Volume 11, Number 2—February 2005
Dispatch
Comparing Aberration Detection Methods with Simulated Data
, Timothy Browne*, G. Matthew Seeman*, and Aaron T. Fleischauer*
Table
By method, overall sensitivity and specificity and time to detection
| Type of method | Name | Sensitivity (%) | Specificity (%) | Time to detection (d)* |
|---|---|---|---|---|
| Historical methods (at least 5 y historical data) | Seasonally adjusted CUSUM† | 82.5 | 88.7 | 1.272 |
| Historical limits‡ | 43.9 | 96.3 | 2.942 | |
| Nonhistorical methods (<3 y historical data) | C1–MILD§ | 60.1 | 97.0 | 1.122 |
| C2–MEDIUM¶ | 71.2 | 97.0 | 1.319 | |
| C3–ULTRA** | 82.3 | 95.4 | 1.307 |
*Time to detection must be interpreted with sensitivity because time to detection does not include missed outbreaks.
†The seasonally adjusted CUSUM method sums the positive differences of the current value from the mean for a period similar to the current value over 5 years.
‡The historical limits method compares the current sum of 4 time periods to the mean of the sum of 15 totals of 4 time periods surrounding the current point of interest over 5 years.
§The C1–MILD method is based on CUSUM, but the calculations reduce to the current value being greater than the mean plus 3 standard deviations (SD), with the mean and SD based on the past 7 days.
¶The C2–MEDIUM method is based on CUSUM, but the calculations reduce to the current value being greater than the mean plus 3 SD, with the mean and SD based on the past 7 days shifted by 2 days.
**The C3–ULTRA method is based on CUSUM, summing the positive difference of the current value from the mean for 3 days, with the mean and SD based on the past 7 days shifted by 2 days.