In critically ill SLEDD patients, continuous infusion dosing regimens were the most suitable type of infusion to maximise PK/PD target attainment, taking into account efficacy and toxicity thresholds. Leveraging data reflecting the real-world situation in ICUs, the identification of the most suitable dosing regimens was achieved by successfully developing a population PK model for meropenem in critically ill patients receiving SLEDD and evaluating PK/PD target attainment in this population across a wide range of clinically relevant dosing regimens. The associated simulation results were translated into a clinically usable dosing nomogram.

Splitting elimination into two distinct processes, i.e. the extracorporeal elimination and remaining renal pathway, was successful. CLSLEDD contributed to more than half (55%) of the total meropenem CL of 6.4 L/h. While not directly quantified through dialysate and pre-/post-filter measurements, the assessment of TDM data during clinical routine on-SLEDD and off-SLEDD periods allowed for a differentiation between and quantification of CLSLEDD and CLREN. Overall, all PK model parameters were plausible compared to previously reported parameters. Our dialysis-associated CL of 3.49 L/h corresponded well with previously reported values of 1.2 L/h and 4.8 L/h for intermittent KRT [27, 28]. Using significantly higher ultrafiltration rates (500 mL/h, our study: 280 mL/h) and blood/dialysate flows (250 mL/min, our study: 180 mL/min), a higher SLEDD-associated CL of 7.9 L/h was reported in [8]. Another study including renally impaired patients reported a lower total CL of 3.52 L/h without dissection into a dialysis-associated CL and a renal CL [19], yet the PK analysis was based on a population receiving SLEDD, the less efficient CKRT or KRT (each group about one third).

In terms of V, Mouton et al. [29] reported 21 L in healthy non-continuous kidney replacement volunteers. In comparison, Chung et al. reported 14.3 L for V of the central and 17.7 L for V of the peripheral compartment in CKRT patients, which is similar to the V of 33.8 L observed in our study [30]. One- and two-compartment disposition models have been previously described in the literature for meropenem, which supports the identification of a one-compartment model in this study to characterise the PK of meropenem in critically ill patients [31,32,33].

The meropenem CL of healthy volunteers has been shown to range from 11–14 L/h [27, 28]. Since our study population suffered from severe renal impairment (Table 1), we expected our estimated CLREN value to be substantially lower: the estimated value of 2.87 L/h corresponded to a glomerular filtration rate of 4.13 mL/min (CLCRurine). Thus, at CLCR values of healthy individuals, e.g. 120 mL/min, CLREN reached the expected range for healthy volunteers at 11.2 L/h, highlighting the plausibility of the estimated effect size in our covariate model, namely a 25% decrease in CLREN for a 10 mL/min deterioration in glomerular filtration. The identification and quantification of CLCRurine as a time-varying covariate influencing CLREN, considered potential changes in renal CL of meropenem in individual patients over the course of therapy. This covariate–parameter relationship aligned with previously developed non-dialysis meropenem models in which kidney function has been shown to significantly influence meropenem CL [32]. The residual meropenem CL of 2.87 may be attributable to a reported increase in active tubular secretion and proportion of non-renal elimination and/or metabolism of meropenem in renally impaired critically ill patients [27, 34].

For the first time, this study quantified the variability of meropenem CLSLEDD, most likely due to machine setting changes, within the different on-SLEDD periods of a critically ill patient implemented as IOV. This was only possible due to the availability of a larger number of data points from multiple on-SLEDD periods per patient. In previously developed population PK models of meropenem therapy in critically ill SLEDD patients this aspect was not considered [8, 19]. The high variability between the on-SLEDD periods (109%CV) underlined the probable impact of distinct SLEDD settings (e.g. duration, interval, dialysate flow rates, blood flow rates, ultrafiltration rate, and dialyser type) on the SLEDD-associated CL of meropenem. However, similar to Braune et al. [8], none of those settings or other factors were identified as significant covariates on CLSLEDD, despite weak correlation trends in the graphical analysis between CLSLEDD, the dialysate flow rate, and the ultrafiltration rate, respectively (Fig. S5). That we could not identify this relationship as statistically significant might also be likely due to a rather small variation in different SLEDD settings and the variable but sparse sampling scheme. Nevertheless, IOV could be associated with the SLEDD elimination of meropenem, reducing RUV by 13%. The additional inclusion of IIV on CLSLEDD did not result in an improvement of the model.

Critically ill patients are at high risk of excessive meropenem serum concentrations, quickly reaching nephrotoxic (Cmin > 44.45 mg/L) or neurotoxic levels (Cmin > 64.2 mg/L), which can be prevented via a dosing strategy guided by the defined PK/PD target window [12, 35]. The toxicity threshold proposed by Imani et al. was selected because the analysed patient population is urgently dependent on their remaining kidney function. The applied target window offers practical and clinically oriented assistance based on the latest guidelines [9, 14].

Within the simulation plots, the bell-shaped curves per dosing regimen in Fig. 5 resulted from the relationship between CLCRurine and meropenem exposure: lower CLCRurine led to higher meropenem concentrations, increasing the probability of exceeding the toxicity threshold (i.e. 44.45 mg/L), whereas higher CLCRurine resulted in lower meropenem exposure, reducing the probability of exceeding the efficacy threshold (i.e. 8 mg/L). The dosing simulation analyses for a clinically relevant range of CLCRurine values between 0 to 40 mL/min revealed that continuous infusions of 2 g or 3 g had the highest PTA values compared to short-term and prolonged infusion regimens. Our results are consistent with those of Westermann et al., who reported that patients with KRT and 2 g/24 h continuous meropenem infusion achieved a PTA of 95% [19].

This study focused on meropenem dosing regimens in the rare population of highly vulnerable critically ill patients undergoing SLEDD, resulting in a small total number of included patients. However, the extensive data collected per patient (median of 14 samples, from on-SLEDD as well as off-SLEDD periods) facilitated the robust estimation of PK parameters. Our developed dosing nomogram allows clinicians to determine the most suitable dosing regimen for patients undergoing SLEDD, aligning with the conditions of our study, yet commonly applied (7 h on-SLEDD periods q24h, blood flow 131–240 mL/min, dialysate flow 120–240 mL/min, ultrafiltration rate 107–565 mL/h and blood volume 26.8–109 L). Based on our analyses, we advise not to use estimated glomerular filtration rate formulas with our nomogram, only CLCRurine. In clinical scenarios with patients exhibiting no residual diuresis and thus CLCRurine cannot be calculated, the dosing recommendations from the first column of the dosing nomogram (i.e. CLCRurine = 0 mL/min), can still be employed. Before our developed model and dosing nomogram can be used more broadly in clinical practice, a prospective clinical validation with external data must be performed.

In the clinical study, no individual pathogen or MIC determinations were acquired [15] which could in a next step further optimise meropenem exposure during SLEDD therapy. In order to apply the finding to other SLEDD settings than in this clinical trial, future studies should focus on broader ranges of durations, intervals, and flow rates to identify influential parameters on the SLEDD-associated elimination of meropenem.

Overall, our developed PK model for SLEDD therapy in critically ill patients has the potential to be integrated into easy-to-use model-informed precision dosing instruments [20] after completion of the prospective validation. Thereby, our model can enable healthcare professionals to decide on dosing adjustments when SLEDD therapy is applied directly at the patient's bedside. By integrating TDM data, Bayesian data analysis approaches could be combined with our model to enhance the precision of meropenem exposure predictions, facilitating more accurate and tailored dosing recommendations [20, 36,37,38,39].