The search strategies, selection criteria and evidence report were formulated according to the recommendations of Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) (Additional file 1: Table S1), and the study was registered on the PROSPERO database (ID: CRD42022367060).

The Cochrane Library, PubMed, SinoMed, Google Scholar, Embase and Web of Science databases were comprehensively searched for relevant studies published until March 1, 2023. The titles and abstracts were preliminarily assessed, after which the full texts of the relevant studies were acquired. The reference lists of the selected articles were also scrutinized for any extra potential studies. The following search string was constructed by combining the terms related to patients and interventions: [(Urinary calculi OR Kidney stone OR Renal calculi OR Urolithiasis) AND (Contrast-enhanced ultrasound OR Ultrasound OR Ultrasonic contrast) AND (Percutaneous nephrolithotomy OR Percutaneous)].

No language-based limitations were imposed.

To determine the studies to be included, the abovementioned search strategy was constructed in accordance with the PICOS framework. [P (patients): patients with non-dilated collecting system kidney stones who required PCNL surgery; I (intervention): CEUS-guide PCNL performed; C (comparator): conventional US-guided PCNL used as a comparison; O (outcomes): perioperative outcomes and complications; S (study type): prospective and retrospective case–control studies as well as randomized controlled trials (RCTs)]. The exclusion criteria were as follows: (1) no relevant data available for meta‐analysis; (2) non-comparative studies and (3) conference abstracts, case reports, letters and any other unpublished articles.

Using Endnote X9 (London, UK), two reviewers (WL and KP) distinguished the conclusive literature by eliminating duplicates, perusing title-level abstracts and performing a full-text audit based on the set inclusion and exclusion criteria for all the incorporated studies. A senior researcher (YS) was consulted in case of disparities. The data extraction process was subsequently implemented, trailed by the ordering of the study data using pre-set Excel tables.

Two reviewers (WL and YL) independently extracted the following details: first author name, year of publication, type of study, country, number of patients, age, body mass index (BMI), gender, PCNL position, lithotripsy technique, CEUS guided PCNL technique, contrast agent used, stone characteristics and follow-up. Furthermore, the following outcomes were retrieved: perioperative outcomes including puncture time, the stone-free rate (SFR), decrease in hemoglobin levels, hospital stay duration, operation time, single-needle puncture success rate, and postoperative complications.

The Cochrane Handbook for Systematic Reviews of Interventions was used to assess the risk of bias [12]. The assessment comprised the analysis of selection bias, attrition bias, blinding and sample size. For all the studies included in this meta-analysis, the level of evidence (LE) was independently estimated according to the criteria provided by the Oxford Centre for Evidence-Based Medicine [13]. The evaluation was completed by two separate reviewers (KP and YS), and any discrepancies were resolved through negotiation.

Review Manager 5.3 (Cochrane Collaboration, Oxford, UK) was used for the meta-analysis [14]. Odds ratios (ORs) with 95% confidence intervals (CIs) was used for dichotomous variables. Standard mean differences (SMDs) and Weight mean differences (WMDs) with 95% confidence intervals (CIs) were used to present continuous variables. McGrath et al.’s formula was utilized to translate the median and range of data for the means and standard deviations [15]. If there was significant heterogeneity at I2 > 50%, random-effect models were selected according to the Cochrane review principles. In all other situations, fixed-effect models were applied. P < 0.05 was considered statistically significant.

Potential sources of heterogeneity, if significant, were explored using sensitivity analysis, performed using leave-one-out methods. However, this condition was not applied when the comparison involved fewer than three studies [16].

Because of the insufficient test power when fewer than 10 studies are included, linear regression-based publication bias detection methods, such as the Begg and Egger tests, may be inaccurate [17]. Thus, we used the visual symmetry of the funnel plot to roughly estimate the publication bias.