Table 3.
Final negative binomial regression (‘forecasting’) model for monthly Ross River virus notifications in the Mildura local government area, Victoria, Australia. Trained on data for the period July 2000–June 2011, validated on data for the period July 2012–June 2015
| Variable (units) | Lag (months) | IRR | s.e. | 95% CI | P value |
|---|---|---|---|---|---|
| log2 (maximum vapour pressure, hPa) | 1 | 13·7 | 6·01 | 5·76–32·4 | <0·001 |
| log2 (precipitation, mm) | 4 | 1·44 | 0·13 | 1·21–1·71 | <0·001 |
| Number of days with precipitation >1 mm | 6 | 1·16 | 0·08 | 1·01–1·33 | 0·036 |
| Est. | s.e. | 95% CI | P value* | ||
| Dispersion parameter (α) | 1·05 | 0·28 | 0·62–1·78 | 0·009 | |
| Forecasting performance | |||||
| Financial year | Forecast cases (95% PI) | Observed cases | ρP | Forecast outbreak alerts | Observed outbreak alerts |
| 2011/2012 | 30 (0–122) | 27 | 0·59 | 2 | 3 |
| 2012/2013 | 11 (0–52) | 30 | 0·32 | 2 | 6 |
| 2013/2014 | 19 (0–88) | 16 | 0·58 | 1 | 1 |
| 2014/2015 | 14 (0–67) | 37 | 0·38 | 0 | 6 |
IRR, Incidence rate ratio; s.e., standard error of IRR; CI, confidence interval; PI, prediction interval; ρP, Pearson's correlation coefficient.
AIC = 367·6142, n = 132, degrees of freedom = 5, log likelihood (model = −178·807; null model = −226·598), maximum likelihood R2 = 0·515, deviance-based goodness-of-fit (P = 0·81).
* P value for dispersion parameter estimated using a likelihood ratio test that α is non-zero.