Literature Search and Study Selection
The results of the literature search are depicted in the PRISMA flow chart (Fig. 1). 698 records were screened for eligibility. For the analysis I of LR for WTLM, 14 studies met inclusion criteria, reporting on a total of 212 patients with WTLM of which 93 underwent LR. For the analysis II that focused on the general outcome of WTLM patients with all kinds of local treatment, eight studies with a total of 254 patients with WTLM were included (among them five studies that were also included in analysis I).
In two studies, specific information on the outcome of patients with surgery for liver metastases was incomplete [4, 5]. Information was requested from the corresponding author in these cases, but the request was not met. As double reporting of patients could not be ruled in the two studies by Liné et al. [7, 27], the corresponding author was contacted. The request for clarification of which patients were reported in both studies was met. Thus, only three of the ten reported patients from the older and less comprehensive study  were included in this review and double reporting was avoided.
Overview of Included Studies, with Critical Appraisal and Risk of Bias Assessment
No randomized-controlled trials, propensity-score or matched-pair analyses were found.
There were three studies in which oncological trial registries were analyzed [3, 4, 7], one multi-center  (MC) and two single-center [5, 28] (SC) retrospective observational studies (ROS). The other eight studies were case reports or series [27, 29,30,31,32,33,34,35]. Five studies were comparative in that they included both, patients with surgical and patients with non-surgical treatment of WTLM [3,4,5,6,7]. However, only three of them adequately reported outcomes for these two groups separately to allow for an explorative comparison of interventions [3, 6, 7]. None of these five studies were conceptualized to specifically investigate the effect of surgery vs. no-surgery on the outcome of patients with WTLM. This implied that the two groups were not comparable regarding sample size, data on extent of disease and other possibly confounding factors. Selection bias was low, and the follow-up was long enough in the six ROS. The only study to report the extent of liver metastasis, type of resections and individual outcome for all patients, was the study by Liné et al. . The case reports increased the risk of publication/reporting bias and tended to present favorable outcomes. Moreover, most of the case series or reports had a high selection bias. On the other hand, individual patient data were detailed in most case reports. However, data on the extent and location of LM were mostly insufficient throughout the included studies, which limited the comparison of the interventions.
Eight studies were found that reported subgroup outcome of a cohort >5 WTLM patients (irrespective of the kind of local treatment). Five of the eight included studies have also been included in analysis I and are discussed in this Sects. [3,4,5,6,7]. Another two ROS-SC studies were found that included pediatric patients with stage IV WT presenting with lung and/or liver metastasis [8, 10]. Both studies had no comparative design and local treatment for metastasis was either radiotherapy (RT) or not performed at all. The third study was a retrospective analysis of a large oncological trial registry . This study compared a group of 236 patients with stage IV WT (synchronous lung and/or liver metastasis) to a group of 244 patients with relapsed WT and metachronous lung and/or liver metastasis. Local treatment of metastases was only briefly addressed in this study, and it was not reported how many patients received LR and/or RT as local therapy for LM. However, this study by Breslow et al. provided good quality regarding the prognostic difference of synchronous or metachronous WTLM. All three studies had adequate follow-up periods and a low selection bias. In summary, the quality of the existing studies on WTLM is low and most studies bear a moderate to high ROB. The studies were rather heterogenous, some with small caseload, and all had a retrospective design. A study designed to compare surgical vs. non-surgical treatment of LM has not been conducted yet. Results of the ROB are depicted in Tables 1 and 2.
Result of Analysis I: Liver Resection for Hepatic Metastases of Wilms’ Tumor
The 14 included studies reported on 212 pediatric patients with WTLM. Mean follow-up among the studies was 46 months (median 54, SD 31). Synchronous LM were found in 169 patients. 30 patients had a metachronous hepatic relapse. For 13 patients, the time of diagnosis of LM was not specified. Concurrent pulmonary metastases were present in 125 patients (59%). 67 patients had liver metastasis as only metastatic site (32%), and information on metastatic pattern was missing for 20 patients (9%). Table 3 gives an overview of the included studies of analysis I.
Liver resection for WTLM
In 93 of the 212 patients (44%), LR was performed. 29 (31%) atypical or wedge resections and 24 (26%) anatomical major LRs were reported. In 40 cases (43%), data on the type of surgical procedure were missing. Complete tumor resection was confirmed in 36 operations (39%). Incomplete resection was reported in 16 patients (17%). For 41 patients, there were no adequate data on the resection status (44%). Hepatic re-resections were performed in 16 children (17%). (See Table 4)
Among the 93 patients with LR, 60 had synchronous LM (65%), 20 had metachronous LM (22%), and for 13 patients (14%), the time of diagnosis of LM was not reported. Data on overall survival was available for 65 of all 93 patients who underwent LR. For those patients, OS was 69% and 20 deaths occurred in the reported follow-up period (31%). In the subgroup of patients with synchronous LM, OS was 75%. In the subgroup of patients with metachronous LM, OS was 65%. For the 36 patients with confirmed tumor free resection margins, OS was 92% (33 patients survived, three deaths in the reported follow-up period). (See Table 5; Fig. 2.)
Non-surgical treatment of WTLM
119 patients of the 212 did not undergo LR. For most of them, no subgroup outcome was reported. For 32 patients without LR, data on OS were available. OS was 53% (17 patients), 47% (15 patients) died of progressive disease.
Chemotherapy and timing of Liver resection
Information on applied chemotherapeutic regimens and the timing of LR was available for 70 of the 93 patients with LR. 48 patients (69%) received neoadjuvant chemotherapy before LR was performed. In 27 of those, LR was performed together with resection of the primary tumor. In 21 patients, LR was performed as a separate operation after neoadjuvant chemotherapy and resection of the primary tumor. In 22 patients (31%), upfront surgery with resection of the primary tumor and concurrent LR was performed before starting any kind of chemotherapy, followed by adjuvant chemotherapy in all 22 cases. OS of these different subgroups was not available based on the data of the included studies.
Five studies included both, patients with LR for WTLM and patients without surgery. Only the study by Liné et al. presented details on the treatment algorithm , that led to the decision of performing or not performing LR. The other studies were lacking those data. The study with the largest cohort of patients with WTLM (only synchronous) did not report the outcome of the LR subgroup . The only subgroup analyses performed in this NWTS study by Ehrlich et al. was a comparison of Event free survival (EFS) between 22 patients with LR for LM during the primary tumor operation versus 75 patients without primary LR. The group with LR had better EFS (5-year EFS 86%) compared to the group without (5-year FS 68%), but this difference was not significant. Another 4 patients received LR later during therapy, making for a total of 25 patients who underwent LR in this cohort. However, no outcome was reported for this subgroup of 25 patients with LR. Fuchs et al. analyzed 45 patients with WTLM treated within the Société Internationale d’Oncologie Pédiatrique/International Society of Paediatric Oncology (SIOP)/ Gesellschaft für pädiatrische Onkologie und Hämatologie (GPOH) studies . 29 had synchronous and 16 metachronous LM. In total, 21 patients underwent LR and 24 were treated without surgery for LM. OS was 57% in the group with surgery (12/21) and in 50% in those without surgery for LM (12/24), the difference being not significant (OR 1.3, 95%-CI 0.3-5.1, p = 0.631). OS was 100% in the 11 patients with complete resection of LM. In the study by Berger et al., 2 of the 6 patients with LM underwent LR. OS of those patients was 100%, while it was 25% for patients without LR . Liné et al. presented 18 patients with synchronous WTLM of whom 14 received LR . OS was 86% for these patients with LR (12 patients). The analysis of Aronson et al. did not include subgroup outcomes of the 3 patients with LR and the 13 with non-surgical treatment of LM . Overall, the resection rate for WTLM varied from 16 to 79% (median 33.3%, SD 21) across the five studies.
High-risk histology (HR) was defined as blastemal predominance or diffuse anaplasia. Other subtypes were defined as non-HR (including low-risk and intermediate-risk). In five studies, information on histologic subtypes of the patients’ primary tumor and/or liver metastases were given [3,4,5,6,7]. Thus, information on histology was available for 184 of 212 WTLM patients (with or without LR). In the largest study on WTLM patients, children with HR were excluded from the analysis . In total, 24 patients (13%) patients had HR, and 171 non-HR (93%). Diffuse Anaplasia was found in 11 patients (6%). Separate reports on the outcome of the histologic subgroups were only partly available. In the study by Fuchs et al., OS among the 9 patients with HR was 22%. Liné et al. analyzed 4 patients with HR: two with blastemal predominance and two with diffuse anaplasia. All four underwent LR. Both patients with blastemal subtype were long-term survivors, while the two patients with diffuse anaplasia both suffered from recurrence and died.
Results of Analysis II: Outcome of Patients with Liver Metastasis of Wilms’ Tumor – Irrespective of the Kind of Local Treatment
Studies with large cohorts of patients with metastatic WT that included at least five cases with WTLM and reported subgroup outcome were eligible for this analysis. All kinds of treatments for WTLM were included, meaning that patients either underwent no local therapy for LM (chemotherapy only), received LR plus chemotherapy, RT plus chemotherapy, or a combination of LR, RT and chemotherapy. Eight studies were found that reported the subgroup outcome of patients with WTLM (four of them have already been included in analysis I). A total of 273 WTLM patients were analyzed, of which 224 had synchronous, and 49 metachronous LM. 44 patients (16%) had HR. Patients with metachronous LM had significantly more often HR compared to those with synchronous LM (33% vs. 13%, OR 3.4, 95% CI 1.6 – 7.4, p < 0.001). The local treatment regimens differed among the studies. Radiotherapy (RT) was the only applied local treatment of LM in two studies [8, 10]. One trial predominantly applied local RT and only few received LR . In one cohort, RT and LR were applied with a similar rate . A higher rate of LR and lower administration of RT was used in another two studies [3, 7]. In two studies, the only applied local treatment for LM was surgery [5, 6]. The outcome varied among the studies, with OS ranging from 13 to 89%. The weighted mean OS of all WTLM patients across the seven studies was 55% (SD 29). For patients with synchronous LM, OS was significantly higher than for those with metachronous LM (63% vs. 22.5%, OR 4.2, 95%-CI 2.0-9.3, p < 0.001). Table 6 gives an overview of the included studies in analysis II. (See Figs. 3 and 4)
Multiple Meta-Regression Analysis
The 7 studies [3,4,5,6,7,8, 10, 36] with subgroups of WTLM patients < 5 were included in a multiple meta-regression model for OS with 4 covariates, as explained in the 2 section. I2 (residual heterogeneity) was 29% (See Table 7). 94% of the heterogeneity of the effect sizes among the trials (i.e., differences in OS), were explained by the multiple meta-regression model (R2 = 94.31%). Liver resection rate was the only significant predictor of OS, meaning that higher rates of liver resection were associated with higher OS (coefficient 0.819, p = 0.038). See Table 7.