Skip to main content
  • Poster presentation
  • Published:

Continuous prediction of glucose-level changes using an electronic nose in critically ill patients


Many if not most critically ill patients are treated with insulin during their stay in the ICU [1]. Intensive monitoring of the blood glucose level is a prerequisite for efficient and safe insulin titrations in these patients [2]. Current continuous glucose measurement techniques rely on subcutaneous glucose measurements [3] or measurements in blood [4]. We hypothesized that changes in volatile organic compound (VOC) concentrations in exhaled breath reflect changes in the blood glucose level. Changes in VOC concentrations can be analyzed continuously using a so-called electronic nose (eNose) [5]. Our aim was to investigate exhaled breath analysis to predict changes in glucose levels in intubated ICU patients.


Exhaled breath was analyzed in 15 intubated ICU patients who were monitored with a subcutaneous CGM device. eNose results were compared with subcutaneous glucose measurements and linear regression models were built, including subject-specific models, and whole-sample models. The models were validated using temporal validation by training the model on the first 75% of measurements and prospectively testing on the last 25% of measurements. Performance of the models was measured using an R2 value, Clarke error grids (CEG) and rate-error grid analysis (R-EGA).


Changes in VOC concentrations were associated with changes in subcutaneous glucose levels. R2 performance had a mean value of 0.67 (0.34 to 0.98) for subject-specific models, and a mean value of 0.70.(0.52 to 0.96) for the model for the whole sample. However, when externally validating the model, the predictive performance dropped to a mean R2 of 0.19 (0.00 to 0.70) for subject-specific models, and 0.04 for the model for the whole sample. Point accuracy in CEG was mostly good with >99% in zones A and B; trend accuracy, as visualized with R-EGA, was low.


Exhaled breath prediction of glucose levels seems promising. However, performance of the current models is too low to be used in daily practice.


  1. Schultz MJ, et al.: Intensive Care Med. 2010, 26: 173-174.

    Article  Google Scholar 

  2. Preiser J-C, et al.: Intensive Care Med. 2009, 35: 1738-1748. 10.1007/s00134-009-1585-2

    Article  CAS  PubMed  Google Scholar 

  3. Kosiborod M, et al.: Crit Care. 2013, 17(2):462.

    Google Scholar 

  4. Schierenbeck F, et al.: Diabetes Technol Ther. 2013, 15: 26-31. 10.1089/dia.2012.0169

    Article  CAS  PubMed  Google Scholar 

  5. Rock F, et al.: Chem Rev. 2008, 108: 705-725. 10.1021/cr068121q

    Article  PubMed  Google Scholar 

Download references

Author information

Authors and Affiliations


Rights and permissions

Reprints and permissions

About this article

Cite this article

Leopold, J., Van Hooijdonk, R., Bos, L. et al. Continuous prediction of glucose-level changes using an electronic nose in critically ill patients. Crit Care 18 (Suppl 1), P437 (2014).

Download citation

  • Published:

  • DOI: