Dr. Erkan Hassan

Monthly Writings

Evaluations and reviews of the latest in the field.

5 Ways to Address Vital Sign Alert Artifacts

SUMMARY:

  • Automated devices cannot differentiate artifact values from actual values.

  • Adequate tools to identify artifact values are lacking

  • Although several methodologies exist, no one method is found to be consistently superior.

REVIEW

  • Collection of automated vital signs is now a common occurrence in a variety of settings, which offers great value as part of data sets as well as use in alerts.

  • Although bedside staff can easily identify and deal with artifact values, automated devices lack the ability to differentiate values as abnormal versus artifactual secondary to mechanical, electrical or physiologic causes (i.e. movement, electrical interference, loose sensors).

  • Artifactual vital sign alerts can lead to alarm fatigue, alarm desensitization, patient safety issues and provider burnout.

  • Since vital signs are used in a large number of early warning scores indicating impending deterioration, artifact values may result in:

    • If artifact values are used in early warning detection, may go unnoticed

    • If the value is real abnormality, but considered to be an artifact (and potentially discarded), the patient may go untreated.

  • Most approaches focus on one of two types:

    • Artifact Reduction

    • Artifact Detection

  • Artifact reduction methods are primarily used with imaging techniques.

  • Artifact reduction and detection methods can be used for imaging as well as physiologic values such as vital signs.

Vital sign artifact strategies

Strategy #1:  Cutoff Values

  • All values beyond a specified range is artifact

 

Strategy #2: Standard Deviation

  • All values beyond 3 standard deviations are artifact

 

Strategy #3: Interquartile Range

  • Interquartile range calculated for each vital sign for each patient

  • All values outside 3 multiples of range is artifact

     

Strategy #4: Outlier Value

  • Based on local ranges, values greater than 1.5 times are treated as artifact

 

Strategy #5: Neural Networks

  • Trained and validated on dataset separate from patient evaluation data.

 

Evaluation of 11,699 vital signs in surgical and ICU patients:

  • Each value was assessed by 5 human experts to be actual or artifact

  • The 5 methods were then used to detect automated artifact values

  • Sensitivity, specificity, positive predictive value and negative predictive value was determinied for each methodology

  • Results indicate no single method performed superior to any other.

  • Neural Network had the higher sensitivity with a specificity similar to others

  • Table below indicates which method had at least a 70% value for each vital sign in ICU patients.

CONCLUSIONS:

  • Vital sign artifact values may be difficult to detect via automated mechanisms.

  • No single method was found to be consistently superior across various vital signs and clinical scenarios.

  • Neural network systems may be a potential avenue to explore further.

  • Clinicians should maintain a high index of awareness when evaluating vital sign alerts. 

  • Automated systems to eliminate artifact data should not be relied upon.

Erkan Hassan