Can Model-Informed Precision Dosing of Beta-Lactams in Critical Care Improve Clinical Outcomes?

Abstract

Background

Critically ill patients exhibit acute physiological changes leading to marked antibiotic exposure variability and an increased probability of either subtherapeutic or potentially toxic exposures. Beta-lactam antibiotics are often affected by these physiological changes, leading to variable and often subtherapeutic exposures. Therefore, methods to increase the probability of a therapeutic exposure may be beneficial for critically ill patients. Model Informed Precision Dosing (MIPD) is an emerging strategy using advanced computer models to individualise dosing in this vulnerable population with or without therapeutic drug monitoring (TDM). By tailoring therapy to the individual patient scenario, MIPD aims to improve clinical outcomes by enhancing the likelihood of effective and safe therapy.

Summary

MIPD-guided beta-lactam antibiotic dosing may improve the probability of attaining target therapeutic exposures and clinical cure rates compared to standard fixed dosing in critically ill patients. Moreover, MIPD allows clinicians to target higher beta-lactam antibiotic exposures while minimising the risk of toxicity, when the goal of treatment is both improving clinical cure rates and suppressing the probability of antibiotic resistance emergence. No differences in mortality and adverse event reduction compared to standard dosing have been currently identified in the literature. However, the exact potential of MIPD has yet to be determined given the significant limitations in the current evidence base, including heterogeneous study designs, small sample sizes, inconsistent implementation strategies, and variable outcome definitions. Clinical practice and future research should optimise MIPD performance by selecting appropriate and innovative beta-lactam antibiotic exposure models, as well as identify methods to instigate early MIPD dosing and timely TDM dose adjustments. This may further improve clinical and economic outcomes for critically ill patients.

Key Messages

MIPD is a clinical decision making tool that uses pharmacokinetic modelling to guide beta-lactam dosing, with or without TDM input. A small number of observational and randomised beta-lactam MIPD intervention studies suggest that clinical outcomes for critically ill patients are improved. Technical, casemix, implementation, and study design limitations likely contribute to inconsistent findings in studies comparing MIPD to standard care. Stronger signals in improved outcomes are correlated with appropriate pharmacokinetic model and pharmacokinetic/pharmacodynamic (PK/PD) target selection, accompanied with timely dose adjustments in critically ill patients with significant pharmacokinetic variability and/or complex pharmacodynamic profiles. Future studies should more robustly define key features that will optimise patient outcomes including: optimal target exposure, characteristics of patients most likely to benefit, preferred minimum inhibitory concentration (MIC) parameter, favoured beta-lactam infusion duration, MIPD technical and implementation factors, and study design elements.

Introduction

Beta-lactam antibiotics are commonly used to treat life-threatening bacterial infections in critically ill patients [1]. Bacterial killing efficacy and patient-oriented clinical outcomes, such as improved rates of clinical cure, are related to the percentage of the dosing interval that the biologically available (unbound or free) beta-lactam antibiotic concentration exceeds the bacterial minimum inhibitory concentration (%fT_>MIC) [2, 3]. Although beta-lactam antibiotics are not commonly associated with clinically relevant adverse effects, recent evidence suggests that there is the potential for concentration-related toxicity [4]. Given that beta-lactam antibiotic concentrations may vary up to 500-fold between critically ill patients, it is highly likely that patients may experience either subtherapeutic or potentially toxic exposures [5]. Despite an awareness of these issues, standard beta-lactam antibiotic dosing remains current standard of care as a “one size fits all” approach. Methods to tailor the beta-lactam antibiotic dose to the specific individual patient may, therefore, enhance clinical outcomes by improving microbiological eradication and minimising the probability of toxicity [6].

Therapeutic drug monitoring (TDM) has emerged as a strategy to guide beta-lactam antibiotic dosing. TDM typically involves assaying a patient’s blood to determine the beta-lactam antibiotic concentration at a presumed “steady-state,” a time when the exposure variability is likely minimal. This enables the concentration to be compared with the target and a linear dose adjustment undertaken [2, 3, 7]. However, traditional TDM dose adjustment does not consider the dynamic pharmacokinetics observed in critically ill patients with fluctuating organ function and implementation of varying therapies. Moreover, delays in sampling while waiting for “steady-state” likely preclude rapid dose adjustment and increase the period of time that the patient may have a subtherapeutic beta-lactam exposure [7–9]. To help overcome these challenges, pharmacokinetic models embedded into software using either standalone or integrated with TDM, can be employed as part of model-informed precision dosing (MIPD). MIPD is a clinical decision support system which aims to optimise drug exposure by accounting for the anticipated inter- and intra-patient variability observed in critically ill patients, thereby improving the probability of target attainment [7, 10–12]. Despite its theoretical advantages, the impact of MIPD on clinical outcomes remains uncertain, with mixed findings reported in the literature [13–19]. This review aimed to synthesise the current evidence for MIPD-guided beta-lactam dosing and to inform both future clinical and research directions aiming to improve outcomes for critically ill patients.

Methods

We have performed a narrative review of model-informed precision dosing guiding beta-lactam antibiotic therapy for critically ill patients. In this review, we have provided an overview of the currently available evidence for the role in MIPD in achieving specific PK/PD targets, patient outcomes (mortality, clinical cure, and microbiological eradication), and addressed potential practical aspects that are relevant for implementing MIPD for beta-lactam antibiotics at the bedside. Relevant observational and controlled clinical trials, systematic reviews, and meta-analyses investigating the implementation of MIPD for beta-lactam antibiotics, were identified through a comprehensive search of the PubMed database from inception until August 2025 [20–24]. Citations and reference lists of relevant meta-analyses and systematic reviews were searched for additional studies. Study methodology was assessed using the Critical Appraisal Skills Programme (CASP) checklist [25]. Key variables were extracted from background literature, including patient characteristics, pathogen profiles, beta-lactam infusion modes, MIPD technical and implementation factors, plus study design elements.

Beta-Lactam Antibiotic Pharmacokinetic and Pharmacodynamic Principles

Critically ill patients exhibit altered beta-lactam antibiotic pharmacokinetics increasing the probability of subtherapeutic or toxic exposures at a given antibiotic dose. Moreover, patients in the intensive care unit (ICU) may be infected by bacterial isolates that are less susceptible to the beta-lactam antibiotic. These factors mean that the divide between a therapeutic and potentially toxic exposure is narrowed [6, 26–29].

Pathophysiological changes commonly observed in critically ill patients alter beta-lactam antibiotic pharmacokinetics, thereby affecting the dose required to reach target exposures. Critical illness affects the beta-lactam antibiotic volume of distribution (Vd) and clearance (CL). The Vd of hydrophillic antibiotics, such as beta-lactam antibiotics, may be increased severalfold due to fluid shifts resulting from capillary leak and administration of intravenous fluids. The CL may be increased due to augmented renal clearance from a hyperdynamic circulatory state and reduced with renal failure. The result of these changes is high inter- and intra-individual pharmacokinetic variability, which may be further complicated by medical interventions such as the implementation of renal replacement therapy and extra-corporeal membrane oxygenation [5, 6, 26–31].

Pre-clinical and clinical studies of hospitalised patients support both microbiological and clinical cure outcomes with a %fT>_MIC ranging from 40 to 70% for most beta-lactam antibiotics [2, 5, 27, 32–37]. In contrast, clinical studies in critically ill patients admitted to the ICU have identified improved clinical cure, suppression of antimicrobial resistance and reduced length of stay with higher target exposures up to 100% fT>_4xMIC [2, 8, 19, 35, 38]. Additionally, ICU patients may be infected with bacterial isolates that are up to fourfold less susceptible than other hospitalised patients [39]. Given the increased pharmacokinetic variability exhibited by critically ill patients, combined with the likely higher infecting organism MIC, it is not surprising that standard beta-lactam antibiotic dosing fails to achieve the minimum target therapeutic threshold (100% fT>_MIC) in approximately 35% of patients [5, 26, 40]. Dosing strategies that can individualise beta-lactam antibiotic therapy are, therefore, likely to improve patient outcomes. One such method is the use of MIPD. However, the integration of MIPD into clinical practice may be challenging [7].

The Rationale of MIPD in Critically Ill Patients

MIPD involves the application of previously generated pharmacokinetic models combined with patient-specific variables, such as renal function, and potentially with TDM to account for inter- and intra-patient pharmacokinetic variability [10–12]. Patient pharmacokinetic parameters are calculated by accompanying software, to simulate the probability of target attainment (PTA) for a given dosing regimen and specific desired target exposure guided by the clinician [12].

An a priori prediction refers to the initial estimation of drug exposure and associated dosing regimen prior to the availability of measured drug concentrations. In contrast, an a posteriori prediction incorporates TDM data to refine the previously estimated individual pharmacokinetic parameter estimates, enabling more precise dose adjustments [10–12].

The probability that the model software selects a dosing regimen that correlates with the desired therapeutic target depends on multiple factors. These may include the internal and external validity of the pharmacokinetic model, intra-patient variability in organ function and therapeutic interventions, beta-lactam protein binding and infusion mode, as well as TDM pre-analytical and analytical factors (shown in Fig. 1; online suppl. Table 1; for all online suppl. material, see https://doi.org/10.1159/000550822). Optimising these factors is critical to achieving high-quality MIPD predictions [7, 10–12, 41, 42].

Fig. 1.

Variables and associated components related to the pharmacokinetic model achievement of target exposure in clinical practice. ECOFF, epidemiological cut-off value; SOFA, sequential organ failure assessment; APACHE, acute physiology and chronic health evaluation; RRT, renal replacement therapy; ECMO, extracorporeal membrane oxygenation; MIC, minimum inhibitory concentration; PTA, probability of target attainment.

PTA through the Use of MIPD

MIPD may both improve the PTA during the beta-lactam antibiotic course, as well as reduce the time taken to achieve the therapeutic target exposure (Table 1). However, evidence is mixed and heterogeneous study design limits drawing firm conclusions.

Table 1.

Comparison of MIPD study PK/PD targets, MIPD performance, and clinical interpretation

Study, design, and number of participants	Patient population	MIPD antibiotics and intervention	PK/PD target	MIPD performance	Clinical interpretation
Machado et al. [16] (2017), retrospective cohort, n = 140	Age (years): Control: median 31, range [1–75] Intervention: median 31, range [2–90] Weight (kg): median 70, range [45–102]f Renal Function: unavailable Illness Severity Scores: Control: APACHE II median 18, range [8–36]; SOFA unavailable Intervention: APACHE II median 20, range [5–26]; SOFA unavailable	Piperacillin-tazobactam, imipenem, and meropenem TDM guided by non-compartmental model dosing recommendations (PK Solutions 2.0)	Total concentrations of piperacillin 100% fT>MIC, imipenem and meropenem 60% fT>MIC. Estimated MICs ranging from 0.5 mg/L, 2 mg/L, and 4 mg/L for all beta-lactams. Plus 1 mg/L and 8 mg/L for piperacillin	PTA calculated from an undisclosed timepoint based on estimated MICs. For clinical relevance, the following PTA was achieved based on a EUCAST ECOFF for Pseudomonas aeruginosa: piperacillin (8 mg/L): 60%, imipenem (2 mg/L): 100%, meropenem (2 mg/L) 100%. Control: unknown Time to target attainment: Intervention, Control: unknown	Culture positivity: 100% No significant difference between intervention and control groups Intervention (n = 77): clinical cure 60%; 14-day mortality 16%; hospital mortality 39% Control (n = 63): clinical cure 52%; 14-day mortality 14%; hospital mortality 36% Clinical cure definition: resolution of signs and symptoms of the infection within 14-day of treatment without changing or adding to the antimicrobial treatment and without the initiation of additional drugs within 48 h after discontinuation of therapy
Roggeveen et al. [17] (2022), two centre RCT feasibility study, n = 252	Age (years): Control: mean 67, IQR [56–74] Intervention: mean 68, IQR [60–75] Weight (kg): Control: mean 80, IQR [70–92] Intervention: mean 80, IQR [70–91] Renal Function (KDIGO Stageg): Control: 0–56.7% 1–23.3% 2–12.55% 3–7.5% Intervention: 0–60.6% 1–24.2% 2–6.8% 3–8.3% Illness Severity Scores: Control: APACHE unavailable; SOFA median 10, IQR [7–12] Intervention: APACHE unavailable; SOFA median 10, IQR [7–13]	Meropenem and ceftriaxone TDM informed by Bayesian software (Autokinetics)	Total concentrations of meropenem and ceftriaxone 75% fT>4×MIC (ECOFF). EUCAST ECOFFs based on combinations of prevalent pathogens and the antibiotics studied (but not disclosed)	Initial achievement of target exposure: Intervention (n = 115): Meropenem 12/19 (63%); ceftriaxone 37/39 (95%). Control (n = 91): Meropenem 12/15 (80%); ceftriaxone 31/32 (97%) Overall achievement of target exposure: Intervention, Control: unknown Time to target attainment (h): Intervention: Meropenem median 4.97, IQR (0–20.0); ceftriaxone median 14.9, IQR (6.4–18.4). Control: Meropenem median 0.08, IQR (0–18.6); ceftriaxone median 18.2, IQR (10.2–18.9)	Culture positivity not reported No significant difference between intervention and control groups Intervention (n = 132): mortality: ICU 34.1%; hospital 34.1%; new onset AKI 33.3%; received CVVH 13.6%; 96 h SOFA score median 10.0, IQR [6.0–15.2]; 96 h delta SOFA score median 0.0, IQR [−3.0 to 4.0]; hospital LOS (days) median 14.0, IQR [4.6–30.2]; ICU LOS (days) median 3.8, IQR [1.6–11.7] Control (n = 120): mortality: ICU 36.7%; hospital 39.2%; new onset AKI 41.7%; received CVVH 15.8%; 96 h SOFA score median 10.0, IQR [5.0–19.2]; 96 h delta SOFA score median 0.0, IQR [−2.0 to 4.0]; hospital LOS (days) median 12.2, IQR [3.4–28.4]; ICU LOS (days) median 3.6, IQR [1.0–11.0] Safety dosing group analysis (groups based on cumulative dose within first 24 h post-randomisation compared to standard licensed dose) High dose (>200% of standard dose, n = 36): new onset AKI: 44.4%; received CVVH: 5.6%; delta SOFA score at 96 h median −1.0, IQR [−3.0 to 4.0]; ICU LOS (days) median 2.3, IQR [1.8–7.0]; hospital LOS (days) median 12.7, IQR [5.4–26.5]; ICU mortality 30.6%; hospital mortality 30.6% Control (100%, n = 120): new onset AKI: 41.7%; received CVVH: 15.8%; delta SOFA score at 96 h median 0.0, IQR [−2.0 to 4.0]; ICU LOS (days) median 3.6, [1.0–11.0]; hospital LOS (days) median 12.2, IQR [3.4–28.4]; ICU mortality 36.7%, hospital mortality 39.2% Low dose (<50% of standard dose, n = 24): new onset AKI: 12.5%; received CVVH: 20.8%; delta SOFA score at 96 h median 0.0, IQR [−1.5 to 4.0]; ICU LOS (days) median 3.5, IQR [1.1–11.2]; hospital LOS (days) median 29.3, IQR [1.1–37.4]
Ewoldt et al. [18] (2022), multicentre RCT, n = 388	Age (years): Control: median 64, IQR [53–71] Intervention: median 65, IQR [57–71] Weight (kg): Control: median 80, IQR [70–90.1] Intervention: median 82.0, IQR [70–100] Renal function (eGFRh (mL/min/1.73 m2): Control: median 60.1, IQR [33.1–98.2] Intervention: median 65.3, IQR [32.1–110] Illness Severity Scores Control: APACHE IV median 70, IQR [50–91]; SOFA median 8, IQR [5–10] Intervention: APACHE IV median 71, IQR [53–89]; SOFA median 8, IQR [5–11]	Amoxicillin, cefotaxime, ceftazidime, ceftriaxone, cefuroxime, flucloxacillin, meropenem, and piperacillin TDM informed by Bayesian software (InsightRx™ version 1.15.16)	100% fT>MIC (ECOFF) of the following values for each drug: Unbound concentrations of ceftriaxone 1 mg/L and flucloxacillin 1 mg/L. Total concentrations of amoxicillin 8 mg/L, cefotaxime 4 mg/L, ceftazidime 8 mg/L, cefuroxime 8 mg/L, meropenem 2 mg/L, piperacillin 16 mg/L	24 h achievement of target exposurea: Intervention (n = 189): 55.6%. Control (n = 199): 60.9% Overall achievement of target exposurea: Intervention: average of sampling on days 1, 3, 5, 7 of 61.6%. Control: average of sampling on days 1, 3, 5, 7 of 57.3% Time to target attainment: Intervention, Control: unknown	Culture positivity: Intervention 78% (n = 148); Control 77% (n = 154) No significant difference between intervention and control groups. Intervention not cost effective Intervention (n = 189): ICU LOS (days) median 10, IQR [5–20]. Day 5 measurements: SOFA score median 3, IQR [0–6]; C-reactive protein (units not provided, likely mg/L) median 79, [41–162]; WBC (units not provided, likely ×109/L) median 13.5, IQR [9.5–18.7]; QALY at 6 months median 0.78, IQR [0.57–0.89]; mortality (%): ICU 21.6; hospital 28 Control (n = 199): ICU LOS (days) median 8, IQR [3–19]. Day 5 measurements: SOFA score median 1.5, [0–7]; C-reactive protein (units not provided, likely mg/L) median 79, IQR [41–162]; WBC (units not provided, likely ×109/L) median 12.9, [9.8–17.1]; QALY at 6 months median 0.72, IQR [0.51–0.85], mortality (%): ICU 18.1, hospital 25.6 The MIPD group incurred incremental costs of EUR 5,312 and an average decrease in 6-month QALYs of 0.03 (range: −0.5 to 0.5) compared to standard dosing, resulting in an incremental cost-effectiveness ratio (ICER) of EUR −177,066 per QALY. This negative ICER indicates that MIPD was more costly and less effective than standard dosing under the observed conditions
Chai et al. [13] (2023)i,j, prospective cohort study, n = 70	Age (years): Subtherapeutic: median 44, IQR [26–56] Supratherapeutic: median 68, IQR [63–76] Weight (kg): Subtherapeutic: median 85, IQR [73–120] Supratherapeutic: median 93, IQR [68–97.5] Renal Function (CrClh (mL/min)): Subtherapeutic: median 163.8, IQR [137.9–230.7] Supratherapeutic: median 69, IQR [37.1–78.3] Illness Severity Scores: Subtherapeutic: APACHE II median 15.5, IQR [9.3–18]; SOFA median 5, IQR [4–7] Supratherapeutic: APACHE II median 16, IQR [10–24]; SOFA median 5, IQR [4.5–7.5]	Piperacillin-tazobactam and meropenem TDM informed by Bayesian software (ID-ODS™)	Unbound meropenem and piperacillin concentrations 100% fT>1–8×MIC. Species EUCAST clinical breakpoint MIC used. In absence of pathogen identification, the EUCAST clinical breakpoint for Pseudomonas aeruginosa was used for piperacillin-tazobactam (16 mg/L) and meropenem (2 mg/L)	24 h achievement of target exposureb: Intervention (n = 20): meropenem 95%. Intervention (n = 13): piperacillin-tazobactam 85% Overall achievement of target exposureb: unknown Time to target attainment: unknown	Culture positivity: 73% (n = 33) Composite clinical cure outcome and mortality reported For participants with repeat TDM (non-therapeutic with standard dosing) (n = 50): clinical cure 94%, discharged from ICU alive 94%, completed beta-lactam course 40%, de-escalation of beta-lactam therapy 14%, ICU mortality 6% Clinical cure defined as a composite outcome of: discharged from ICU alive, completed beta-lactam course, de-escalation of beta-lactam therapy
Chai et al. [14] (2024)j, prospective, non-randomised multicentre interventional trial, n = 156	Age: unavailable Weight: unavailable Renal function: unavailable Illness Severity Scores: Phase 1: APACHE III median 65, IQR [50–77]; PELOD-2 median 4, IQR [2–9]; SOFA unavailable Phase 2: APACHE III median 65, IQR [51.5–76.3]; PELOD-2 median 7, IQR [4–10]; SOFA unavailable	Cefotaxime, flucloxacillin, piperacillin-tazobactam and meropenem TDM informed by Bayesian software (ID-ODS™)	100% fT>MIC. Measured pathogen MIC used. In absence of pathogen identification, the following ECOFFs were used: Unbound cefotaxime 2 mg/L, flucloxacillin 2 mg/L, piperacillin-tazobactam 16 mg/L or meropenem 2 mg/L concentrations 100% fT>MIC	24 h achievement of target exposure: Intervention (n = 48, 52 courses): Phase 2 participants’ initial standard dosing course: 65%. Non-therapeutic Phase 2 participants were then assigned to MIPD dosing (n = 18, 20 courses) with a 24 h achievement of target exposure: 85% Control (n = 108, 123 courses): Phase 1 participants (represented by a different cohort of patients at a different timepoint to Phase 2 participants) initial standard dosing course: 71% Overall achievement of target exposure: Beta-lactam concentrations were collected every 24 h up to 96 h. However, only 24 h achievement of target exposure reported Time to target attainment (h)c: Intervention: Phase 2 median 14.3, IQR [7.3–32.0]. Control: Phase 1 median 8.7, IQR [5.9–96]. However, for participants who failed to achieve target exposures within 24 h: Intervention: Phase 2 median 36.4, IQR [29.8–42.6]. Control: Phase 1 median 96, IQR [96]	Culture positivity: Phase 1 33% (n = 108); Phase 2 25% (n = 48) No difference in mortality and adverse events reported for Phase 1 and Phase 2 participants. An improvement in ICU LOS for Phase 2 participants was found. The MIPD intervention was deemed cost-effective Phase 1 (n = 108): 29% In-hospital mortality up to 60-day 10.1%; in-hospital mortality up to 60-day of participants with subtherapeutic concentrations at 24 h 2.8%; adverse events attributable to intervention 0%; ICU LOS (days) 12.89 (95% CI: 7.92–17.87) Phase 2 (n = 48): In hospital mortality up to 60 days 10.4%; in-hospital mortality up to 60 days of participants with subtherapeutic concentrations at 24 h 11.1%; adverse events attributable to intervention 0%; ICU LOS (days) 10.76 (95% CI: 7.72–13.80) Benefit-cost ratio of intervention for Phase 2 participants: 173.27 (95% CI: 133.86–212.69). For every USD 1 invested in the intervention, there was a return of USD 173.27 in benefits
Haefliger et al. [15] (2025)k, retrospective observational monocentric study, n = 78	Age (years): mean 64, SD±16 Weight (kg): mean 79 SD±18 Renal Function (CrClh (mL/min)): mean 86 SD±52 Illness Severity Scores: APACHE unavailable; SOFA unavailable	Piperacillin-tazobactam simulations using prior TDM datapoints and Bayesian software (TUCUXI)	100% fT>4×MIC of obtained MIC using piperacillin total concentrations. In absence of isolate MIC, a target trough concentration range of 8–32 mg/L was used, which ensures 100% fT>4×MIC for the majority of Enterobacteriaceae identified at the centre (primarily Escherichia coli and Klebsiella pneumoniae), which have an EUCAST ECOFF of 8 mg/L	Initial PTAd: A priori dose 29%, a posteriori dose 83%, identical dosage for all (4,000 mg 8-hourly) 32%, original initial dosage 32% Overall PTAd: A posteriori dose and adjustment 88.5%, original initial dosage and adjustment 43.5% Time to target attainment: unknown	Culture positivity: 14%. No clinical outcomes
Roberts et al. [19] (2025), a prospective cohort study, n = 25	Age (years): mean 56.6, SD±17 Weight (kg): median 85, IQR [76–100] Renal Function (eGFRh (mL/min/1.73 m2): median 88, IQR [65–90] Illness Severity Scores: APACHE II mean 17.1, SD±5.4; SOFA unavailable	TDM informed by Bayesian software (ID-ODS™)	100% fT>4×MIC using obtained MIC. In MIC absence, the EUCAST clinical breakpoint for Pseudomonas aeruginosa was used for unbound piperacillin-tazobactam (16 mg/L) and unbound mero-penem (2 mg/L) concentrations The minimum unbound meropenem and piperacillin PK/PD exposure targets for empiric therapy were a minimum concentration (Cmin) >8 mg/L and >64 mg/L, respectively. Unbound or free Cmin was capped at 45 mg/L and 360 mg/L for meropenem and piperacillin, respectively, for safety reasons	24 h achievement of target exposure: Intervention (n = 14): piperacillin-tazobactam 86%. Intervention (n = 11): meropenem 100% Overall achievement of target exposure: unknown Time to target attainment (h): median 23.3, IQR [19.4–41.52]	Culture positivity: 16% (n = 25) Clinical cure, treatment failure, infection improvement, antimicrobial resistance, adverse events and mortality reported Intervention (n = 25): at antibiotic cessation: clinical resolution 48%, failure 16%, clinical improvement 36%. 7-day post antibiotic cessation: clinical resolution 68%, failure 4%, improvement 28%. Day-14 since intervention: antimicrobial resistance 4%, no adverse events attributable to study antibiotics, ICU mortality 4% (unrelated to antibiotic adverse events 13 days into admission)e Clinical cure defined as: a resolution of all signs and symptoms related to infection Infection improvement defined as: a marked or moderate reduction in severity and/or number of signs or symptoms related to infection

AKI, acute kidney injury; APACHE, Acute Physiology and Chronic Health Evaluation; CI, confidence interval; CrCl, creatinine clearance; CVVH, continuous veno-venous hemofiltration; ECOFF, epidemiological cut-off value; ECMO, extracorporeal membrane oxygenation; eGFR, estimated glomerular filtration rate; EUCAST, European Committee on Antimicrobial Susceptibility Testing; fT>MIC, fraction of time that the unbound beta-lactam concentration remains above the MIC during a dosing interval; IQR, interquartile range; KDIGO, Kidney Disease Improving Global Guidelines; L, litre; LOS, length of stay; mg, milligram; MIC, minimum inhibitory concentration; PELOD-2, Paediatric Logistic Organ Dysfunction-2; PK, pharmacokinetics; PTA, probability of target attainment; QALY, quality adjusted life years; RCT, randomised controlled trial; RRT, renal replacement therapy; SD, standard deviation; SOFA, Sequential Organ Failure Assessment; TDM, therapeutic drug monitoring; WBC, white blood cell count.

^aOnly collective achievement of target exposure for all beta-lactams and ciprofloxacin provided.

^bUnknown achievement of target exposure after initial clinician dosing. Repeat TDM 24–48 h after interventional dose change which does not directly correlate to 24 h achievement of target exposure.

^cIn participants not achieving 24 h achievement of target exposure (n = 36, Phase 1), (n = 18, Phase 2), MIPD for Phase 2 was associated with a reduction in median time to target antibiotic exposure (96 h [96–96 h], 36.4 h [29.8–42.6 h]), respectively.

^dAll MIPD data simulation based on two retrospective TDM samples provided at unknown intervals resulting from original clinical dosing (undisclosed). Overall PTA for a posteriori dosing based on MIPD dose simulation using initial (83%) and subsequent (94%) TDM measurements. Whereas overall PTA for clinical dosing calculated from initial clinical guided dose (32%) and clinical dose adjustment following first TDM (55%).

^eFor comparative purposes, proportions of clinical cure and infection improvement 7-day post antibiotic cessation were combined for clinical cure comparisons.

^fCalculated collectively for imipenem, piperacillin and meropenem patients based on online supplementary material.

^gKDIGO [43].

^heGFR, CrCl [44].

ⁱData only provided for participants exhibiting non-therapeutic exposures.

^jPhase 1 all participants with dosing determined by clinical team, Phase 2 only participants who failed to achieve target exposures within 24 h with MIPD dosing recommendations.

^kData from 80 courses of intermittent piperacillin infusions with two TDM samples, dose adjustment strategies applied using simulations with, without, or with partial TDM data.

Most studies examined the PTA or achievement of the target exposure for the first 24 h of the treatment course. There were mixed findings between a priori, a posteriori and standard dosing (Table 1). Greater PTA than standard care was identified only for a posteriori dosing. For instance, simulated a priori dosing in Haefliger et al. [15] resulted in a PTA of 29%, compared to 32% for standard dosing and 83% for a posteriori dosing. A similar increase in the PTA for patients receiving a posteriori dosing was identified in a pre-post study whereby 85% of phase 2 participants who received MIPD-TDM guided dosing following an initial failure to achieve the target exposure within 24 h of beta-lactam antibiotic commencement, compared to 65% of all participants receiving standard dosing regimens [14]. Only the study conducted by Ewoldt et al. [18] directly compared a posteriori dosing for attainment of target exposure throughout the entire treatment course, identifying no statistically significant difference (61.6% vs. 57.3%, respectively) [9, 13, 14, 36, 45–49]. There was appreciable inter-study heterogeneity given the variability in: reporting outcome metrics, beta-lactam antibiotic studied, TDM measurement of free/total beta-lactam concentrations, beta-lactam infusion mode, target exposure definition (defined by variable %T_>MIC, whereby the MIC was determined by a laboratory measured MIC, epidemiological cut-off, clinical breakpoint or estimation with automated methods such as the Vitek^®), the patient cohorts included, plus the implementation of MIPD and TDM methodology [9, 13–19, 40, 45, 46, 48, 50] (shown in Fig. 1; Table 1; online suppl. Tables 1–3).

Moreover, specific study design may limit the comparison between MIPD and standard care (e.g., pre-post study designs and the variability of included beta-lactam antibiotics) (Table 1). MIPD technical factors such as the use of potentially suboptimal pharmacokinetic models also impact the study results [9, 15–18, 51–65] (online suppl. Table 2). An example of this occurred in one study, whereby a renal replacement therapy specific pharmacokinetic model was not included despite 15.3% of patients in the intervention group required renal replacement therapy [18]. Whereas MIPD implementation factors such as delays exceeding 24 h for MIPD dosing intervention and/or TDM turnaround time may have been a contributory factor in suboptimal target attainment rates in some studies [13–19, 40] (shown in Fig. 1; online suppl. Tables 1 and 3).

Studies Evaluating MIPD Impact on Clinical Cure and Mortality

Current evidence does not routinely support improved clinical cure or survival rates with MIPD. Only one study which used exposure targets of 60–100% fT>_MIC reported clinical cure rates that compared MIPD and standard dosing, which showed a modest improvement (8%) in the cure rate of study participants receiving MIPD [16]. Mortality outcomes were reported by most studies [13, 14, 16–19], but none identified a statistically significant difference between MIPD and standard dosing, with ICU and in-hospital mortality rates almost as high as 40% (Table 1).

Study design and confounding factors further complicate interpretation of mortality outcomes. Cohort studies without randomisation resulted in imbalanced groups (Table 1). For example, the intervention group in the study conducted by Machado et al. [16], had lower illness severity scores which may have contributed to a marginal reduction in mortality [66]. Conversely, higher severity scores in Chai et al. [14] Phase 2 participants, likely explain the increased mortality in the 24 h subtherapeutic cohort (2.8% in Phase 1 versus 11.1% in Phase 2). Variations in target exposure definitions across studies and limited details about the MIC used in some cases (Table 1), may reduce the effectiveness of MIPD when low MICs are present since standard dosing is likely sufficient to achieve target exposure in these patients. Adding to this variability is the heterogeneous patient populations, associated variable infection sites and pathogens, plus inconsistent clinical cure definitions and mortality assessment time points [13–19] (Table 1; online suppl. Table 2).

Studies Evaluating the Impact of MIPD on Organ Dysfunction, Length of Stay, Economic Outcomes, and Antibiotic Resistance Emergence

Other clinically relevant outcomes include prevention of organ dysfunction, length of ICU stay (LOS), cost savings, and antimicrobial resistance prevention. In a sub-analysis of one randomised controlled trial [17] only 5.6% of participants receiving MIPD guided high doses (twofold standard dose) received renal replacement therapy during treatment compared to 20.8% in the low-dose (half standard dose) and 15.8% in the control (standard dose) group. This finding, coupled with an observed improvement in delta Sequential Organ Failure Assessment (SOFA) scores among high-dose participants, suggests that MIPD-facilitated dosing may reduce organ dysfunction rates if a higher exposure if selected. Importantly, no statistically significant differences in adverse event rates were found between MIPD and standard care groups, even when high doses were implemented [17].

ICU LOS (Table 1) results were mixed, with one MIPD randomised controlled trial [18] identifying an increased median ICU LOS for the MIPD group (6.9 days) compared to standard care (5.2 days). On the other hand, another randomised controlled trial [17] identified a reduced ICU LOS by 1.3 days for participants receiving MIPD. A pre-post study [14] showed that MIPD reduced the ICU LOS by 2.1 days for those who received MIPD (Table 1; online suppl. Table 2). These observations may be attributable to target selection, achievement of target exposure and MIPD dosing recommendation turnaround time (shown in Fig. 2, Table 1; online suppl. Table 2) [8, 9, 35, 38, 45, 46, 67].

Fig. 2.

A representation of how suitable pharmacokinetic/pharmacodynamic target selection, pharmacokinetic model suitability and rapid MIPD application may reduce LOS compared to standard dosing through cumulative and conditional outcome improvements. LOS, length of stay; MIPD, model informed precision dosing.

Cost-effectiveness is another key consideration when evaluating investment in MIPD. LOS formed a substantial component of cost analyses in both Ewoldt et al. [18] and Chai et al. [14] and LOS variability likely contributed to the inconsistent cost-effectiveness results (Table 1; online suppl. Table 2). Ewoldt et al. [18] reported a negative incremental cost effectiveness ratio of €-177,066 per quality adjusted life year for MIPD, compared to standard care treated patients. Whereas in Chai et al. [14] identified that for every $1 invested in MIPD treatment on top of standard care costs, there was a return of $173.27 in clinical benefits, mostly due to a reduced LOS. Additionally, cost evaluation approaches differed notably between studies. Ewoldt et al. [18] did not disclose the specific cost constituents used, while Chai et al. [14] included TDM and MIPD costs but did not capture avoidance of adverse event costs; a metric previously considered in other economic evaluations [68–71]. These omissions may have led to underestimation of potential savings and further complicate interpretation of MIPD’s economic value.

Data on antimicrobial resistance are sparse. Roberts et al. [19] reported a 14-day resistance incidence of 4%, lower than the 30-day resistance rate of 8% in a TDM study [8] and the 28-day rate of 12% in a standard dosing study [72]. However, differences in targets (i.e., 100% fT≥_1–4×MIC in the TDM study compared to 100% fT≥_4×MIC in others) and varying timepoints limit direct comparability [8, 19, 72].

Clinical Practice Implications

From a clinician’s perspective, MIPD offers a practical pathway to personalise beta-lactam therapy in critically ill patients, balancing efficacy with safety. However, the utility of MIPD at the bedside depends on several factors: the selection of the pharmacokinetic model, the usability of the software interface, beta-lactam antibiotic target exposure selection, timely access to TDM and dose adjustment with accountment of pre-analytical and analytical issues, appropriate patient selection, plus the clinical expertise to interpret and implement pharmacokinetic model suggestions (online suppl. Table 1) [2, 7, 9, 10, 12, 41, 42, 45, 46, 51, 73, 74].

Pharmacokinetic models should be critically evaluated for relevance to patient cohorts and enquiries made with providers about validation steps [12, 41, 51, 73, 74]. Target selection should be guided by infection site, organism susceptibility, and toxicity risk. Although 100% fT>_MIC is a widely accepted minimum, higher thresholds (e.g., 100% fT>_4xMIC) may be necessary for deep-seated infections or to suppress the emergence of antibiotic resistance [2, 8, 19, 72, 75]. Dose adjustments should ideally occur within 24 h from initiation of the antimicrobial therapy and upon any change in organ function or clinical status, to account for dynamic pharmacokinetic changes in ICU patients [7, 45, 46, 76]. When rapid adjustment is not feasible, clinicians should rely on a priori dosing strategies, toxicity thresholds, and TDM/MIC data when available to guide dosing [2–4, 10, 77, 78]. To facilitate MIPD integration, clinicians must navigate governance and regulatory considerations. Studies such as Roberts et al. [19] suggest that higher daily dosing can be safely implemented when supported by multidisciplinary MIPD teams and institutional oversight [10] (online suppl. Table 1). In practice, some ICUs have low MICs [13, 17], making target attainment feasible even with conventional dosing. Consequently, the likely proportional benefit of MIPD depends on the clinician’s standard dosing practices (including any TDM utilisation), patient cohort and local susceptibility patterns [5, 9, 26, 27, 40, 42, 50]. To maximise MIPD benefit, it should be prioritised for critically ill patients with considerably altered pharmacokinetics (septic shock, burns, reduced or augmented renal function, extremes of weight, extracorporeal therapies, high disease severity, suspicion of toxicity) or those requiring higher exposures (MICs close to the susceptibility breakpoint, Gram-negative ventilator associated pneumonia, deep-seated infections, and foreign-body infections) [9, 13, 14, 36, 45–49]. Given MIC laboratory determination delays and increased infecting bacterial MICs encountered in ICU patients, it is justified to use the clinical breakpoint of the least susceptible potentially causal organism (e.g., Pseudomonas aeruginosa for a nosocomial respiratory infection) when considering the initial empiric MIC [79, 80]. Toxicity thresholds with causality assessment should guide dosing decisions especially at higher, possibly unlicensed exposures [2–4, 77, 78, 81].

Recommendations for the Design of Further MIPD Studies

From a clinical researcher perspective, the advancement of beta-lactam MIPD necessitates rigorous study design, novel methodologies and implementation outcomes assessment. Although no study to date has demonstrated worse clinical outcomes [13–19], there remains considerable scope for building a translational evidence base which better addresses trial design, execution, and reporting [9, 45, 48, 82].

Frameworks can guide both research implementation and practice [83, 84] by building upon existing evidence [13–19]. Defining the clinical contexts in which MIPD provides the greatest benefit is also valuable, such as targeted investigations in critically ill subpopulations. Associated future studies should explore optimal modelling strategies such as the integration of bacterial pharmacodynamic models that account for the specific infecting bacterial isolate [10–12, 85]. Another potential opportunity is to understand the relationship between time to target attainment and clinical cure, as well as the safety implications of higher-than-licensed dosing regimens in specific patient cohorts [8, 13, 14, 19, 38, 47]. Additionally, assessment of earlier initiation of MIPD optimised dosing guided by 100% fT>_MIC or higher targets may improve outcomes when stratified by illness severity [7–9, 45, 46, 86], which must be supported by streamlined sample collection processes and validated analytical workflows [42, 87–91]. In parallel, the development and validation of innovative sampling strategies, artificial intelligence-assisted model adaptation, and covariate selection may enhance precision and scalability [7, 10, 11, 51, 76, 85, 92].

Harmonised outcome definitions [93–97] and assessment of all-cause outcomes using validated causality criteria [81] within pre-specified statistical analysis plans may improve signal detection. With potentially modest benefits, such as improvement of mortality in beta-lactam continuous infusion studies [98], adequate sample sizes are imperative and may require the use of composite endpoints to alleviate associated challenges [93].

Understanding MIPD performance across diverse clinical settings is essential for scalable adoption. However, this will require standardisation of study endpoints [93–97], susceptibility testing [79, 80], availability of validated pharmacokinetic models [10–12, 41, 50, 51], treatment and TDM practice protocols to enable context-specific outcome identification [21, 67, 93, 97], plus stakeholder engagement to establish cost-effectiveness thresholds [99, 100]. The success of the “myPKFiT” platform software may provide an insight that can be adapted to guide these processes [101–106].

Conclusion

MIPD of beta-lactam antibiotics in critical care shows potential to improve clinical outcomes, particularly in subgroups of critically ill patients with complex pharmacokinetic and pharmacodynamic profiles. While current evidence does not demonstrate harm, widespread adoption is limited by methodological variability, inconsistent implementation, and underpowered studies.

Clinical benefits such as improved cure rates, reduced organ dysfunction, shorter hospital stays, and decreased emergence of resistance are more likely when key interdependent factors are optimally aligned. These factors include the selection of appropriate pharmacokinetic/pharmacodynamic targets, the use of validated pharmacokinetic models tailored to patient-specific characteristics, accurate pathogen susceptibility data, and use of continuous or extended beta-lactam infusions. Additionally, operational elements such as the ability to initially dose with MIPD, timely TDM, accurate analytics, and rapid turnaround of dose recommendations are just as essential to ensure effective implementation.

From both clinical and research perspectives, MIPD offers a promising strategy to personalise therapy and support antimicrobial stewardship. To fully answer if beta-lactam MIPD can improve outcomes in critically ill patients, future studies must adopt rigorous designs, standardise intervention components and evaluate outcomes across clinical, microbiological, safety, and economic domains. Suggested actions from this framework should guide both clinical practice and future research, ensuring that MIPD evolves into a reliable tool for improving ICU outcomes.

Acknowledgments

M.U. would like to acknowledge post-doctoral support from the Fundación Alfonso Martín Escudero through the Fúndame fellowship. J.R. would like to acknowledge funding from the Australian National Health and Medical Research Council for a Centre of Research Excellence (APP2007007) and an Investigator Grant (APP2009736) as well as an Advancing Queensland Clinical Fellowship. N.R. would like to acknowledge funding from Medical Research Future Fund 2022-2026 Quality, Safety and Effectiveness of Medicine Use and Medicine Intervention by Pharmacists. A.H. would like to acknowledge a Queensland Health Clinical Research Fellowship (4303825440). No further financial support or sponsorship was received for this contribution. The funders had no role in the design, data collection, data analysis, and reporting of this study. The authors acknowledge the use of Microsoft Copilot to assist in the preparation of Figure 2. This figure were generated based on author-provided input using artificial intelligence assisted image generation tools, with human oversight and verification. Microsoft Copilot is a product of Microsoft Corporation. This publication is independent and not affiliated with Microsoft.

Conflict of Interest Statement

The authors have no conflicts of interest to declare.

Author Contributions

R.O. conducted the literature review and drafted the initial version of the manuscript. A.H. and M.U. contributed to the conceptual structure of the manuscript and critically reviewed subsequent drafts. M.P.H.M., M.C., N.R., J.R., and J.L. provided input on the initial structure and content scope and reviewed draft manuscripts.