Delirium detection using GAMMA wave and machine learning: A pilot study.
Malissa Mulkey, Thomas Albanese, Sunghan Kim and 2 others
PMID 36321335WHAT IT FOUND
In 13 ventilated older adults, a handheld EEG headband classified delirium status with 87 to 97% accuracy using gamma wave ratios.
This is a small pilot, not a validated tool. Nurses should continue using standard delirium screening until cut points are established.
Key findings
01Thirteen of seventeen enrolled patients had usable EEG data for this analysis.
02Gamma/delta and gamma/theta power ratios were higher in participants with delirium, with means of 0.452 and 0.813 respectively, compared to 0.238 and 0.390 in those without.
03Machine learning models using these ratios achieved classification accuracies between 87% and 97% in this small sample.
STILL TO COME
How it was doneWhat they foundWhat it means for RNs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
The sample size was very small (13 patients analyzed from 17 enrolled), making the results preliminary and susceptible to chance. The study was conducted at a single academic medical center, limiting generalizability. Four patients were excluded due to EEG signal artifact, suggesting practical difficulties in obtaining clean data in a busy ICU. The design was correlational, so it cannot prove that EEG changes cause or precede delirium, only that they are associated with it. Overfitting was a significant risk in the machine learning models, particularly for SVM and random forest, despite cross-validation techniques. The study did not establish specific cut-off values for the EEG ratios that could be used for clinical decision-making.
Declared interests
The study was supported by the National Institutes of Health (N.I.H., Extramural). The paper does not declare any conflicts of interest with the device manufacturer (Ceribell Inc.).
The easy way to misread this
Do not interpret the 87 to 97% accuracy rates as evidence that this device is ready for clinical use. These figures come from a pilot study of only 13 patients with no independent validation set, and the authors explicitly state that nurses should continue using validated screening methods until well-demarcated cut points are established.