DATA SYNTHESIS AND SIMULATION OF DOSING SCENARIOS
Data were collated using Microsoft Excel. Our primary analysis was a mixed narrative and meta-analysis. Descriptive analytics and multiple regression modelling were performed using IBM SPSS Statistics 28. In cases where specific values were not reported (e.g. ultrafiltration rate, filtration fraction, volume of distribution) these were calculated using the available data and standard formulae where possible. Statistical heterogeneity was not assessed as the Cochrane Q Test and I2 statistic are not applicable to our dependent variable which is continuous rather than dichotomous.
Simulations were undertaken in R using tidyverse and linpk packages[29-31] . Individual patient level data was used for elimination rate constant and volume of distribution. Elimination rate constant was taken directly from published manuscripts or derived from other published pharmacokinetic parameters, where available. Volume of distribution was not available for 8 patients (Chappell). Simulated patients (n=10,000) were created from these values using the MASS package and assuming a multivariate log-normal distribution of parameters [32] . For pharmacokinetic profiles, a one compartment model was assumed. Profiles were simulated to 72-hours, on the assumption that steady state would be achieved by then and that this early period of treatment was likely to be the most important in achieving therapeutic concentrations. We utilised a target trough concentration range of 12 – 46 mcg/mL as this is the most widely used therapeutic index [6, 25, 34, 38] .