Fixed effects included | WAIC | ΔWAIC | dSE | Model weight in ensemble |
|---|
Data from all studies (n=16) |
None | 2222.5 | - | - | 0.62 |
Severity | 2223.5 | 1.0 | 1.1 | 0.38 |
Data from studies (n=14) with demographic information |
None | 1796.6 | - | - | 0.33 |
Age | 1797.4 | 0.8 | 1.2 | 0.22 |
Severity | 1798.8 | 2.2 | 1.0 | 0.11 |
Sex | 1798.9 | 2.2 | 0.9 | 0.11 |
Severity + sex | 1799.7 | 3.0 | 1.4 | 0.07 |
Age + sex | 1799.7 | 3.1 | 1.1 | 0.07 |
Age + sex + severity | 1800.7 | 4.0 | 1.9 | 0.04 |
Age + severity | 1800.7 | 4.0 | 1.7 | 0.04 |
- Three fixed effects (severity, age, sex) were added to linear regression models of viral load over time, separately or in combinations (see the “Materials and methods” section), to determine the extent to which they explained variation present in the data. All studies included in this analysis described the severity of disease for each patient. However, patient-level demographic information was only provided in 15 of the 17 studies. Hence, a separate analysis was carried out for these studies. In each analysis, the best-fitting model is the one with the lowest WAIC value, as indicated in bold. The value of ∆WAIC indicates the difference in goodness of fit between each model and the one that provided the best fit to the data, whilst dSE denotes the standard error of the difference in WAIC between each model and the best-fitting one. In each analysis, we provide the Akaike weight of each model in the ensemble of candidate models that were analysed (Materials and methods). As a sensitivity analysis, we also assessed the goodness of fit by leave-one-out cross-validation [52], which resulted in very similar conclusions (see Supplementary Table 4)
