataset, confirming its robustness.

Conclusions:

• The developed GBDT model is highly effective for detecting specimen mix-ups.
• This improved accuracy facilitates more efficient and centralized mix-up detection in clinical laboratories.
• Implementation of this model can lead to significant improvements in patient safety by reducing errors.