PMC

Legend: OR: log₁₀odds ratio; LR: log₁₀likelihood ratio. Log₁₀positive and negative likelihood summed make the log₁₀odds ratio, shown on a log₁₀odds scale. The width between test-treatment threshold and decision threshold is determined by the negative likelihood ratio of the test. Mutatis mutandis, the distance between the test treatment and the decision threshold are given by the positive likelihood ratio.

Online training and evaluation scenario. Here, the model predicts at regular intervals during an ICU stay (we show predictions in 1-h intervals). For sepsis cases, there is no prima facie notion at which point in time positive predictions ought to be considered as true positive (TP) predictions or false positive (FP) predictions (mutatis mutandis, this applies to negative predictions). For illustrative purposes, here we consider positive predictions up until 1 h before or after sepsis onset (for a case) to be TP.

Study design. Abbreviations: n, number of evaluable subjects aimed at. To account for possible withdrawals and drop outs, about 50 subjects were randomized in cohort 1; PK, collecting of blood samples for PK profiling; QD, once daily; BID, twice daily. ^a DRSP/EE or DRSP/E2 treatment was continued until day 21 at least and could be prolonged until day 28 to synchronize visits for PK profiling. For simplicity, the PK profile day is always referred to as 'day 28'. The same applies, mutatis mutandis, to 'day 38'. ^b The group sequential approach allowed to stop the study for futility (or success, if applicable) after the planned interim analysis

More precisely, in panel a we show the set of data LM_αi versus Γ_αi, where α _i here runs over all the x provinces in Rheinland-Pfalz; the solid black line represents the best-fit obtained over the whole set of data pertaining to Germany. Notice that the best-fit is the same already presented in . In panel b we show ρ _αi obtained by dividing each data point LM_αi in panel a by the related expected value f(Γ_αi). Notice that the values of ρ _αi are scattered randomly around 1. The same passages are applied, mutatis mutandis, in panels c and d; the legend and the best fit are the same as in . Notice that now data pertaining to Campania and to Emilia-Romagna are systematically above and below, respectively, the best-fit line. Accordingly, the related values of ρ _αi are always larger and smaller than 1, respectively.