Objectives

This systematic review and meta-analysis aims to determine whether ICG fluorescence imaging-guided intraoperative bile leak detection decreases clinically relevant postoperative bile leakage after hepatectomy, defined as ISGLS grade ≥B, compared with a conventional assessment without ICG [10]. The secondary aims are to evaluate the effects of ICG imaging on any bile leakage, bile leak-related interventions, major postoperative complications, length of postoperative hospital stay, and mortality and to explore heterogeneity across ICG administration routes.

Protocol and reporting standards

This protocol has been prepared in accordance with the PRISMA-P guidelines [20] (see S1 Checklist). The final systematic review will be reported according to PRISMA 2020 [21]. The protocol has been registered in PROSPERO (CRD420261291065). Any amendments will be documented with rationale and date-stamped updates in the PROSPERO record and described in the final manuscript.

Eligibility criteria

Exclusion criteria

Liver transplantation cohorts; paediatric-only cohorts; studies in which hepatectomy-specific outcomes cannot be extracted; and studies in which postoperative bile leakage mainly results from biliary reconstruction or biliary–enteric anastomosis (such as hepaticojejunostomy/choledochojejunostomy or extrahepatic bile duct resection) will be excluded from the primary quantitative synthesis. Studies focused on pancreatic surgery or hepaticojejunostomy-related leakage will be used only as background literature or narrative context, where relevant, and will not be eligible for quantitative synthesis of bile leakage from the hepatic transection plane or biliary stump after hepatectomy.

Intervention

Intraoperative ICG fluorescence imaging is specifically used to detect bile leaks and guide immediate repair during hepatectomy, focusing on the hepatectomy transection plane and/or biliary stump, regardless of the ICG administration route (intravascular and/or intrabiliary injection) and the fluorescence imaging system (i.e., any near-infrared fluorescence platform).

Operational definition: The report explicitly describes (1) preoperative and/or intraoperative ICG administration via an intravascular route (venous or portal venous) and/or via an intrabiliary route, with near-infrared fluorescence visualization of ICG-containing bile and (2) the use of fluorescence findings to identify suspected bile leakage sites and guide intraoperative management, including targeted suturing, clipping, sealing, or other forms of repair/reinforcement.

Exclusion by intent: Studies will be excluded when ICG fluorescence imaging is used only for non-bile leak purposes—including tumour localization/navigation, segmental delineation, perfusion assessment, or vascular imaging—without an explicit intent to detect bile leakage or guide repair. Studies will not be excluded when bile leak detection is an explicit component of the fluorescence-guided intraoperative strategy.

Primary intent of ICG use: For included studies, the primary intent of ICG administration will be classified as follows: (1) primary bile leak detection, in which ICG is administered specifically to detect bile leakage and guide intraoperative repair; (2) multipurpose or other primary fluorescence-guided intent, in which ICG is administered primarily for another purpose, such as tumour localization, segmental demarcation, anatomical resection, perfusion assessment, or vascular imaging, but bile leak detection is explicitly described as a component of the intraoperative strategy; and (3) unclear intent, when the primary intent cannot be determined from the report. Studies without any explicit intent to detect bile leakage or guide repair will be excluded.

Definitions of ICG administration routes (operational)

We will categorize ICG administration for bile leak detection based on the administration route and specific technique used.

1. A. Intravascular (systemic venous and/or portal venous) administration (IV)
   • ICG is delivered via systemic venous access and/or through the portal venous system, aiming for it to be excreted into bile and visualized intraoperatively using near-infrared fluorescence imaging.
   • Representative techniques include intravenous bolus injection, intravenous infusion, intraportal (portal venous) injection, and single or repeated dosing.
   • The timing and dose of ICG administration will be recorded, including preoperative dosing (i.e., hours to days before surgery) or intraoperative dosing (i.e., at skin incision, at the start of parenchymal transection, during transection, or after transection).
2. B. Intrabiliary administration (IB)
   • ICG is administered directly into the biliary tree to locally label bile, which is then visualized through fluorescence.
   • We subclassify the technique as follows:
     • IB-1 (via the cystic duct/gallbladder/biliary stump cannulation): ICG is injected through a cannulated cystic duct, gallbladder, or biliary stump during surgery.
     • IB-2 (transpapillary endoscopic route): ICG is injected via a transpapillary biliary catheter placed pre- or intraoperatively.
     • IB-3 (preoperatively placed percutaneous transhepatic route): ICG is injected via a percutaneous transhepatic biliary drainage catheter or via a catheter placed during percutaneous transhepatic cholangiography.
     • IB-4 (direct intraoperative puncture/needle injection): ICG is directly injected into the bile duct under direct vision (when clearly described), including intraoperative transhepatic puncture with intrabiliary injection when a percutaneous transhepatic access route is created de novo during surgery.
3. C. Mixed/combined administration (MIXED)
   • Studies involving both intravascular and intrabiliary ICG administration within the same intervention protocol (either sequentially or concurrently) will be classified as MIXED.
4. D. Unclear/insufficiently described administration route (UNCLEAR)
   • If the administration route cannot be identified from the report (i.e., it is insufficiently described), we will classify it as UNCLEAR and attempt to contact the corresponding author. If the issue remains unresolved, the study will be retained in the main synthesis but excluded from administration route-based subgroup analyses.

Comparators

A conventional intraoperative bile leak assessment performed without ICG fluorescence imaging, namely, the surgeon’s judgement under white light, with or without additional leak tests (such as saline or air injection, nonfluorescent dye injection, or the “white test” using fat emulsion), or standard care as defined by each study, will be used as the comparator.

Study design

We will include randomised controlled trials and comparative nonrandomised studies (prospective or retrospective) that evaluate ICG fluorescence imaging-guided intraoperative bile leak detection and compare the results with a control group without ICG fluorescence imaging. The included studies must report outcomes for both groups to enable a comparative estimation of the effects. Single-arm case series and other noncomparative designs will be excluded from the quantitative synthesis; when relevant, they will be summarized narratively to describe the evidence base.

Context

Any hospital setting and country.

Information sources and search strategy

Electronic searches will be conducted in MEDLINE (via PubMed), Embase, Cochrane CENTRAL, and Web of Science Core Collection from inception to the date of the final search. We will also search ClinicalTrials.gov and the WHO International Clinical Trials Registry Platform (ICTRP), screen the reference lists of the included studies and relevant reviews, and conduct forward and backward citation tracking. No language or date restrictions will be applied during the search process, and no study design filter will be used. Articles published in a language other than English will be translated as needed (e.g., by native speakers or using professional services when available), and key data items will be verified by a second reviewer. Full search strategies for all databases and registries are provided in S2 Appendix.

Study records

Outcomes

Risk of bias assessment

Randomised trials will be evaluated using RoB 2 [23]. Comparative nonrandomised studies will be evaluated using ROBINS-I V2 [40, 41], following a prespecified target trial framework. ROBINS-I V2 was selected because the review question concerns the comparative effect of an intraoperative intervention, namely ICG fluorescence imaging-guided bile leak detection, rather than the effect of an exposure. The target trial will be conceptualized as a comparison between intraoperative ICG fluorescence imaging-guided bile leak detection and conventional intraoperative assessment without ICG fluorescence imaging among patients undergoing hepatectomy. The minimum required confounding domains for nonrandomised studies will include surgical period, surgeon or institutional experience, institutional perioperative practice including enhanced recovery pathways and drain management policy, extent of hepatectomy, surgical approach, background liver disease or cirrhosis, cholestasis or biliary obstruction, biliary reconstruction or extrahepatic bile duct resection, surgical indication or tumour type, and history of preoperative chemotherapy. Co-interventions related to intraoperative bile leak assessment, including the white test, saline or air injection, and nonfluorescent dye tests, will also be considered during risk-of-bias assessment. Two reviewers will independently assess the risk of bias; disagreements will be resolved through consensus or by consulting a third reviewer (YF or MU).

Data synthesis

We will conduct a meta-analysis when at least two studies are sufficiently comparable. Randomised controlled trials (RCTs) and comparative nonrandomised studies (NRSI; prospective or retrospective) will be synthesized in separate primary meta-analyses because of systematic differences in confounding structures and the risk of bias [24,25]. When both evidence streams are available for an outcome, we will present pooled estimates for RCTs and NRSI side by side. Any pooled analysis that combines RCTs and NRSI will be considered exploratory and reported only as a secondary analysis with cautious interpretation [24,26].

For dichotomous outcomes, the risk ratio (RR) will be used as the effect measure. A random-effects meta-analysis will be used as the primary approach. Between-study variance (τ²) will be estimated using the restricted maximum likelihood (REML) method [27], and 95% confidence intervals (CIs) for the summary risk ratios will be calculated using the Hartung–Knapp–Sidik–Jonkman method where applicable [28]. Statistical heterogeneity will be assessed using τ² and I² statistics [29,30]. When a sufficient number of studies are available, heterogeneity and effect modification will be explored through subgroup analyses, as outlined in the next section. For outcomes with limited data, we will use prespecified methods for zero-event studies. When a study reports zero events in one arm, we will apply a Haldane-type correction (default: 0.5 added to each cell) to facilitate RR estimation [31]. Studies with zero events in both arms provide no information for relative effect estimates and will be excluded from RR pooling for that outcome but will be summarized descriptively [31]. For nonrandomised studies, adjusted effect estimates will be prioritized when available and synthesized using the generic inverse variance approach. Adjusted odds ratios, risk ratios, and hazard ratios will be extracted on the logarithmic scale with their standard errors when reported or calculable. Because odds ratios, risk ratios, and hazard ratios represent different effect measures, they will not be combined in the same pooled analysis unless a clinically and statistically justified conversion is possible. If multiple adjusted models are reported, we will prioritize the model that best accounts for the prespecified confounding domains. Adjusted and unadjusted estimates will not be pooled together in the same primary synthesis. When adjusted estimates are unavailable, unadjusted event data will be analysed separately as an exploratory or sensitivity analysis, and the limitations of residual confounding will be explicitly considered [32].

For continuous outcomes, including the length of postoperative hospital stay, we will use the mean difference (MD) as the effect measure when outcomes are reported on a common scale. If outcomes are reported on different scales or are not directly comparable, quantitative synthesis will not be performed, and the results will be summarized narratively. For studies reporting medians with interquartile ranges or ranges without means and standard deviations, we will preferentially request additional data from the authors. If such data are unavailable, we will consider estimating means and standard deviations using established methods [33,34] and will assess the robustness of findings in sensitivity analyses by excluding studies that required such conversions.

Where quantitative synthesis is not appropriate because of substantial clinical or methodological heterogeneity, incompatible outcome definitions, or incomplete reporting that precludes estimation of the effect, findings will be summarized narratively and presented in structured tables [35,36]. Following SWiM guidance, studies will be grouped by outcome, study design, ICG administration route, primary intent of ICG administration, and clinically relevant surgical factors when applicable. For each outcome, we will summarize the direction and magnitude of effects, available effect estimates or event risks, confidence intervals where reported, risk of bias, and consistency across studies. For the primary outcome (clinically relevant bile leakage, ISGLS grade ≥B), quantitative pooling will be performed only when at least two studies are sufficiently comparable in design, population, and outcome definition. If pooling is not feasible, conclusions will be based on the consistency, direction, magnitude, and certainty of the available evidence rather than on statistical significance alone. Statistical analyses will be conducted using R, primarily with the metafor and meta packages [37].

Subgroup analyses

Subgroup analyses will focus mainly on the primary outcome and key secondary outcomes for which a sufficient number of studies are available. The prespecified subgroup variables are listed below. The findings from the subgroup analyses will be interpreted cautiously owing to limited statistical power and potential residual confounding.

• Primary intent of ICG administration: primary bile leak detection vs. multipurpose or other primary fluorescence-guided intent with explicit bile leak assessment. Studies with unclear intent will be excluded from this subgroup analysis and summarized descriptively.
• ICG administration route: intravascular vs. intrabiliary injection.

Studies using both routes within the same intervention protocol will be analysed as a separate MIXED subgroup when the data allow; studies with unclear routes will be excluded from the administration route-based subgroup analyses.

• Exploratory subgroups (if reporting permits):
  • – Surgical approach (open vs. minimally invasive). When data permit, minimally invasive surgery will be further categorized as pure laparoscopic, robotic, or robot-assisted/hybrid approaches. These categories will be examined descriptively and, if sufficient comparative data are available, in exploratory subgroup analyses.
  • – Extent of hepatectomy (major vs. minor). For the purpose of subgroup analysis in this review, major hepatectomy will be defined as resection of three or more Couinaud segments, and minor hepatectomy will be defined as resection of fewer than three Couinaud segments.
  • – Resection type (anatomical vs. non-anatomical/parenchymal-sparing resection, when reported). Studies in which the resection type is unclear will be summarized descriptively and excluded from this subgroup analysis.

These exploratory analyses will be interpreted carefully, recognizing the limited power and possible residual confounding [38,39].

Sensitivity analyses

• Excluding studies at serious/critical risk of bias (ROBINS-I V2) [40,41];
• Preferentially pooling adjusted estimates for nonrandomised studies when available [32];
• Excluding studies without feasible ISGLS grade [10] ≥B mapping for the primary outcome;
• Restricting the primary analysis to cohorts in which hepatectomy without extrahepatic bile duct resection constitutes the primary surgical population, excluding studies or subgroups with extrahepatic bile duct resection when separable.

Assessment of reporting bias

If at least 10 studies are available for an outcome, we will construct funnel plots and consider statistical tests for small-study effects while acknowledging limited power [42,43]. If fewer than 10 studies are available, funnel plots and statistical tests for small-study effects will not be performed because of insufficient power and unreliable interpretation. In such cases, reporting bias will be assessed qualitatively by considering study size, direction and magnitude of effects, trial registry records, conference abstracts, unpublished or ongoing studies, and discrepancies between protocols or registrations and published reports, where available. Funnel plots will be included as supplementary figures when applicable.

Certainty assessment

The certainty of evidence for key outcomes will be assessed using GRADE [44].

Guarantor

TH will act as the guarantor of the review.

Ethics and dissemination

This study does not involve human participants and uses data from published studies; therefore, ethical approval is not necessary. The findings will be shared through peer-reviewed publications and conference presentations.

Patient and public involvement

Patients and the public were not involved in the design, conduct, reporting, or dissemination plans of this research.