Using machine learning to detect noncredible cognitive test performance.

Area of Science:

• Neuropsychology
• Artificial Intelligence
• Data Science

Background:

• Assessing performance validity in neuropsychological testing is crucial.
• Advanced algorithmic methods, including machine learning (ML), may enhance validity assessments.

Purpose of the Study:

• To investigate the utility of unsupervised ML in assessing performance validity during neuropsychological evaluations.
• To determine if ML can accurately cluster patient data into valid and invalid performance groups.

Main Methods:

• An unsupervised ML model analyzed data from 359 adult outpatients.
• Data included performance validity test scores, medical/psychiatric history, referral reason, litigation, and disability status.
• The model's clusters were compared against established validity ratings.

Main Results:

• The ML model identified two clusters: valid and invalid performance.
• The model achieved excellent predictive accuracy (AUC = .92).
• Performance validity test scores were most influential, followed by medical history, referral reason, and disability status.

Conclusions:

• Unsupervised ML can accurately assess performance validity using diverse neuropsychological evaluation data.
• This approach may overcome limitations of traditional validity assessment methods.
• Unsupervised ML is adaptable to digital technologies, potentially improving future validity assessments.