Fig 1.
Estimated probability of virus positivity for 7 different viruses using RCS, periodic RCS and CS, and cosinor models with one and two harmonics using data from Horton et al.
The step functions are the observed weekly proportions of virus positive samples, the lines are the estimated probabilities estimated using the 5 different models.
Table 1.
Analysis of the complete dataset of Horton et al.
Estimates are obtained with repeated 10-fold cross-validation.
Table 2.
Repeated analysis of subsets of 500 units; the number of knots is based on AIC; estimates (Brier score, c index, calibration intercept and slope) are obtained on the data not included in the model estimation, power is evaluated on training data.
Fig 2.
Simulation results for the alternative cases.
Each row shows the results from one of the three alternative simulations settings, each column refers to one of the models. The black curves show the probabilities from the model generating the data, the average estimates obtained using the five models are shown with gray lines; dashed lines are average limits of the 95% confidence intervals. Spline models were fitted using 5 parameters. See methods for details on the simulation settings.
Table 3.
Simulation results from the alternative and null case; the probabilities in the alternative case are simulated from a sine function, shown in Fig 2.
Table 4.
Simulation results from the second alternative case, where the probabilities are simulated from a periodic function with two peaks, shown in Fig 2.
Table 5.
Simulation results from the third alternative, where the probabilities are simulated from a periodic function with complex pattern, shown in Fig 2.