For each model, we calculated the deviance information criterion (DIC) (Spiegelhalter et al. 2002), which can be viewed as a Bayesian analogue of the Akaike Information Criterion (AIC; Akaike 1973).
The OUV measure calculates the deviance between genomic oligonucleotide frequencies and approximated oligonucleotide frequencies using the considered oligonucleotide's mononucleotide frequencies.
We assessed the goodness of fit of the model to the data by calculating the residual deviance, 9 the heterogeneity of treatment effects estimated from the median between trial variance τ observed in the posterior distribution, and the consistency of the network determined by use of inconsistency factors, 17 defined as the difference in log rate ratios derived from direct and indirect comparisons.
For all MTC analyses, we measured the goodness of fit of each of the models to the data by calculating the residual deviance and comparing it against the number of unconstrained data points, where the number of unconstrained data points was obtained by summing up the number of study arms across all studies included in our analyses.
The significance of each of the factors was obtained by calculating the decrease in deviance when the factor was included in the model (given the null model including age and sex only) and comparing this to the appropriate chi-square distribution.
The value of each lagged media series in predicting mortality was determined by calculating the proportion of the deviance explained, similar to the R2.
The contribution of nonhuman surveillance systems to variability in human risk was determined by calculating the proportion of the deviance explained (R).
We calculated the percent deviance explained by season, day-of-week, and holiday patterns by estimating the deviance from a null GLM and a full GLM model with the season/temporal trend, day-of-week, or holiday term; the full deviance was subtracted from the null deviance and then divided by the null deviance.
Again, we ran each parameter set three times and calculated the mean deviance.
(2) The regression coefficients from step 1 were used to calculate the partial deviance for the respective validation subsample, that is, to assess how well the outcomes in that subsample were predicted by a given model (Binquet et al, 2008).
In order to assess whether there is a seasonal effect of influenza, we performed a likelihood ratio test by calculating the difference of statistical deviance between the time-varying coefficient Poisson model (equation 3) and the main effect model (equation 2) [ 12].
