Accuracy of bedside capillary blood glucose measurements in critically ill patients.

OBJECTIVE: To compare the accuracy of fingerstick with laboratory venous plasma glucose measurements (laboratory glucose) in medical ICU patients and to determine the factors which interfere with the accuracy of fingerstick measurements. PARTICIPANTS: The study included 80 consecutive patients aged 58+/-7 years, BMI 29.5+/-9.0, and APACHE II score 15+/-6 (277 simultaneous paired measurements). MEASUREMENTS: This prospective observational study compared fingerstick measurements to simultaneously sampled laboratory glucose once a day in patients in our medical ICU (twice daily if on an insulin infusion). Data recorded included patient demographics, admission diagnoses, APACHE II score, BMI, daily hematocrit, arterial blood gasses, chemistry results, concomitant medications (including vasopressors and corticosteroids), and upper extremity edema. Accuracy was defined as the percentage of paired values not in accord (>15 mg dl(-1)/ 0.83 mmol(-1)l(-1) difference for laboratory values <75 mg dl(-1)/4.12 mmol(-1)l(-1) and >20% difference for laboratory values >or=75 mg/dl). Outliers (blood glucose difference >100 mg dl(-1)/5.56 mmol(-1) l(-1)) were excluded from the correlation and distribution analyses.
RESULTS: Mean fingerstick glucose was 129+/-45 mg/dl (7.2+/-2.5 mmol/l) and mean laboratory glucose 123+/-44 mg/dl (6.8+/-2.4 mmol/l). The correlation coefficient between the two values was 0.9110 (Clinical and Laboratory Standards Institute threshold 0.9751). The mean difference (bias) between the two methods was 8.6+/-18.6 mg/dl (0.48+/-1.0 mmol/l) and limits of agreement +45.8 and -28.6 mg/dl (+2.5 and -1.6 mmol/l). Fifty-three (19%) paired measurements in 22 patients were not in accord (CLSI threshold <or=5%). In 44 (83%) of these paired measurements fingerstick glucose was greater than laboratory glucose.
CONCLUSIONS: The findings suggest that capillary blood glucose as measured by fingerstick is inaccurate in critically ill ICU patients and does not meet the CLSI standard. It is unclear whether the sampling method, device used, or both contributed to this inaccuracy. The wide limits of agreement suggest that fingerstick measurements should be used with great caution in protocols of tight glycemic control.