Table 2. Comparison of the performance of detection algorithms following Seasonal Autoregressive-Integrated Moving-Average (SARIMA) smoothing (resulting in fitted and residuals) values using transmissible gastroenteritis virus (TGEV) polymerase chain reaction (PCR)-negative submissions from January 1st, 2011, through October 10th, 2013.
EARS (Exponentially Weighted Anomaly Score), EWMA (Exponentially Weighted Moving Average), and Farrington.
| Alarm algorithms | Model parameters | # monitored weeks | Total alarms | False alarms number (%)a | Number of true alarms (starting April 1st 2013)b | Time (Weeks) to detect PEDV emergence (April 29th, 2013) |
|---|---|---|---|---|---|---|
| CUSUM fitted | Initial value of standard error = 0; Number of standard errors to be considered out of control = 5; Number of shifts to be detected = 1 | 51 | 10 | 5 (10%) | 5 | 9 |
| CUSUM residuals | Initial value of standard error = 0; Number of standard errors to be considered out of control = 5; Number of shifts to be detected = 1 | 51 | 1 | 0 (0%) | 1 | 4 |
| EARS C1 fitted | Method = C1; Lambda = 0.5; Alpha = 0.001; Baseline = 7 weeks | 51 | 0 | NA | NA | NA |
| EARS C1 residuals | Method = C1; Lambda = 0.5; Alpha = 0.001; Baseline = 7 weeks | 51 | 1 | 0 (0%) | 1 | 4 |
| EARS C3 fitted | Method = C3; Lambda = 0.5; Alpha = 0.001; Baseline = 11 weeks | 51 | 0 | NA | NA | NA |
| EARS C3 residuals | Method = C3; Lambda = 0.5; Alpha = 0.001; Baseline = 11 weeks | 51 | 1 | 1 (2%) | 0 | NA |
| EWMA fitted | Baseline = 1 years; Lambda = 0.4; Sigma = 3 | 51 | 11 | 6 (12%) | 5 | 9 |
| EWMA residuals | Baseline = 1 years; Lambda = 0.4; Sigma = 3 | 51 | 1 | 0 (0%) | 1 | 4 |
| Farrington fitted | Baseline = 1 year; Weeks before and after current week = 3 weeks; Past week weight = 1; Number of past weeks not included = 3; Alpha = 0.05 | 51 | 7 | 3 (6%) | 4 | 9 |
| Farrington Flexible fitted | Baseline = 1 year; Weeks before and after current week = 1 weeks; Weight of past week = 2.58; Number of past weeks not included = 3; Alpha = 0.05 | 51 | 4 | 0 (0%) | 4 | 4 |
a False alarm rate referred to the frequency of incorrect alarms (false positive) divided by the total number of monitored weeks. For PEDV emergence, given that the true alarms were expected to begin after April 1st, 2013 (one month earlier than the first PEDV disease cases observed in the ISU VDL, on April 29th 2013), any alarms generated in any weeks between January 2010 and March 29th, 2013 were considered false alarms.
b True alarms referred to alarms generated when there was a true increase in weekly negative submissions within the period of monitoring. For PEDV emergence, the true alarms were expected to begin after April 1st, 2013 (one month earlier than the first PEDV disease cases observed in the ISU VDL, on April 29th, 2013).
c Time to detect an outbreak refers to the number of weeks between the first alarm and the week reporting PEDV emergence (April 29th, 2013).