Table 2

Algorithm summary performance statistics for false alarm rates approximating 0.001

Algorithm

false alarm rate

median (mean) sensitivity

Day 1 sensitivity

Day 2 sensitivity

Days 1–7 sensitivity

CARL

median (mean) timeliness

median (mean) adjusted timeliness


EARS C1

0.0008

0 (0.27)

0.13

0

0.13

1.0

36.5 (49.0)

123 (113.1)

EARS C2

0.0008

0 (0.40)

0.13

0

0.13

1.0

44.5 (52.7)

90 (96.0)

EARS C3

0.0008

0 (0.47)

0.13

0

0.13

1.0

51.0 (52.4)

73 (93.4)

NBC 7-day

0.0008

1 (0.80)

0.13

0.07

0.27

1.8

19.5 (32.3)

21 (31.1)

NBC 14-day

0.0008

1 (0.80)

0.07

0

0.13

2.0

33.5 (39.0)

32 (36.5)

NBC 28-day

0.0008

1 (0.93)

0.07

0

0.2

3.0

16.5 (26.3)

18 (26.8)

NBC 56-day

0.0008

1 (0.80)

0

0

0.13

5.0

19.0 (31.0)

20 (30.1)

p-value¥

-

<0.0001

-

-

-

-

0.91

0.0003


Test threshold that produced a false alarm rate ≤ 0.001

Detection on the first outbreak day is equivalent to a timeliness of 0 days

¥Friedman rank sum test

CARL: Conditional Average Run Length – conditional on the detection of the outbreak during the first 7 days

EARS: Early Aberration Reporting System

NBC: Negative binomial cusum with an out of control state defined as 2 standard deviations greater than the mean

Watkins et al. BMC Medical Informatics and Decision Making 2008 8:37   doi:10.1186/1472-6947-8-37

Open Data