Skip to main content
. 2020 Dec 17;15(12):e0242878. doi: 10.1371/journal.pone.0242878

Table 6. Univariate DL analysis in the LEMA 0–8 dataset.

AUC(95% CI) Top-Left Threshold (Prob) Threshold (Values) Normal Values Specificity Sensitivity Accuracy
LACTATE 0.92(0.889–0.951) 0.01 0.3–2.7 mmol/L <2 mmol/L 0.897 0.826 0.887
PH 0.867(0.831–0.902) 0.03 7.32–7.52 units 7.35-7-45 units 0.899 0.764 0.878
BICARBONATE 0.842(0.807–0.877) 0.07 21–33 mEq/L 24–30 mEq/L 0.823 0.75 0.812
POTASSIUM 0.770(0.731–0.809) 0.06 3.4–4.4 mEq/dL 3.5–5 mEq/dL 0.749 0.666 0.735
CALCIUM 0.709(0.668–0.750) 0.03 7.9–9 mg/dL 8–10.5 mg/dL 0.636 0.675 0.642
GLUCOSE 0.683(0.622–0.744) 0.05 85–160 mg/dL 70–130 mg/dL 0.737 0.606 0.717
CHLORIDE 0.659(0.616–0.702) 0.1 102–111 mEq/L 95–106 mEq/L 0.676 0.602 0.665
SODIUM 0.619(0.574–0.665) 0.08 137–143 mEq/L 135–145 mEq/L 0.631 0.549 0.619

The DL-derived optimal probability cutoff was used to estimate the parameter values discriminating ICU patients who remained stable from those who physiologically deteriorated and died in 0–8 hours. Lactate was the most reliable predictor value, with an AUC of 0.92, followed by pH and bicarbonate, with AUCs of 0.867 and 0.842, respectively.