Implementation of a kidney protection strategy to prevent acute kidney injury after major surgery in high-risk patients identified by biomarkers: a systematic review and individual participant data meta-analysis of randomized controlled trials

Abstract

Purpose

Acute kidney injury (AKI) is a common complication after major surgery and is associated with increased morbidity and mortality. Kidney protection strategies may help prevent moderate or severe AKI in high-risk patients. This study aims to assess the effect of the Kidney Disease: Improving Global Outcomes (KDIGO) kidney protection strategy for the prevention of AKI in patients after major surgery.

Methods

We conducted a systematic review and individual participant data (IPD) meta-analysis of randomized controlled trials (RCTs) comparing the kidney protection strategy recommended by international guidelines consisting of hemodynamic and fluid status optimization, avoidance of nephrotoxins or radiocontrast agents, regular monitoring of kidney function, and glycemic control to standard care in high-risk patients after major surgery with an enrichment strategy based on renal biomarkers. The primary outcome was moderate or severe AKI (KDIGO stage ≥ 2) within 72 h after surgery. MEDLINE via PubMed, Web of Science, and the Cochrane Central Register of Controlled Trials were searched from January 1, 2000, to September 1, 2025. References of eligible trials and related reviews were hand-searched. Two reviewers independently assessed trial quality using the Cochrane Risk of Bias tool version 2.0. Certainty of the evidence was assessed using GRADE. IPD were pooled. Odds ratios (ORs) and mean difference with 95% confidence intervals (CIs) were computed with one-stage IPD meta-analysis. Heterogeneity was assessed by I² and Cochran’s Q.

Results

We identified four RCTs, two single-center trials and two multinational-multicenter trials. We pooled IPD from all four trials. The final cohort included 1,851 participants with 921 participants in the intervention group and 930 participants in the control group. Moderate or severe AKI occurred significantly less frequently in the intervention group (162/918 participants (17.7%)) compared to the control group (252/929 participants (27.1%)) (OR 0.55, 95% CI 0.44–0.70; p < 0.0001). There was no evidence of heterogeneity across studies (p = 0.7309, I² = 0.0%, τ² = 0). Secondary endpoints varied across trials and did not demonstrate major differences between groups. When measured, the intervention tended to result in fewer persistent AKI events and larger decreases in renal tubular stress biomarkers.

Conclusion

The implementation of a kidney protection strategy reduces the rates of moderate or severe AKI in biomarker-enriched high-risk patients after major surgery compared to standard of care, while the incremental clinical value of biomarker-guided selection itself remains uncertain.

Registration

The study was registered at the international, prospective register of systematic reviews PROSPERO. Identifier: PROSPERO 2025 CRD420251138328.

Visual abstract

Supplementary Information

The online version contains supplementary material available at 10.1007/s00134-026-08399-1.

Introduction

Acute kidney injury (AKI) is a common and serious complication after major surgery, associated with increased morbidity, mortality, prolonged hospitalization, and long-term risk of chronic kidney disease (CKD) [1, 2]. Early identification of patients at high risk and the application of a kidney protection strategy have the potential to mitigate these adverse outcomes [3]. The Kidney Disease: Improving Global Outcomes (KDIGO) AKI guideline recommends application of a kidney protection strategy in patients at high risk for AKI, including optimization of volume status and hemodynamics, advanced hemodynamic monitoring, avoidance of nephrotoxins and radiocontrast agents, when possible, as well as glycemic control and regular assessment of kidney function [4]. These recommendations are based on clinical risk assessment and are intended as good clinical practice for patients judged to be at increased perioperative AKI risk.

In recent years, biomarkers have been proposed to improve identification of patients at particularly high risk for AKI and to enable targeted initiation of preventive measures [5]. The renal tubular stress biomarkers, tissue inhibitor of metalloproteinases-2 and insulin-like growth factor binding protein 7 ([TIMP-2]*[IGFBP7]), identify patients at high risk for AKI and can predict the development of moderate and severe AKI [6]. Previous randomized controlled trials (RCTs) evaluated whether applying the kidney protection strategy in patients at high risk identified by elevated levels of [TIMP2]*[IGFBP7] can reduce the occurrence of AKI when compared with standard care [7–10]. These trials were designed as biomarker-enriched prevention studies and therefore inform the effectiveness of bundle implementation in a selected high-risk population rather than the incremental prognostic or clinical utility of the biomarker itself.

While these trials have reported concordant results, consistent findings across multiple RCTs are most robustly synthesized using meta-analytic approaches. Such evidence informs clinical practice by enabling generation of large sample sizes, combining treatment effect estimates from different studies and assessing risk of bias of included studies, as well as between-study heterogeneity of treatment effects (non-random variation) [11, 12]. Furthermore, meta-analysis can guide identification of patient groups at high risk or susceptible to certain interventions, due to heterogeneity of treatment effect. However, in intensive care trials, a heterogeneity of study populations, clinical phenotypes, and treatment effects is often high [13, 14]. The cause of between-study heterogeneity can only reliably be investigated using individual participant data (IPD), but not study-level data [12]. IPD meta-analysis allows harmonization of AKI definitions, investigation of treatment effect heterogeneity across subgroups (e.g., surgery type, baseline risk, biomarker levels), and more precise estimation of overall effect and low-incidence secondary outcomes [15]. However, IPD meta-analysis also requires careful consideration of the independence, diversity, and external validity of contributing trials.

No IPD meta-analysis on the implementation of the kidney protection strategy recommended by KDIGO has yet been published. The aim of the present IPD meta-analysis of RCTs was to assess the effects of this kidney protection strategy within biomarker-enriched high-risk patients after major surgery, without addressing the independent clinical utility of biomarker testing in this cohort.

Methods

Study design

This systematic review and meta-analysis followed the Preferred Reporting Items for Systematic reviews and Meta-Analyses for IPD (Table S1), as well as the PRISMA-IPD checklist for abstracts [16, 17]. The protocol was registered on PROSPERO (CRD420251138328) on September 11th 2025 and followed a pre-specified statistical analysis plan as also registered via PROSPERO.

Eligibility criteria

We included all RCTs that enrolled adult patients undergoing major surgery (cardiac and/or non-cardiac) and fulfilled all of the following criteria:

1. Use of validated AKI biomarkers to identify patients at high risk for AKI (enrichment strategy)

2. Randomization of participants to a structured KDIGO-based kidney protection strategy versus standard care

3. Collection of postoperative AKI outcomes according to KDIGO criteria.

Search strategy

We searched MEDLINE via PubMed, Embase, CINAHL (EBSCO), SCI-EXPANDED, SSCI, A&HCI, CPI-SSH, ESCI (Web of Science), and CENTRAL (Cochrane Library) from Jan 1 st 2000 until 1 st Sept 2025, using a search algorithm developed for the purpose of this study and adapted to each database (Table S2 in the Supplement). The following supplementary searches were conducted to ensure all eligible trials were identified and included:

1. Trial registries: ClinicalTrials.gov and WHO ICTRP to identify ongoing/unpublished RCTs.

2. Grey literature: Web of Science Conference Proceedings; major intensive care and nephrology society abstracts (SCCM, ESICM, ASN, ISN)

3. Hand searches: Reference lists of included studies and relevant reviews; forward citation tracking (Web of Science/Google Scholar).

Selection of studies and risk-of-bias assessment

Selection was conducted independently by two reviewers (TvG and MS) on titles and abstracts first and then on the full text. After removing the duplicates, titles and abstracts were screened and selected according to eligibility criteria. The remaining publications were checked for completeness and plausibility before they were combined in a database (EndNote). For each included RCT, the corresponding author was contacted to provide fully anonymized IPD as well as format, coding and definition of all variables. Risk of bias in each trial was evaluated by two independent reviewers (MM and DT) using the updated risk-of-bias (RoB) tool developed by Cochrane (Cochrane RoB 2.0 tool, see Supplement). The first author (TvG) guided a consensus process for any conflicting evaluations with the two authors performing RoB assessment (MM and DT). Certainty of the evidence for all outcomes was assessed (TvG, MM, DT) using the GRADE approach with GRADEpro GDT (GRADEpro GDT: GRADEpro Guideline Development Tool [Software]; McMaster University, 2015 (developed by Evidence Prime, Inc.; Available from gradepro.org).

Individual participant data acquisition and management

Three of the included studies were performed or coordinated by authors from our study group (94% of the included patients); hence, IPD were accessible and databases similar. We invited the principal investigator of the fourth RCT identified to share de-identified individual participant-level data. Requested variables included demographic data, comorbidities, surgical details, intraoperative hemodynamics, biomarker levels, indicators of kidney protection strategy initiation and compliance, details of kidney protection strategy components, postoperative outcomes, and follow-up duration. We standardized variables across trials. We standardized variables across trials; however, given all trials were coordinated by German study sites, units were already reported in International Systems of Units (SI) units only and did not require transformation. AKI stages were defined according to KDIGO 2012 criteria in all included studies. No missing data were imputed.

Study outcomes

The primary endpoint was incidence of moderate or severe AKI (KDIGO stages 2 or 3) within 72 h after major surgery. AKI was defined according to KDIGO criteria, using both serum creatinine and urinary output data. Secondary outcomes included overall AKI incidence (KDIGO stage ≥ 1), AKI stages, occurrence of transient or persistent AKI, need for renal replacement therapy (RRT), intensive care unit (ICU) length of stay, hospital length of stay, as well as long-term outcomes of RRT requirement, mortality and a combined endpoint of Major Adverse Kidney Events (MAKE) at day 90. MAKE was defined as the composite of death, any use of RRT during the first 30 or 90 days (MAKE₃₀, MAKE₉₀) after randomization, and persistent renal dysfunction (defined as serum creatinine ≥ 1.5 × baseline value) at day 90. Change in biomarker values at 12 h after randomization was also analyzed. All analyses were predefined and registered with the study registry prior to the start of the project.

Statistical analysis

Statistical analyses were performed for each outcome of interest using IPD. Baseline characteristics are summarized by trial and overall. Categorical variables are presented as absolute and relative frequencies and continuous variables as median and 25% and 75% quantile. An intention-to-treat analysis was used for all outcomes, whereby all patients were analyzed in the groups to which they were randomized. The measures of treatment effect were odds ratio for moderate or severe AKI (KDIGO stages 2 or 3) within 72 h after major surgery, overall AKI incidence (KDIGO stage ≥ 1), AKI stages, occurrence of transient or persistent AKI, need for RRT, as well as long-term outcomes of RRT requirement and mortality and a combined endpoint of MAKE₃₀ and MAKE₉₀ and mean difference for change in biomarker values at 12 h after randomization, intensive care unit (ICU) length of stay, and hospital length of stay.

Primary analysis was a one-stage IPD meta-analysis. A two-stage IPD meta-analysis was conducted as a sensitivity analysis. For mortality, only a two-stage individual patient data meta-analysis was performed, as no IPD was available from the BigpAK-1 study. Heterogeneity between studies was assessed using Cochrane’s Q homogeneity test, as well as ${\tau }_2$ and $I^2$. If significant heterogeneity was detected, a random effects model was used.

One-stage individual patient data meta-analysis consisted of a linear mixed model for mean differences and a generalized mixed model for odds ratios. The model included a random intercept for trial and in case of heterogeneity a random effect for intervention. Logit was used as the link function in the generalized linear model with the exception of AKI stages, where cumulative logit was used.

We conducted sensitivity analyses for the primary outcome both in an as-treated, as well as in a per-protocol population. While the intention-to-treat (ITT) population (primary analysis set) included all randomized patients as attributed by randomization. The per-protocol population compared patients receiving the full KDIGO kidney protection strategy to those who did not receive all components of the kidney protection strategy, whatever the randomization arm. The per-protocol analysis evaluated participants based on randomization group, assessing participants to be treated per protocol if all bundle components were implemented.

In subgroup analyses, we explored whether the effect of the KDIGO kidney protection strategy on the primary outcome varied according to baseline kidney function (high and low baseline eGFR; low eGFR was defined as a baseline eGFR < 60 ml/min/1.73 m²) or type of surgery (cardiac and non-cardiac surgery). Within each subgroup, one-stage IPD meta-analysis for the primary and secondary outcomes was performed.

Analyses were performed with the use of SAS 9.4 (SAS Institute, Cary, North Carolina, USA) and R 4.4.1.

Results

Selection process and general characteristics

From the 708 references identified by the search strategy, we included 4 RCTs fulfilling our eligibility criteria: The PrevAKI RCT, The PrevAKI-Multicenter RCT (“PrevAKI-2”), The BigpAK Study (“BigpAK-1”), and the BigpAK-2 Trial (Fig. 1). Reasons for exclusion are reported in Fig. S1. At time of literature search, the BigpAK-2 RCT was not yet published; however, its study protocol and statistical analysis plan were identified in the literature search. Given the study was conducted by our research group, access to the final version of the manuscript as accepted for publication and to the database was possible. The four included trials provided IPD for all randomized patients (921 patients in the intervention group, 930 patients in the control group), and there was no eligible trial not providing IPD. Study design and cohorts of included studies are summarized in the Supplement (Table S3). Interventions were highly standardized across studies and are summarized in the Supplement (Table S4). Approximately 94% of included participants originated from trials coordinated by the same investigative group (PrevAKI, PrevAKI-2, BigpAK-2), while the remaining trial (BigpAK-1) evaluated a similar biomarker-guided strategy in a closely aligned perioperative setting.

Fig. 1.

Literature search and study workflow

All of the included studies measured urinary TIMP-2 and IGFBP7 in the commercially available NephroCheck® assay marketed by BioMérieux/Astute Medical.

Comparison of patient characteristics at randomization did not show a baseline imbalance between groups (Table 1).

Table 1.

Baseline characteristics of included studies and of the combined cohort

	Intervention (n = 921)	Control (n = 930)
Patient demographics
Age (years), median (Q1, Q3)	71 (62, 77) [N = 910]	70 (62, 76) [N = 924]
Female sex, n (%)	300 (33.0) [N = 910]	287 (31.0) [N = 924]
Preoperative serum creatinine (mg/dl), median (Q1, Q3)	0.90 (0.75, 1.10) [N = 775]	0.90 (0.78, 1.10) [N = 797]
Preoperative eGFR (ml/min/1.73 m2), median (Q1, Q3)	79.0 (61.0, 96.4) [N = 775]	82.0 (62.1, 101.5) [N = 735]
[TIMP-2]*[IGFBP7] at randomization, median (Q1, Q3)	0.63 (0.41, 1.15) [N = 775]	0.63 (0.43, 1.15) [N = 920]
Comorbidities
American Society of Anesthesiology (ASA) Score, no. (%)a
1 (healthy)	3 (0.4) [N = 775]	15 (1.9) [N = 791]
2 (mild or moderate illness)	139 (17.9) [N = 775]	146 (18.4) [N = 791]
3 (severe general illness)	520 (67.1) [N = 775]	499 (63.0) [N = 791]
4 (life-threatening general illness)	113 (14.6) [N = 775]	131 (16.6) [N = 791]
Hypertension, n (%)	630 (74.3) [N = 848]	601 (70.0) [N = 852]
Congestive heart failure, n (%)	216 (30.2) [N = 715]	212 (28.6) [N = 741]
Chronic obstructive pulmonary disease, n (%)	91 (10.7) [N = 849]	99 (11.5) [N = 862]
Chronic kidney disease, n (%)	170 (18.6) [N = 916]	159 (17.2) [N = 927]
Surgical category, n (%)
General/abdominal	253 (27.7) [N = 912]	269 (29.1) [N = 926]
Cardiac	471 (51.6) [N = 912]	465 (50.2) [N = 926]
Vascular	87 (9.5) [N = 912]	92 (9.9) [N = 926]
Thoracic	25 (2.7) [N = 912]	27 (2.9) [N = 926]
Neuro	5 (0.6) [N = 912]	2 (0.2) [N = 926]
Orthopedics	10 (1.1) [N = 912]	13 (1.4) [N = 926]
Obstetrics/gynecology	10 (1.1) [N = 912]	14 (1.5) [N = 926]
Urology	25 (2.7) [N = 912]	20 (2.2) [N = 926]
Plastics	1 (0.1) [N = 912]	3 (0.3) [N = 926]
Other	25 (2.7) [N = 912]	21 (2.3) [N = 926]

Risk of bias was judged as “low” or “with some concern” in the included trials (Fig. S2) with concerns arising from randomization process or deviations from the intended intervention in one study each (Table S5).

Primary outcome

Moderate or severe AKI occurred significantly less frequently in the intervention group (162/918 patients (17.7%)) compared to the control group (252/929 patients (27.1%)) (odds ratio [OR] 0.55, 95% CI 0.44–0.70; p < 0.0001, high certainty of the evidence). Results were similar between one-stage and two-stage models. The absolute risk reduction was 9.4%, resulting in a number needed to treat of 10.6 to prevent one case of moderate or severe AKI. There was no evidence of heterogeneity across studies (I² = 0.0%, τ² = 0, p = 0.7309). Figure 2 displays the forest plot of the primary outcome in the intention-to-treat population. Table 2 displays the primary and secondary outcomes.

Fig. 2.

Forest plot of the primary outcome in the intention-to-treat population

Table 2.

Primary and secondary outcomes

	Intervention (n = 921)	Control (n = 930)	Effect estimate (95% CI)	p value
Moderate or severe AKI within 72 h
Moderate or severe AKI, n (%), [N = 1847]	162 (17.7) [N = 918]	252 (27.1) [N = 929]	OR 0.55 (0.44, 0.70)	< 0.0001
Secondary outcomes: renal endpoints
Any AKI within 72 h
Any AKI, n (%), [N = 1847]	385 (41.9) [N = 918]	460 (49.5) [N = 929]	OR 0.71 (0.46, 1.08)
AKI stages
Stage 1, n (% of any AKI), [N = 995]	228 (48.0) [N = 475]	196 (37.6) [N = 520]	OR 0.53 (0.42, 0.67)
Stage 2, n (% of any AKI), [N = 995]	106 (22.3) [N = 475]	197 (37.9) [N = 520]
Stage 3, n (% of any AKI), [N = 995]	50 (10.5) [N = 475]	66 (12.7) [N = 520]
Duration of moderate or severe AKI
Persistent (> 48 h), n (%) [N = 467]	52 (24.8) [N = 210]	73 (28.4) [N = 257]	OR 0.83 (0.55, 1.26)
Secondary outcomes: clinical endpoints
Change in biomarker values during 12 h following initial measurement, median (Q1, Q3), [N = 1271]	0.23 (− 0.17, 0.69) [N = 640]	0.21 (− 0.26, 0.64) [N = 631]	MD 0.11 (− 0.17, 0.40)
RRT up to day 30, n (%), [N = 1702]	36 (4.3) [N = 848]	38 (4.5) [N = 854]	OR 0.95 (0.60, 1.52)
RRT up to day 90, n (%), [N = 1679]	34 (4.1) [N = 833]	38 (4.5) [N = 846]	OR 0.90 (0.56, 1.44)
Deaths until day 30 (two-stage model), n (%), [N = 1675]	31 (3.7) [N = 833]	32 (3.8) [N = 842]	OR 0.98 (0.59, 1.62)
Deaths until day 90 (two-stage model), n (%), [N = 1626]	45 (5.6) [N = 804]	43 (5.2) [N = 822]	OR 1.07 (0.70, 1.65)
ICU length of stay, median (Q1, Q3), [N = 1834]	3.0 (1.2, 6.1) [N = 910]	2.9 (1.1, 6.0) [N = 924]	MD 2.0 (− 1.76, 5.76)
Hospital length of stay, median (Q1, Q3), [N = 1830]	14.0 (9.0, 25.4) [N = 909]	14.6 (9.5, 25.0) [N = 921]	MD 1.37 (− 2.16, 4.89)
Major adverse kidney event until day 30 (MAKE30), n (%), [N = 1591]	76 (9.6) [N = 793]	65 (8.2) [NN = 798]	OR 1.19 (0.84, 1.69)
Major adverse kidney event until day 90 (MAKE90), n (%), [N = 1522]	83 (11.0) [N = 758]	79 (10.3) [N = 769]	OR 1.07 (0.77, 1.48)

Numbers and percentages are provided where they are not missing. It is, therefore, possible that the figures do not add up to the total number of the cohort. Outcomes were calculated as one-stage models unless otherwise indicated (mortality outcomes)

ICU intensive care unit, AD absolute difference, OR odds ratio, HR hazard ratio, HL Hodges–Lehmann estimator, MD mean difference, SD standard deviation

Secondary outcomes

Secondary outcomes did not differ between groups (Table 2). Overall rates of AKI were similar between groups. However, patients in the intervention group had more stage 1 AKI, whereas controls had more moderate or severe AKI (KDIGO stage 2 or 3). There were no differences in death, dialysis, or MAKE₉₀ endpoints. ICU and in-hospital length of stay did not differ. Forest plots for all secondary outcomes are presented in the supplementary appendix (Figs. S2–S1).

Sensitivity analysis

Moderate or severe AKI in the per-protocol cohort was similar to the ITT analysis (OR 0.45, 95% CI 0.32–0.63) with consistent effect directions in all included studies (Fig. S12, Table S6). Similar results were observed in a per-protocol analysis (OR 0.38, 95% CI 0.26–0.56) with consistent effect directions in all included studies (Fig. S13, Table S7). Finally, we conducted a sensitivity analysis of the ITT cohort excluding BigpAK-2 study data as this study predominately contributed patients to the IPD meta-analysis (Fig. S14). This analysis found consistent results with an OR of 0.51 (95% CI 0.36–0.73).

Subgroup analyses

We performed four subgroup analyses (Fig. S15). First, the treatment effect was assessed in subgroups with low baseline eGFR (defined as an eGFR of < 60 ml/min/1.73 m²) and high eGFR (defined as an eGFR ≥ 60 ml/min/1.73 m²). Second, the treatment effect was assessed in subgroups based on surgical category (cardiac and non-cardiac surgery). The treatment effect was consistent among all investigated subgroups; however, treatment effects were stronger in the subgroup of cardiac surgery (OR 0.48, 95% CI 0.35–0.66), but not in the subgroup of patients with low eGFR (OR 0.60, 95% CI 0.42–0.86).

Ancillary analyses

Given the clinical relevance of RRT as a patient-centered and health-economic outcome, we additionally performed post hoc sample size calculations to estimate the number of participants required to detect an RRT reduction based on observed event rates and effect estimates from adherence-dependent analyses (Table S8).

Certainty of the evidence

The Summary of Findings (SoF) Table reporting the evaluation of the quality of evidence for the primary and secondary outcomes is presented in the Supplement (Table S9). The certainty of the evidence was high for the primary outcome (both in the intention-to-treat and per-protocol cohorts) and duration of RRT. We found moderate certainty of the evidence for the development of any stage of AKI. The certainty of the evidence was low or very low for all other secondary outcomes.

Discussion

This IPD meta-analysis includes primary source data of four RCTs of patients after major surgery with high AKI risk, as identified by urinary renal biomarkers. We found strong evidence supporting the use of a KDIGO kidney protection strategy in surgical patients at high risk for AKI to reduce rates of moderate or severe AKI compared with standard care. Importantly, the included trials were designed as biomarker-enriched prevention studies rather than evaluations of the incremental diagnostic or prognostic value of [TIMP-2]*[IGFBP7]. Consequently, the present findings should be interpreted as evidence supporting the effectiveness of structured KDIGO-based preventive care in a selected high-risk population, rather than proof of benefit attributable to biomarker-guided decision-making itself.

We found an absolute risk reduction of 9.4%, resulting in a number needed to treat of 10.6 to prevent one case of moderate or severe AKI, indicating a clinically meaningful effect on AKI severity. This demonstrates a significant reduction in rates of moderate or severe AKI with the study intervention, which is in line with findings of the included studies. This effect was consistent both when employing one-step or two-step models, indicating no significant heterogeneity of treatment effect alongside low parameters of between-study heterogeneity (e.g., very low I² test value). Notably, adherence to the kidney protection strategy emerged as a critical determinant of effectiveness. Compliance with the full care strategy remained incomplete across trials (e.g., approximately 47% in BigpAK-2), and exploratory per-protocol and as-treated analyses suggested directionally consistent reductions in severe AKI and RRT when bundle delivery was achieved. These findings underscore that successful implementation, rather than risk identification alone, is central to real-world clinical benefit. As expected, the as-treated analysis demonstrates an imbalance between patients that received the full intervention. This imbalance is consistent across all included studies and reflects the reality that individual components of the intervention were not mandated, but rather clinicians were required to consider each element. It may have been impossible to implement certain parts of the study intervention in many patients; for example, if conflicting clinical goals exist (e.g., requirement for application of nephrotoxic antibiotic therapy if no alternative anti-infective agent is available). This emphasizes the difficulty of implementing an allegedly easy-to-implement compound clinical intervention in critically ill patients. It is noteworthy that, despite not having received the full intervention, many of the patients deemed as control patients in the per-protocol cohort received some or even many parts of the study intervention, but failed to achieve adherence to all components of the kidney protection strategy.

Nevertheless, no differences were observed in intention-to-treat analyses of secondary outcomes exploring long-term benefits such as mortality, RRT, or MAKE at day 90 after randomization. Although the more pronounced decrease in renal tubular stress biomarkers [TIMP-2]*[IGFBP7] suggests treatment response with the study intervention, this is not a patient-centered outcome. The intervention group had slightly lower rates of persistent AKI as compared to controls, but this difference was not significant. Although the analysis included more than 1,800 patients, this IPD meta-analysis remains underpowered to detect modest but clinically meaningful differences in long-term clinical outcomes, such as mortality and MAKE. This is based on low event rates, and significantly larger trials would be needed to explore the differences in these outcomes [18]. The absence of statistically significant effects on long-term outcomes in this cohort of approximately 2000 patients should, therefore, not be interpreted as evidence of futility or lack of biological relevance. Finally, our exploratory sample size calculations demonstrate considerable reductions in day 30 and day 90 rates of RRT in both per-protocol and as-treated analyses. This emphasizes not only the importance of care strategy adherence, but also demonstrates that considerably smaller sample sizes would be required for these outcomes when high rates of care strategy adherence can be achieved. Therefore, while prevention of moderate or severe AKI remains a clinically meaningful endpoint on its own, downstream benefits may require substantially larger pragmatic trials focused on implementation fidelity and long-term follow-up. Importantly, moderate or severe AKI itself represents a clinically relevant outcome, as AKI severity is strongly and consistently associated with mortality, long-term kidney dysfunction, and healthcare utilization. We would argue that prevention of higher-stage AKI, therefore, has intrinsic clinical value, even in the absence of demonstrable effects on downstream endpoints within the available sample size. In addition, MAKE₉₀ was not designed for and should not be used in prevention trials [19]. To detect an effect on MAKE90 in prevention trials, more than 10,000 patients are needed in such trials. Because RRT represents a clinically meaningful and economically relevant endpoint, we also performed additional sample size calculations based on observed RRT incidence and adherence-dependent effect estimates. These analyses indicated that very large trials would still be required to demonstrate statistically significant RRT reduction with low fidelity to the study intervention, but realistic sample sizes in cohorts with high adherence to the preventive strategy. Therefore, the occurrence of AKI is an adequate endpoint in prevention trials [20]. Based on data from this IPD meta-analysis, we performed sample size calculations for an RCT with 1:1 randomization, able to detect a difference in outcomes as observed in our data with a power of 90. To detect a difference in MAKE30, a sample size of 6340 participants per group would be required, and this number grows to 31,573 participants per group for MAKE90. To detect a mortality benefit by day 90, a sample size of 59,268 participants per group would be required. Differences in renal replacement therapy requirement by day 90 would require 38,786 participants per group. These analyses support our statement that significantly larger trials would be needed to demonstrate differences in such outcomes, and this appears likely hard to conduct.

Notably, all included studies identified high-risk patients by using the tubular stress biomarkers TIMP-2 and IGFBP7. These biomarkers identify patients at high risk for AKI and predict the development of moderate and severe AKI [5]. However, the biomarker signal does not indicate irreversible damage, and AKI may still be prevented with an early implementation of the KDIGO kidney protection strategy.

Our results are consistent with previous aggregated data meta-analyses in the field; however, no prior meta-analysis included the recently published BigpAK-2 study data [21–24]. Besides adding the data of this recently published study, our IPD meta-analyses go beyond previous studies and provide stronger evidence based on IPD of four RCTs. IPD meta-analyses provide more robust and flexible analyses than aggregated data meta-analyses because they use raw participant-level data, allowing for better adjustment of confounding variables and assessment of potential heterogeneity of treatment effect.

GRADE methodology assessed certainty of the evidence. Certainty of the evidence was found to be high for moderate or severe AKI, as well as for duration of RRT and moderate for overall AKI rates. Besides this, certainty of the evidence remains low for all other investigated outcomes, despite IPD meta-analysis of four RCTs. This was mostly due to wide confidence intervals of outcome measures, varied direction of treatment effect across studies, the impossibility of blinding for this intervention, as well as low event rates for clinical outcomes, such as mortality, MAKE, or long-term RRT requirement. Risk of bias was low or with some concerns in most study domains. Most notably, blinding to the intervention is not possible, imposing some risk of bias. Although formal risk-of-bias assessments only showed low risk or some concerns, we acknowledge that structural proximity between investigators, trials, and assessors may limit full independence of interpretative judgments. Independent external assessment would further strengthen certainty in future IPD meta-analyses.

Meta-analyses of IPD can also explore outcomes in important subgroups and suggest which population may derive the greatest benefit of a specific intervention, which is limited in aggregated data meta-analyses [25]. In our study, we found that the treatment effect was consistent in all investigated subgroups, but especially pronounced in patients undergoing cardiac surgery. These findings are in line with prior studies identifying these subgroups to be at especially high risk to experience AKI [26–29].

This study has several limitations. First, a key limitation of this IPD meta-analysis is that the included trials are conceptually aligned and originate largely from a closely connected investigative environment. While this internal consistency explains the absence of heterogeneity and supports reproducibility within this framework, it limits external validity. The findings may not be generalizable to centers with different baseline standards of perioperative care, monitoring capabilities, staffing models, or access to biomarker testing, particularly in low-resource settings. Second, even an IPD study cannot mitigate limitations of included studies. For example, although patients were representative of those undergoing major surgery in Europe, the study population may not be representative of patients in low- and middle-income countries or jurisdictions with different ethnic distributions or background care. The intervention itself also warrants careful interpretation. KDIGO-recommended kidney-protective strategy reflects good clinical practice for patients at increased AKI risk. An important conceptual consideration relates to the positioning of the KDIGO kidney protection strategy as a biomarker-triggered intervention rather than a universally applied standard of care. Given the high negative predictive value of this biomarker panel, routine application of an intensive, protocolized care strategy to biomarker-negative patients, who have a low residual risk, may not align with the intent of targeting preventive efforts to those most likely to benefit [30]. Importantly, biomarker negativity did not mandate withholding individual components of kidney-protective care, which could still be applied based on clinical judgment and other indications. Nevertheless, the present analysis primarily informs the effectiveness of delivering a structured KDIGO-based care strategy to a biomarker-defined high-risk subgroup, and does not resolve the question of optimal preventive strategies for patients with similar clinical risk profiles but without biomarker elevation.

In the included trials, biomarkers were used for patient enrichment and the strategy was delivered as a whole in patients in the intervention group. Consequently, the present analysis primarily informs the effect of the kidney protection strategy to an enriched biomarker-positive patient population, rather than the value of these measures for all patients. The implications for patients with similar clinical risk who do not meet the biomarker threshold remain uncertain. Third, performance bias is another important consideration. Blinding of a multifaceted kidney protection strategy is not feasible, and biomarker-positive patients randomized to the intervention likely received increased clinical attention. Although this reflects real-world implementation of such strategies, it introduces an unavoidable risk of bias. In this specific context, biomarker-based identification of high-risk patients may have also introduced differential clinical attention as a non-measured co-intervention, including intensified monitoring, reassessment, and clinician engagement. Such effects cannot be disentangled from bundle components and may partially inflate observed treatment effects. Fourth, all included studies excluded patients with advanced CKD (stage 4) or end-stage kidney disease. Hence, we could not investigate the effectiveness of the kidney protection strategy in such patients as part of this IPD meta-analysis. However, the AKI risk may not be modifiable any more in patients with very advanced CKD, and they may therefore not be suited for simultaneous recruitment with patients without CKD or less severe stages of CKD (stage 1–3). However, future trials will be required to investigate whether the implementation of the KDIGO kidney protection strategy in advanced CKD patients can prevent AKI and improve long-term outcomes. Fifth, assessing multiple secondary outcomes introduces risk for false-positive findings. Hence, all secondary outcome analyses must be regarded as exploratory analyses. Finally, we could not perform cost-effectiveness analyses because such data were not available except for only one study, rendering need for meta-analysis redundant.

While [TIMP-2]*[IGFBP7] has demonstrated robust performance for early detection of tubular stress and risk stratification in patients at high risk for acute kidney injury, its broader implementation warrants consideration of cost and availability, particularly in resource-limited settings. The assay requires dedicated platforms and carries a considerable per-test cost compared with clinical risk assessment, which may limit accessibility in low- and middle-income countries; however, this needs to be calculated against potential cost-savings due to less human work required and potentially lower cost for treatment of AKI. These factors may impact the external validity of our findings across healthcare systems with differing economic and infrastructural capacities. Nonetheless, targeted use of [TIMP-2]*[IGFBP7] in carefully selected high-risk populations may still prove cost-effective by enabling earlier preventive strategies and potentially reducing downstream costs associated with established AKI. Future studies should focus on external validation in diverse geographic and economic settings, as well as formal cost-effectiveness analyses, to better define the role of [TIMP-2]*[IGFBP7] in global AKI care pathways.

Despite these limitations, the present IPD meta-analysis provides the most comprehensive synthesis to date of biomarker-based KDIGO kidney protection strategies in surgical patients. The findings demonstrate internal consistency and reproducibility and support prevention of moderate or severe AKI as a meaningful and achievable target. Future research should focus on pragmatic, multicenter implementation studies in more diverse healthcare settings, evaluation of cost-effectiveness and feasibility, and trials powered for patient-centered outcomes. In conclusion, this IPD meta-analysis of four RCTs provides high-certainty evidence that implementation of a structured KDIGO kidney protection strategy can reduce moderate or severe AKI in biomarker-enriched high-risk patients undergoing major surgery. Clinical effectiveness appears closely linked to adherence with care strategy delivery, while the independent incremental value of biomarker-guided remains uncertain. While downstream patient-centered benefits were not demonstrated, this sample size did not achieve adequate power to do so, and prevention of higher stage AKI itself represents a clinically relevant outcome. Further large-scale and pragmatic studies are required to clarify effects on long-term outcomes, generalizability of findings, and cost-effectiveness.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Concept and design: all authors. Acquisition, analysis, or interpretation of data: TvG, EB, MM, DJT, IG, LGF, HG, JAK, AZ. Literature search: MS. Risk of bias assessment: TvG, MM, DJT. Statistical analysis: TvG, EB, and MM. Data access, responsibility, and analysis: TvG, EB, MM, and AZ; all had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Drafting of manuscript: TvG, EB, and MM drafted the first version of the manuscript. All authors provided critical feedback to the draft manuscript. All authors carefully read and approved the final version.

Funding

Open Access funding enabled and organized by Projekt DEAL. No external funding was received for this study. TvG was supported by a rotational position funded by the German Research Foundation (Deutsche Forschungsgemeinschaft (DFG))—grant number 493624047. AZ was supported by the German Research Foundation (ZA 428/28-1, ZA 428/29-1, ZA 428/30-1). The PrevAKI-1 trial was supported by the German Research Foundation (428/6-1 to Alexander Zarbock), the European Society of Intensive Care Medicine (ESICM), the Innovative Medizinische Forschung grant, and an unrestricted research grant from Astute Medical. The PrevAKI-2 trial was supported by the European Society of Intensive Care Medicine (ESICM) and by the German Research Foundation (DFG) (ZA428/14-1, KFO 342/1, ZA 428/18-1, ZA 428/10-1, ME 5413/1-1). The BigpAK-2 trial was supported by an independent research grant from BioMérieux.

Data availability statement

Data are available upon reasonable request by contacting the coordinating author (zarbock@uni-muenster.de).

Declarations

Conflicts of interest

TvG, EB, MM, DT, IG, MS have no potential conflict of interest to declare. AZ has received consulting fees from Astute-Biomerieux, Baxter, Bayer, Novartis, Guard Therapeutics, AM Pharma, Paion, Viatris, Dropshot, Fresenius, research funding from Astute-Biomerieux, Fresenius, Baxter, and speaker fees from Astute-Biomerieux, Fresenius, Baxter. LF received funding and honoraria from Baxter and consulting fees from AstraZeneca, Baxter, and SphingoTec. HG received consultancy fees for trilinear bio ventures, Novartis, and Talphera, speaker fees from bioMerieux and research grants from Baxter and biomMrieux. JAK holds royalties or licenses from CytoSorbents, and J3RM holds stock options of and is contracted by Spectral Medical and received honoraria from AstraZeneca, Bayer, Novartis, bioMéríeux, Mitsubishi Tenabe, and Chugai Pharma. AZ and JAK are named inventors on multiple patents (assigned to the Universities of Münster and Pittsburgh), involving TIMP-2 and IGFBP7, unrelated to the use of these biomarkers for AKI detection or risk stratification. Urinary TIMP-2 and IGFBP7 are measured in the commercially available NephroCheck assay marketed by bioMérieux; author relationships with Astute Medical/bioMérieux are disclosed above.