Protocol for an individual patient data meta-analysis for the assessment of risk factors affecting the development and outcome of functional and structural acute kidney injury after cardiac surgery: the METARIS CS-AKI study

Abstract

Introduction

In addition to serum creatinine (SCr), a kidney functional biomarker, structural damage biomarkers such as neutrophil gelatinase-associated lipocalin (NGAL) may improve identification of patients at risk of adverse outcomes after cardiac surgery. Previous meta-analyses have relied on study-level data, functional acute kidney injury (AKI) definitions and heterogeneous NGAL cut-off concentrations, limiting comparability and clinical interpretation.

Methods and analysis

We will conduct an individual patient data (IPD) meta-analysis of prospective studies investigating NGAL in adult patients undergoing cardiac surgery. Authors of eligible studies will be contacted to provide IPD according to a prespecified data request form. Primary outcomes will be the AKI incidence, initiation of renal replacement therapy and in-hospital mortality after surgery. SCr and NGAL will be analysed both as continuous predictors and using study-specific as well as harmonised IPD-derived cut-off values to characterise AKI phenotypes and predictors of adverse outcomes enabling the quantification of risk across the full spectrum of NGAL and SCr concentrations.

Exploratory analyses using mixed-effects logistic regression models will investigate patient-level characteristics that may modify clinical risk for the development of AKI and related adverse outcomes. The study results will be reported in accordance with the Preferred Reporting Items for a Systematic Review and Meta-Analysis of Individual Participant Data. The findings may improve biomarker-guided risk stratification after cardiac surgery.

Ethics and dissemination

Ethics approval was obtained for all included studies by the respective ethics committees or institutional review boards. This meta-analysis is exempt from additional ethical approval because it uses anonymised data from previously published studies with informed consent already obtained. Findings will be disseminated through national and international conferences and may be published in separate peer-reviewed articles according to the scope of the analyses.

PROSPERO registration number

CRD420261336943.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This study will use an individual patient data (IPD) meta-analysis design, enabling standardised analyses across prospective cardiac surgery cohorts.

• The joint evaluation of a kidney damage biomarker (neutrophil gelatinase-associated lipocalin, NGAL) and kidney functional marker (serum creatinine) may allow more precise characterisation of acute kidney injury (AKI) phenotypes and adverse outcomes.

• Modelling NGAL and serum creatinine as continuous variables rather than dichotomised measures will preserve computational information and improve risk stratification.

• Mixed-effects regression analyses will explore patient-level modifiers of AKI risk and clinically relevant subgroup differences.

• Potential limitations include heterogeneity across included studies, variability in biomarker measurement protocols and possible incomplete availability of individual patient data from eligible cohorts.

Introduction

Acute kidney injury (AKI) is common in hospitalised patients undergoing cardiac surgery and is associated with serious complications and adverse outcomes.¹ Typically, AKI is diagnosed according to consensus criteria that use increases in serum creatinine (SCr) and decreases in urine output.² However, the utility of these criteria is limited by the fact that relative to the onset of AKI, the changes in SCr and urine output are delayed. Therefore, their diagnostic performance is considered suboptimal, with low sensitivity and specificity.³

To address these shortcomings, the Acute Disease Quality Initiative recommended the use of novel kidney biomarkers, in combination with clinical information, to improve diagnostic accuracy for AKI, to identify the different pathophysiological processes involved, to discriminate AKI aetiology and to assess AKI severity.⁴ Such biomarkers measured in urine or plasma may indicate kidney tissue stress or damage before the onset of AKI as identified by SCr.⁵ Accordingly, prospective, observational trials have examined whether including these biomarkers in addition to SCr improves early kidney risk assessment over traditional methods.⁶

Of those, neutrophil gelatinase-associated lipocalin (NGAL) reflects early structural tubular damage, has prognostic value for patients’ kidney outcomes and is one of the most intensely investigated kidney injury biomarkers.^{7 8} Reference ranges derived from healthy individuals are readily available,⁹ and NGAL concentrations measured immediately at patient admission to the emergency department or the intensive care unit, or sampled after cardiac surgery differed between those patients who went on to develop AKI or required commencement of renal replacement therapy (RRT) and those who did not.¹⁰ However, NGAL might not reflect changes in glomerular filtration rate.

Combinations of structural and functional AKI may define subgroups of patients with specific underlying mechanisms (pathophysiological phenotypes of AKI) that increase the risk of adverse outcomes.^{5 11} Interestingly, studies have found that the risk of adverse outcomes is increased in patients with elevated NGAL levels, regardless of whether renal function acutely declined or not.^12–16

In the absence of agreed upon NGAL thresholds, previous studies calculated and reported study-specific NGAL cut-off concentrations. Urine and/or plasma NGAL concentrations above the cut-off were considered ‘NGAL positive’, and those below the cut-off were labelled as ‘NGAL negative’. Such thresholds were based on SCr, using the biomarker concentration with maximum Youden’s index (ie, the cut-off which maximised the combined sensitivity and specificity) for prediction of SCr-based AKI stage 1 or worse.^{7 14 17} Similar methods were adopted in other studies,^{12 13 15 18} and the resulting cut-offs were used to allocate patients into AKI phenotypes according to a 2×2 cell matrix system (figure 1).

Figure 1. Left: Attributed clinical AKI phenotypes plotted in a 2×2 contingency table or ‘matrix’. Functional impairment is attributed to increased serum creatinine concentrations or reduced urine output and defined by positive RIFLE, AKIN or KDIGO criteria. Structural damage is attributed to increased NGAL concentrations above a candidate threshold concentration. Right: Corresponding traditional test classification matrix: NGAL(–)/SCr/UO(–), TN, no damage apparent, no functional kidney impairment apparent; NGAL(+)/SCr/UO(–), FP, subfunctional or subclinical AKI (AKI stage 1 S); NGAL(–)/SCr/UO(+), FN, haemodynamic AKI, volume depletion, diminished kidney functional reserve (AKI stage 1–3A); NGAL(+)/SCr/UO(+), TP, AKI with functional and structural impairment (AKI stage 1–3B). Classification corresponding to traditional SCr/UO-based AKI, yes/no; classification corresponding to NGAL—status positive/negative. AKI stages according to Ostermann et al.⁴ AKI, acute kidney injury; AKIN, Acute Kidney Injury Network classification; FN, false negative; FP, false positive; KDIGO, Kidney Disease: Improving Global Outcomes classification; NGAL, neutrophil gelatinase-associated lipocalin; RIFLE, Risk, Injury, Failure, Loss of kidney function, and End-stage kidney disease classification; SCr/UO, serum creatinine/urine output; TN, true negative; TP, true positive.

We recently aggregated original study data in order to provide applicable cut-off concentrations using a meta-analysis approach.⁷ Adjusting for confounding factors such as the clinical setting and measurement specimen, we found considerable variability among studies’ individual cut-off concentrations, likely attributable to differences in the clinical settings and the original study protocols (figure 2). This variability in cut-off selection, in turn, may lead to differences in patient classification to clinical AKI phenotypes and consecutively, their attributed risk of adverse events depending on the statistical methodology used for cut-off derivation and AKI definition.^{19 20}

Figure 2. Variability of plasma NGAL (a) and urine NGAL cut-off concentrations (b) for the outcome measures AKI (of all stages), severe AKI (RIFLE stages Injury and Failure) and acute RRT initiation derived from Albert et al.⁷ Each bubble represents an individual study’s cut-off concentration within the respective outcome categories, with the bubble size shown proportional to the study’s sample size. AKI, acute kidney injury; KDIGO, Kidney Disease: Improving Global Outcomes classification; NGAL, neutrophil gelatinase-associated lipocalin; RIFLE, Risk, Injury, Failure, Loss of kidney function, and End-stage kidney disease classification; RRT, renal replacement therapy.

Moreover, hypothesis-generating post hoc analyses of our data⁷ suggest a non-linear relationship between NGAL concentrations and specificity for AKI. Rather, there is a near hyperbolic pattern as specificity increases, which potentially limits risk discrimination at lower concentrations and results in a potential NGAL–‘blind range’ that might justify repeat testing or use of delta values (figure 3). We speculate that this pattern reflects typical non-linear biological dose–response behaviour, whereby mild tubular injury leads to limited NGAL release, while more severe injury results in the detection of disproportionately higher biomarker concentrations and steep gains in specificity. This is driven by tubular pathophysiology rather than assay limitations. Clinically, this observation may be analogous to the well-known relationship between SCr and estimated glomerular filtration rate (eGFR) for chronic kidney disease (CKD), in which early declines in kidney function may remain undetected within a similar diagnostic blind range.

Figure 3. Plasma NGAL (left panel) and urine NGAL concentrations (right panel) up to 1200 ng/mL were plotted as a function of diagnostic specificity for the outcome measures AKI, severe AKI and RRT. Specificity values were estimated by non-linear interpolation between predefined anchor concentration and specificity pairs for each outcome derived from Albert et al.⁷ Trends were visualised using GAM, able to fit non-linear relationships through cubic regression splines to capture patterns in the data. Shaded areas represent the approximate 95% CI of the fitted curve at each point calculated from the standard errors of the GAM fit, providing a visual estimate of uncertainty around the predicted NGAL concentration for a given specificity. In the scenarios modelled above, plasma and urine NGAL concentration thresholds only increase modestly as specificity is increased up to approximately 70%, indicating that low or borderline concentrations may provide only limited diagnostic confidence and may therefore be less reliable for both confirming and excluding the development of an outcome of interest. Beyond the 70% specificity mark, the biomarker threshold concentrations rise steeply for higher specificities, meaning that higher biomarker concentrations offer substantially stronger rule-in evidence for adverse outcomes. Consequently, while elevated NGAL concentrations may strongly support an early diagnosis of tubular injury, lower concentrations may warrant continued monitoring or additional testing.⁴⁷ Abbreviations: AKI, acute kidney injury; generalised additive models, GAM; NGAL, neutrophil gelatinase-associated lipocalin; RRT, renal replacement therapy.

Nevertheless, pooled analyses of prospective cohort studies have shown that elevated NGAL concentrations can identify patients at risk of adverse events, even in the absence of SCr increases.⁸ Dividing NGAL concentrations into tertiles and classifying according to the presence or absence of AKI yields six rather than four groups and may further delineate the relative risk for adverse events.²¹

These findings challenge the suitability of dichotomised or trichotomised, cut-off-based classifications and point towards the benefit of modelling incremental biomarker-related risk for adverse events using a continuous, progressive measure.

Additionally, none of the aforementioned analyses adjusted for individual patient characteristics or risk factors, as detailed patient-level information was either not available or not considered in the prior analyses.⁸ Therefore, the influence of patients’ clinical presentation, comorbidities and perioperative risk factors on the development of functional and structural kidney injury remains incompletely understood.^{6 14 22} Previous meta-analyses have neither addressed these issues nor assessed risk associated with elevated NGAL independently of SCr (eg, mortality, RRT initiation).⁸

Yet, several clinical factors are thought to influence the development of cardiac surgery-associated AKI, including gender, CKD, congestive heart failure, reduced left ventricular ejection fraction, diabetes mellitus, chronic obstructive pulmonary disease, history of previous cardiac surgery, type of surgery and cardiopulmonary bypass duration.^23–25

Methodological rationale

Addressing the shortcomings of previous analytical approaches, this study protocol introduces several methodological innovations. We propose collecting individual patient data (IPD) from primary studies in order to conduct meta-regression and exploratory analyses. This will allow the characterisation of clinical factors’ influences on the development of adverse events.²⁶

This approach extends beyond the current established system based on a confusion matrix of NGAL to predict AKI (figure 1) as predictors will be modelled as continuous variables rather than dichotomised, thereby preserving information, maintaining resolution and enabling more refined risk stratification. In addition, for the first time in acute renal risk prediction, both a kidney damage marker (NGAL) and a kidney function marker (SCr) will be jointly considered, capturing two complementary dimensions of acute renal impairment (figure 4).

Figure 4. Conceptual illustration of a risk model combining the continuous parameters NGAL and SCr in combination (in ng/mL and µmol/L, respectively). The figure is intended to demonstrate how study results may be visualised, with the greyscale shading reflecting increasing risk (eg, ORs or predicted probabilities) of adverse renal outcomes such as acute renal replacement therapy or mortality. After modelling of continuous variables is performed, categorisation into groups of incrementally increasing risk useful for clinical decision-making is performed. Tiles of the heat map represent individual groups of elevated risk compared with the reference (1.0) without NGAL or SCr increase, respectively. NGAL, neutrophil gelatinase-associated lipocalin; SCr, serum creatinine.

Study objectives and hypotheses

The METARIS CS-AKI study (individual patient data meta-analysis for the assessment of risk factors affecting the development and outcome of functional and structural AKI after cardiac surgery) will summarise published and unpublished evidence on risk factors associated with functional and structural AKI phenotypes using IPD from prospective clinical studies.

Our a priori hypothesis is that increased postoperative concentrations of SCr and NGAL, alone or in combination, are associated with increased risks of mortality or need for acute RRT initiation. We further hypothesise that:

1. Study-specific cut-off concentrations for NGAL and those from the literature may differ from those derived in a pooled IPD analyses.

2. NGAL and SCr concentrations measured after cardiac surgery will be incrementally predictive across their continuous range for adverse outcomes.

3. Patient characteristics may identify interpretable subgroups with differential risk.

Study aims

1. Cut-off-based analysis:

• Identify NGAL and SCr cut-off concentrations (separately and combined) based on IPD and compare these with previously reported study-specific cut-offs.

• Use these cut-off concentrationss to estimate relative risks, ORs or probability for major adverse outcomes, including:

  • Necessity of acute RRT initiation.

  • In-hospital mortality and mortality during 30 and 90 days follow-up.

  • Serum creatinine/urine output (SCr/UO)-based AKI.

  • Severe stages of AKI (defined as Risk, Injury, Failure, Loss of kidney function, and End-stage kidney disease classification (RIFLE) I+F or Kidney Disease Improving Global Outcomes (KDIGO) stage 2+3).

• 2Continuous predictor analysis:

• Model NGAL and SCr as continuous variables and quantify the incremental change in odds or the probability of adverse outcomes as a function of biomarker concentrations.

• Compare continuous performance with cut-off concentration-based classifications to evaluate information gain.

• Derive relation of sensitivity and specificity versus the NGAL concentrations for the predefined outcome measures.

• 3Exploratory stratified analyses:

• Assess the role of perioperative patient characteristics (eg, CKD, diabetes, arterial hypertension, reduced ejection fraction, type of surgery etc.) as modifiers of the NGAL–outcome relationship.

• Identify clinically interpretable subcategories.

• Compare the performance of additional kidney biomarkers, when available, alongside NGAL and SCr.

Methods and analyses

Analytical framework and contingency strategy

The primary analytical approach of this study will be an IPD meta-analysis. This approach will be pursued if IPD can be obtained from a sufficient proportion of eligible studies to ensure representativeness and validity of pooled estimates. For this purpose, feasibility will be defined a priori as the availability of IPD from at least 65% of eligible studies identified through the systematic literature search. The threshold was chosen in reference to Nevitt et al representing typical historical data retrieval rates across IPD meta-analyses.²⁷

If these criteria are not met, analyses will be conducted using the available datasets as a multicentre pooled cohort of individual participant data. In this case, results will be interpreted as exploratory and hypothesis-generating rather than confirmatory. The analytical methods applied will remain consistent; however, the limitations related to potential selection bias and reduced generalisability will be explicitly acknowledged. The feasibility of each analysis will be assessed at the individual analysis level based on data availability.

Since the outlined hypotheses and corresponding analyses are dependent on IPD and cannot be accomplished with data extracted from the literature, we will not perform any analyses of aggregate literature data regarding the appointed study aims.

Reporting of findings of an IPD meta-analysis will follow the ‘Preferred Reporting Items for a Systematic Review and Meta-Analysis of Individual Participant Data’.²⁸ The protocol was registered with the International Database of Prospectively Registered Systematic Reviews (PROSPERO, registration number CRD420261336943).²⁹ Reporting of this protocol, the ‘Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols’ were followed.³⁰ The item checklist is available as online supplemental material 1.

Identifying eligible studies

Two investigators will independently perform a systematic literature database search in MEDLINE (through PubMed interface, http://www.ncbi.nlm.nih.gov/pubmed), Web of Science, Cochrane Library and Scopus databases to identify potentially relevant articles ever published through 31 October 2026 using the linked search terms and Medical Subject Headings (by modifiers ‘OR’, ‘AND’) and their combinations: ‘acute kidney injury’, ‘acute renal failure’, ‘acute kidney impairment’, ‘AKI’, ‘ARF’, ‘subclinical AKI’, ‘hemodynamic AKI’, ‘NGAL’, ‘neutrophil gelatinase-associated lipocalin’, ‘siderocalin’, ‘renal replacement therapy’, ‘RRT’, ‘acute’, ‘dialysis’, ‘mortality’, ‘adverse’, ‘outcome’. A sample search strategy is provided in online supplemental material 2.

Additionally, reference lists of relevant reviews, articles and congress abstracts meeting inclusion criteria and previously published meta-analyses will be reviewed to identify studies that may have been missed in the database search (backward citation searching). Articles with potential relevance will be retrieved for full-text review. The search strategy will be applied without language restriction.

Studies selected will be restricted to diagnostic test studies, observational studies or cohort studies investigating AKI in adult humans, and RRT or patient mortality in the setting of cardiac surgery using either urinary or plasma NGAL for diagnostic or risk assessment purpose. Study data identified through published conference proceedings will be considered. In contrast, data that are not publicly retrievable, accessible and verifiable through recognised registries will not be considered for inclusion.³¹ The NGAL measurements must be performed using clinical laboratory platforms, point-of-care devices or commercially available ELISA.

Reporting of individual study characteristics

Data reporting will include AKI diagnostic criteria, sample material, characteristics of utilised laboratory detection method such as ELISA, point-of-care device or clinical laboratory platforms and the variables: sample size, number of patients with and without AKI, AKI stages, RRT as well as patient in-hospital mortality, where available with a follow-up of 30 and 90 days, cut-off concentrations with corresponding sensitivity and specificity, the area under the curve (AUC) with corresponding 95% CI for NGAL, SCr and other biomarkers wherever available. In addition, we will describe patient-level-based characteristics of included studies.

Identification of index parameters

Data extracted from individual patient level to perform risk assessments will include index parameters relevant to perioperative cardiac surgery risk assessment (table 1).^{7 8 32–35} For identification of index data, we performed a literature search for eligible cardiac surgery studies assessing NGAL for renal risk prediction. To obtain an overview of potentially available patient data, we extracted patient and study characteristics from eligible studies into a summary table and harmonised the extracted data.

Table 1. Requested data variables.

Patient baseline characteristics	Biomarkers
Age (years)	Neutrophil gelatinase-associated lipocalin (NGAL)
Any lipid disorder, yes=1, no=0	Preoperative plasma NGAL (ng/mL)
Arterial hypertension, yes=1, no=0	Postoperative plasma NGAL closest to ICU admission (ng/mL)
Diabetes mellitus, yes=1, no=0	Plasma NGAL, post-OP day 1 (ng/mL)
Chronic heart failure, yes=1, no=0	Plasma NGAL, post-OP day 2 (ng/mL)
Chronic kidney disease, yes=1, no=0
Any preoperative proteinuria	Preoperative urine NGAL (ng/mL)
Chronic obstructive pulmonary disease, yes=1, no=0	Postoperative urine NGAL closest to ICU admission (ng/mL)
Gender, male=1, female=0	Urine NGAL, post-OP day 1 (ng/mL)
NYHA class (numerical, 1–4)	Urine NGAL, post-OP day 2 (ng/mL)
Other pulmonary disease (ie, asthma), yes=1, no=0
Preoperative left ventricular ejection fraction, EF (%)	(TIMP-2)×(IGFBP7)
Coronary artery disease, yes=1, no=0	Preoperative(TIMP-2)×(IGFBP7)
Peripheral artery disease (PAD), yes=1, no=0	Postoperative(TIMP-2)×(IGFBP7)
Extracardiac arteriopathy, yes=1, no=0
Prior or recent myocardial infarction, yes=1, no=0	Cystatin C (Cys-C)
Pulmonary hypertension, yes=1, no=0	Preoperative cystatin C (mg/L)
Weight (kg), height (cm), BMI (kg/m2)	Postoperative cystatin C closest to ICU admission (mg/L)
(preoperative) Cardiogenic shock, yes=1, no=0	Cystatin C, post-OP day 1 (mg/L)
	Cystatin C, post-OP day 2 (mg/L)
Surgery-associated characteristics
Active endocarditis, yes=1, no=0	C-reactive protein (CRP) (mg/L, mg/dL)
Any valve surgery, only, yes=1, no=0	Preoperative CRP (mg/L, mg/dL)
Aortic valve surgery, yes=1, no=0	Postoperative CRP closest to ICU admission (mg/L, mg/dL)
Mitral valve surgery, yes=1, no=0	CRP, post-OP day 1 (mg/L, mg/dL)
CABG only, yes=1, no=0	CRP, post-OP day 2 (mg/L, mg/dL)
Cardiopulmonary bypass time (CPB time) (in minutes)	CRP, post-OP day 3 (mg/L, mg/dL)
Concomitant procedure (valve and CABG), yes=1, no=0
Emergency surgery, yes=1, no=0	NT-pro BNP or BNP (pg/mL, pmol/L)
IABP, yes=1, no=0	Preoperative NT-pro BNP or BNP (pg/mL, pmol/L)
Previous cardiac surgery, yes=1, no=0	Postoperative NT-pro BNP or BNP closest to ICU admission (pg/mL/pmol/L)
Re-operation, yes=1, no=0	Postoperative NT-pro BNP, post-OP day 1 (pg/mL/pmol/L)
Intraoperative hypotension, yes=1, no=0
Red blood cell supplementation, yes=1, no=0	Kidney injury molecule (KIM-1)
	Preoperative KIM-1 (ng/mL)
Outcome measures	Postoperative KIM-1 closest to ICU admission (ng/mL)
Any postoperative acute dialysis initiation (RRT), yes=1, no=0	KIM-1, post-OP day 1 (ng/mL)
Commencement of RRT on postoperative day X	KIM-1, post-OP day 2 (ng/mL)
In-hospital mortality, yes=1, no=0
Mortality including up to 30 days, yes=1, no=0	Perioperative proenkephalin A (define timing)
Death on postoperative day X	Preoperative proenkephalin A
Acute kidney injury (RIFLE, AKIN, KDIGO—please specify)	Postoperative proenkephalin A
Any stage acute kidney injury, yes=1, no=0 (RIFLE, AKIN, KDIGO)
Stage 1/RIFLE Risk	Perioperative kidney function
Stage 2/RIFLE Injury	Serum creatinine (SCr)
Stage 3/RIFLE Failure	Any long-term baseline or stable CKD serum creatinine if available (µmol/L or mg/dL)
AKI occurred on postoperative day X	Baseline eGFR (mL/min/1.732)
Any AKI defining urine output criterion (please outline)	Preoperative serum creatinine (baseline serum creatinine) (µmol/L or mg/dL)
End-stage kidney disease (subsequent to operation), yes=1, no=0	Postoperative serum creatinine closest to ICU admission (µmol/L or mg/dL)
Fluid overload (please specify your definition), yes=1, no=0	Serum creatinine, post-OP day 1 (µmol/L or mg/dL)
	Serum creatinine, post-OP day 2 (µmol/L or mg/dL)
Preoperative medication	Serum creatinine, post-OP day 3 (µmol/L or mg/dL)
ACEs or ARBs, yes=1, no=0	Serum creatinine, post-OP day 4 (µmol/L or mg/dL)
Lipid-lowering drugs, yes=1, no=0	Serum creatinine, post-OP day 5 (µmol/L or mg/dL)
Any diuretics, yes=1, no=0	Serum creatinine, post-OP day 6 (µmol/L or mg/dL)
Beta blockers, yes=1, no=0	Serum creatinine, post-OP day 7 (µmol/L or mg/dL)
Any insulin, yes=1, no=0	Maximum serum creatinine, post-OP in hospital (µmol/L or mg/dL)
Immunosuppression, yes=1, no=0	Serum creatinine closest to hospital discharge (µmol/L or mg/dL)

ACE, angiotensin-converting enzyme inhibitor; AKI, acute kidney injury; AKIN, Acute Kidney Injury Network classification; ARB, angiotensin receptor blocker; BNP, B-type natriuretic peptide; CABG, coronary artery bypass grafting; CKD, chronic kidney disease; CRB time, cardiopulmonary bypass time; CRP, C-reactive protein; Cys-C, cystatin C; EF, left ventricular ejection fraction; eGFR, estimated glomerular filtration rate; IABP, intra-aortic balloon pump; ICU, intensive care unit; KDIGO, Kidney Disease: Improving Global Outcomes classification; KIM-1, kidney injury molecule-1; NGAL, neutrophil gelatinase-associated lipocalin; NYHA, New York Heart Association classification; PAD, peripheral artery disease; post-OP, post-operation; RIFLE, Risk, Injury, Failure, Loss of kidney function, and End-stage kidney disease classification; RRT, renal replacement therapy; SCr, serum creatinine; [TIMP-2]×[IGFBP7], tissue inhibitor of metalloproteinases-2 × insulin-like growth factor-binding protein 7.

To estimate parameter-overlap across studies, we adopted the conceptual framework of the corrected covered area (CCA) described by Pieper et al,³⁶ originally developed to quantify overlap in umbrella reviews. Using a parameter-by-publication matrix, parameter use was recorded accordingly. To our knowledge, this conceptual application of the CCA framework to parameter overlap has not been previously described. As a central presentation of the results, an evidence map was derived to illustrate the extracted parameters’ overlap within the identified studies. Results of this analysis will be reported in a separate manuscript.

In addition to the extracted parameters, potentially missing characteristics relevant to calculate common risk severity scores/classification scores were selected and included in a final data request sheet matching the clinical setting of cardiac surgery.

Collection of individual patient data

Invitational requests to participate in a prospective IPD meta-analysis will be sent by e-mail to corresponding authors of identified eligible studies. A cover letter detailing the background information and objectives of the IPD meta-analysis will be accompanied by the data collection file provided as a Microsoft Excel sheet (Microsoft Corp., Redmond, Washington, USA). In detail, the requested patient and study characteristics are provided in table 1.

Definition of acute kidney injury

Authors are requested to provide sequential SCr data to enable classification of AKI within 7 days of onset using the KDIGO² and R (risk), I (injury) or F (failure) RIFLE classification.³⁷ If sequential SCr concentrations are not provided, AKI will be defined and classified according to the study-specific definition of AKI using KDIGO, RIFLE or Acute Kidney Injury Network (AKIN)³⁸ classifications.

Definition of acute kidney injury phenotypes

For the cut-off-based analyses, patients’ distribution within the AKI matrix system will be based upon biomarker positivity. We will apply the term ‘NGAL–positive’ or ‘NGAL–negative’ to indicate the absence or presence of kidney structural damage and the term ‘SCr–positive’ or ‘SCr–negative’ to indicate functional impairment. Phenotypes will thus be defined along the structural (NGAL) and functional (SCr) axes (figure 1).

Outcome measures

Outcome measures in the IPD meta-analysis will be the ones listed above, that is, renal replacement therapy initiation, in-hospital mortality, mortality during 30 and 90 days’ follow-up, SCr/UO-based AKI and severe AKI.

Data management and data validation

Data will be sent to the coordinating study investigator as encrypted email attachments. Forwarding of database files will be accepted (eg, SPSS, Stata) but using the individualised Excel data request sheets is preferred. Data are specifically requested without information that may enable back-tracing to an individual patient. Any information directly identifying patients (eg, study numbers, record numbers, names, addresses or similar identifiers), if present, will be deleted by the principal investigation team immediately upon receipt. Anonymised individual patient data will then be inserted into a single database file. To ensure quality of the data received, two readers will independently check the datasets for data entry mistakes, non-plausible data and other data checks such as consistency with published data from the original study references. In case of implausibility, data providers (authors of original studies) will be contacted for clarification.

Harmonisation of data, such as conversion of parameter measurement units, will be performed electronically and verified by two independent readers from the central data management team.

Statistical analysis

All analyses will be performed using mixed logistic regression models with the study identifier (ID) as a random effect. The outcome measure of interest will serve as the response variable. Unless otherwise specified, predictors will include NGAL and/or SCr concentrations, with the measurement selected to a closest harmonised post-surgery timepoint when multiple measurements are available. The robustness of meta-analytic estimates will be assessed through appropriate sensitivity analyses and evaluation of between-study heterogeneity. The predictive performance of regression models will be assessed using repeated cross-validation or, where available, an independent external validation cohort. All results will be reported with appropriate measures of uncertainty, such as CIs, to reflect the precision and reliability of the findings. Missing data will be handled using multiple imputation or complete case analysis, depending on the missingness structure. Logarithmic transformations of continuous variables may be applied when appropriate; additionally, restricted cubic splines may be used to model flexibly non-linear relationships without imposing linearity assumptions on transformed concentrations.

1. Cut-off-based analyses.

The summary area under the receiver operator characteristic curves (AUC-ROC) for the given outcome measures will be calculated. Sensitivity and specificity curves will be modelled against NGAL thresholds. Cut-off concentrations will be calculated and meta-analysed based on the Youden index derived from the AUC-ROC, analogous to established approaches.²⁰ Separate models will be run using NGAL alone, SCr alone and both NGAL and SCr together as predictors. Derived cut-off concentrations will be compared with those results from published meta-analysis,⁷ study-specific thresholds, optimised sensitivity-prioritised/specificity-prioritised cut-off concentrations and SCr-independent cut-off concentrations.¹⁸

2. Continuous predictor analyses.

NGAL and SCr will be modelled both separately and jointly as continuous predictors. Mixed logistic regression models with study ID as a random effect will be fitted using NGAL alone, SCr alone and NGAL together with SCr. In the joint models, potential non-linear effects and interactions will be considered. Reclassification metrics such as integrated discrimination improvement and net reclassification improvement may be used to assess the improvements in the performance of the reference model following the addition of additional risk variables.³⁹ Predicted probabilities of adverse outcomes will then be evaluated across a grid of NGAL and SCr values and visualised as a two-dimensional risk surface (eg, heatmap). For the single-predictor models, incremental changes in odds or predicted probability of adverse outcomes will be quantified per unit increase in NGAL or SCr (figure 4). Continuous modelling approaches will be directly compared with cut-off-based classifications in terms of predictive performance.

3. Exploratory stratified analyses

Risk factors will be identified using mixed logistic regression with study ID as a random effect and all potential covariates (eg, CKD, diabetes, arterial hypertension, reduced EF, surgical type etc.) as predictors. NGAL and SCr will be included as additional predictors in extended models. Risk factors will be determined and sorted by relevance using stepwise backward selection or regularised regression methods.

Analyses will be performed separately for urine and plasma NGAL. All statistical analyses outlined in this protocol were conducted using the R environment for statistical computing (R Foundation for Statistical Computing, Vienna, Austria),⁴⁰ and analyses for the forthcoming protocolised study described herein will likewise be performed within this framework.

Patient and public involvement

Patients or the public are not involved in the design or conduct of the research. The results will be presented at national as well as international conferences. The findings arising from the hypotheses and analyses outlined herein may be presented in separate peer-reviewed manuscripts, reflecting the scope and complexity of the research questions addressed.

Discussion

By leveraging individual participant data, this meta-analysis will extend and refine the findings of prior studies.^{7 8 32} Specifically, IPD will enable meta-regression analyses and extensive exploratory analyses using state-of-the-art statistical methods for outcome prediction and modelling.

The METARIS CS-AKI study will derive and introduce a novel risk classification framework for AKI after cardiac surgery. Complementary to the KDIGO-endorsed classification of CKD, which incorporates urinary albumin as a marker of kidney damage and SCr and eGFR as markers of kidney function,⁴¹ the underlying model will, for the first time, jointly integrate incremental increases in NGAL and SCr concentrations, representing complementary dimensions of structural kidney damage and kidney function, respectively. This two-dimensional risk classification will be used to assess the risk of RRT and mortality after cardiac surgery. The resulting two-dimensional risk estimates will be illustrated in a comprehensible heatmap.

The integration of patient characteristics and clinical risk factors might help elucidate specific constellations of structural and functional kidney impairment and thereby characterise distinct AKI phenotypes, such as those associated with systemic inflammatory conditions,⁴² or those potentially associated with haemolysis and iron-mediated tubular injury.⁴³

Incorporating these risk factors into clinical prediction models could subsequently improve perioperative risk stratification and identification of patients at increased risk of adverse events after cardiac surgery.

Moreover, using continuous variables on an individual patient level enables more nuanced and differentiated statistical modelling in both exploratory and regression analyses; therefore, providing greater analytical precision than dichotomised approaches such as those relying on confusion matrix–based systems for the determination of clinical AKI phenotypes.⁴⁴

Limitations and risk of bias

Data inclusion may be limited by selection bias and data availability bias if only a subset of contacted authors is able to contribute individual study data or provide sufficient information for meta-regression analysis, particularly given anticipated barriers related to dataset age, investigator mobility, institutional restrictions and data availability.⁴⁵

Analyses may also be challenged by the need for data harmonisation due to heterogeneity in original study protocols, including potential limitations in the classification of raw data, such as but not limited to AKI classification,⁴⁶ particularly if insufficient data are available for reassessment from individual patient data, or by the absence of urine output criteria in some studies. We will examine funnel plots to investigate association between study size and effect size which could be due to publication or data availability bias.

Furthermore, biological variability in SCr and NGAL, as well as analytical and inter-assay variability, will need to be considered when interpreting the findings and deriving potential clinical implications. To maximise data availability and avoid insufficient sample sizes for meaningful analyses, all assay types will be included, with sensitivity analyses accounting for the test kit used.

Perspective

The authors of this study suggest that refining risk assessment using this methodological framework may contribute to improvements in the clinical diagnostic management of patients undergoing cardiac surgery. Specifically, this approach may support a more nuanced identification of patient subgroups at increased risk of adverse kidney-related events and may, in turn, inform clinicians’ diagnostic and risk-assessment algorithms.

Trial status

This is a study protocol for an ‘Individual patient data meta-analysis for the assessment of risk factors affecting the development and outcome of functional and structural acute kidney injury after cardiac surgery’ (the METARIS CS-AKI study), which has been registered with the international prospective register of systematic reviews (PROSPERO), version of record 1.0 available at https://www.crd.york.ac.uk/PROSPERO/view/CRD420261336943.

Ethics Statement

All original studies were approved by and conducted under the oversight of an appropriate institutional review board or human research and ethics committee. In addition, the responsible Ethics Committee of Otto-von-Guericke-University Magdeburg was consulted and raised no concerns regarding the assessment of the outlined analysis of retrospective, anonymized data. Accordingly, the present metaanalysis was deemed exempt from formal ethics approval, as it exclusively uses anonymized data derived from previously approved clinical studies for which informed consent had already been obtained. Data sharing is performed in accordance with the individual protocols and approvals governing each original study. Responsibility for compliance with the respective study protocols, applicable ethical requirements, and local data protection regulations rest with the investigators of the original studies. Following transfer, all received data are handled in accordance with applicable national data protection regulations, including the General Data Protection Regulation (GDPR) for European centers.

Amendments

Any important modifications to the study protocol or methodological approach made during the peer-review process or conduct of the study will be clearly documented, justified, and transparently reported alongside the final publication to ensure methodological integrity.

The funder did not have a role in study design, collection, analysis or reporting of data, preparation of the manuscript, or decision to submit the article for publication.