Table 4.

Confusion Matrix for Modified Early Warning Score During Bedside Evaluation in Unplanned Escalation of Care

Actual Prognosis
Predicted PrognosisAliveDeceasedTotals
Alive183 (a or TP)73 (b or FP)256 (r1)
Deceased3 (c or FN)4 (d or TN)7 (r2)
Totals186 (c1)77 (c2)263 (t)

  • ARR, Absolute risk reduction; CI, confidence interval; DP, difference in proportions; FN, false negative; FP, false positive; TN, true negative; TP, true positive.

  • Prevalence=Alive [c1/t]=186/263=0.707 (71%) (CI 0.65 to 0.76); Deceased [c2/t]=77/263=0.293 (29%) (CI 0.241 to 0.35)

  • Kappa=0.049 (CI –0.015 to 0.103).

  • Test statistics not dependent upon prevalence.

  •  Sensitivity=a/c1=183/186=0.984 (CI 0.97 to 0.996)

  •  Specificity=d/c2=4/77=0.052 (CI 0.019 to 0.080)

  •  Positive predictive value=a/r1=183/256=0.715 (CI 0.71 to 0.72)

  •  Negative predictive value=d/r2=4/7=0.571 (CI 0.20 to 0.88)

  •  Positive likelihood ratio=Sensitivity/(1–Specificity)=0.984/(1–0.052)=1.038 (CI 0.99 to 1.08)

  •  Negative likelihood ratio=(1-Sensitivity)/Specificity=(1–0.984)/0.052=0.310 (CI 0.06 to 1.61)

  •  Odds ratio=(a/b)/(c/d)=(183/73)/(3/4)=3.34 (CI 0.61 to 19.4)

  •  Relative risk=(a/r1)/(c/r2)=(183/256)/(3/7)=1.67 (CI 0.89 to 6.08)

  •  Diagnostic odds ratio=[Sensitivity/(1–Sensitivity)]/[(1–Specificity)/Specificity=[0.984/(1–0.984)]/[(1–0.052)/0.052]=3.373 (CI 0.61 to 19.36)

  •  Error odds ratio=[Sensitivity/(1–Sensitivity)]/[Specificity/(1–Specificity)]=(0.984/[1-0.984])/(0.052/[1-0.052])=1,139 (CI 1,711 to 2,553)

  •  Difference in proportions=[(a/r1) – (c/r2)]=[(183/256) – (3/7)]=0.286 (CI –0.09 to 0.60)

  •  Number needed to treat=(1/absolute value of DP) which is equal to (1/absolute value of ARR)=1/0.286=3.49 (CI 1.66 to infinite)

  •  Absolute risk reduction=[(c/r2) – (a/r1)]=[(3/7) – (183/256)]=which is equal to –DP=–0.286 (CI –0.60 to 0.09)

  •  Relative risk reduction=[ARR/(c/r2)]=[–0.286/(3/7)]=–0.668 (CI –5.079 to 0.114)

  •  Youden J value=(Sensitivity+Specificity–1)=(0.984+0.052–1)=0.036 (CI –0.01 to 0.08)

  •  Number needed to diagnose=which is equal to (1/Youden J)=(1/0.036)=27.9 (CI 13.23 to 88.03)

  • Test statistics dependent upon prevalence.

  •  Accuracy=(a+d)/t)=(183+4)/263=0.711 (71%) (CI 0.69 to 0.73)

  •  Misclassification rate=[(c+b)/t]=(3+73)/263=0.289 (29%) (CI 0.27 to 0.31)

  •  Number needed to misdiagnose=[1/(1–Accuracy)]=[1/(1–0.711)]=3.46 (CI 3.24 to 3.67)