Volume 12, Number 3—March 2006
Research
Medication Sales and Syndromic Surveillance, France
Table 2
Forecast accuracy*†
Forecast method of ILI incidence | Forecast horizon (wks) | National level |
Regional level |
||
---|---|---|---|---|---|
Weeks 2003(40)–2004(34) | Weeks 2003(41)–2004(01) (preepidemic and epidemic) | Weeks 2003(40)–2004(34) | Weeks 2003(41)–2004(01) (preepidemic and epidemic) | ||
Using drug-sales data | 1 | 0.93 | 0.96 | 0.70 | 0.64 |
2 | 0.93 | 0.95 | 0.70 | 0.56 | |
3 | 0.91 | 0.85 | 0.69 | 0.54 | |
Current method used by FSN | 1 | 0.90 | 0.96 | 0.73 | 0.67 |
2 | 0.93 | 0.88 | 0.68 | 0.53 | |
3 | 0.91 | 0.74 | 0.65 | 0.40 |
*Defined as the correlation coefficient between observed and predicted influenzalike incidences and calculated on the validation dataset. Correlations were computing for 1-, 2-, and 3-week-ahead prediction obtained from medication sales for 2 periods (p<0.001). Forecast accuracy was compared with that of the method of analogs. At the regional level, the value in the table was obtained by averaging the correlation coefficient over the 21 regions of France. The number in parentheses after a year refers to that week of that year.
†ILI, influenzalike illness; FSN, French Sentinel Network.
1Both authors contributed equally to this research.