Four databases, namely PubMed, Embase, Web of Science, and Cochrane Library, were searched for English articles related to the studied topic from conception to April 16, 2023. The following keywords and MeSH terms were used in PubMed search: hepatocellular carcinoma, hepatectomy, hepatic resection, liver resection, and blood transfusion. The details of the search strategy for all databases are shown in Supplementary Material S1 . References in the identified studies were further searched manually for additional relevant studies.

The inclusion criteria were as follows: (1) articles that reported randomized controlled trials (RCTs) and PSM studies and were published in English; (2) studies that used LR with or without allogeneic blood transfusion for treating HCC patients; and (3) studies that reported overall survival (OS) and/or recurrence-free survival (RFS).

The following types of studies were excluded: (1) studies that focused on autologous transfusion and (2) noncomparative and non-PSM studies, abstracts, case reports, and reviews. Moreover, for studies with overlapped patient cohorts, only the top study (highest quality, largest sample size, or the most recent one) was included, and the remaining studies were excluded.

OS was defined as the time between surgery and death and RFS was defined as the time from surgery to tumor recurrence. Herein, OS and RFS were considered as primary time-to-event outcomes. 1-, 3-, and 5-years overall survival rates and recurrence-free survival rates were defined as the percent of patients survival or without tumor recurrence at 1, 3, and 5 years after liver resection.

The initial quality assessment of each study and the subsequent data extraction were conducted independently by two researchers (LB Hu and XP Shi). For nonrandomized comparative trials, the Newcastle-Ottawa scale (NOS) with a score of up to 9 points (5 or less for low quality; 6–7 for medium quality; and 8 or more for high quality) was used for quality assessment (17).

Predesigned and standardized forms were used to extract the study details from the included studies (e.g., first author, year of publication, patient information, and tumor characteristics). The primary outcomes, including OS, RFS, survival rates (1-, 3-, and 5-year), and RFS rates, were extracted either directly from the original reports or indirectly by estimation with the Kaplan-Meier curve using the Engauge Digitizer software (version 4.1) based on the approach introduced by Tierney et al. (18, 19). Any disagreements between the two independent researchers were resolved by a third researcher (AD Wang).

The inverse variance method was used to determine the hazard ratio (HR) and 95% confidence interval (CI) values. The Mantel-Haenszel method was used to estimate the risk ratio (RR) and 95% CI values. Heterogeneity was assessed using the χ² method (I² values of 25% and 50% indicated low heterogeneity and moderate heterogeneity, respectively). The test model was selected based on the heterogeneity level, and the random-effects model was used for studies with I² > 50% (20). The robustness of the conclusion was determined by sensitivity analysis. Subgroup analysis was based on the Barcelona Clinic Liver Cancer (BCLC) staging system and the balance in the volume of blood loss between the two groups. The publication bias was assessed using funnel plots with Begg’s test and Egger’s test. A p-value of <0.05 was considered to indicate statistical significance. All statistical analyses were conducted in R program (version 4.2.3).

The database search yielded 2,096 articles, of which 2,060 articles were excluded after reviewing the title and abstract ( Figure 1 ). Among the remaining articles, 27 articles did not meet the inclusion criteria and were therefore excluded. Thus, 9 studies were selected for the meta-analysis (8–16).

All the 9 included studies with 12 datasets were PSM studies including 8028 patients, of which 21.50% of patients (1726) received perioperative allogeneic blood transfusion, and the remaining 78.50% of patients did not receive perioperative allogeneic blood transfusion. After the PSM approach with 1:1 match, the includded studies involved 2476 patients, of which 50% patients (1238) received perioperative allogeneic blood transfusion (the PBT group), and the remaining 50% patients did not receive perioperative allogeneic blood transfusion (the NPBT group). Patient characteristics are shown in Table 1 . The median volume of blood loss ranged from 400 to 1870 mL in the PBT group and from 200 to 1200 mL in the NPBT group. The volume of blood loss between the PBT and NPBT groups was balanced in 6 studies (8, 10–12, 15, 16) and not balanced in 3 studies (9, 13, 14). Three studies assessed the effect of blood transfusion on long-term prognosis by stratifying the patients according to the tumor stage (10–12). Four studies excluded patients died in early period post operation (30-day, 90-day, or in-hospital) (9, 11, 12, 15), while other studies included these patients.

The details of the quality assessment of the included studies are shown in Supplementary Material S2 . Studies with a balance in the volume of blood loss between the PBT and NPBT groups scored 9 points, and those with no balance in the volume of blood loss between both groups scored 8 points. Thus, all studies were considered high-quality studies.

The HR values of OS were available for all studies. The random-effects model was used because of significant heterogeneity among the studies. The pooled data showed that patients in the PBT group had lower OS (HR, 1.34; 95% CI, 1.10–1.64; p < 0.01) ( Figure 2 ). The 1-, 3-, and 5-year survival rates were reported in 4, 5, and 9 studies, respectively. Because the studies showed significant heterogeneity in the 3- and 5-year survival rates, the random-effects model was used to pool the results from these studies. Patients in the PBT group had lower 1-, 3-, and 5-year survival rates (RR, 0.89; 95% CI, 0.84–0.96; p < 0.01; RR, 0.7839; 95% CI, 0.6150–0.9992; p = 0.0492; RR, 0.70; 95% CI, 0.51–0.96; p = 0.026, respectively) ( Table 2 ).

The HR values of RFS were available for all studies. The random-effects model was used because of significant heterogeneity among the studies. The pooled data showed that patients in the PBT group had lower RFS rates (HR, 1.29; 95% CI, 1.07–1.56; p < 0.01) ( Figure 2 ). The 1-, 3-, and 5-year RFS rates were reported in 4, 5, and 8 studies, respectively. Because the studies showed significant heterogeneity in 1-, 3-, and 5-year RFS rates, the random-effects model was used to pool the results from these studies. Patients in both groups showed similar 1-, 3-, and 5-year RFS rates (RR, 0.75; 95% CI, 0.51–1.11; p = 0.1475; RR, 0.70; 95% CI, 0.41–1.20; p = 0.1927; RR, 0.63; 95% CI, 0.39–1.02; p = 0.0583, respectively) ( Table 2 ).

The sensitivity analysis revealed good robustness of OS and RFS results ( Supplementary Material S3 ). Funnel plots of the OS and RFS data combined with Begg’s test and Egger’s test indicated no significant publication bias ( Supplementary Material S4 ).

The details of the subgroup analysis are shown in Supplementary Material S5-7 . For patients with BCLC A HCC, those in the PBT group had lower OS and RFS rates (HR, 2.27; 95% CI, 1.61–3.21; HR, 2.11; 95% CI, 1.30–3.41, respectively). The OS and RFS rates in both groups were similar for patients with BCLC B and C HCC ( Supplementary Material S5 ). Regarding the balance in the volume of blood loss between the two groups in the included studies, subgroup analysis showed better OS and RFS rates for the NPBT group regardless of balance or no balance in the volume of blood loss between the two groups ( Supplementary Material S6 ). Subgroup analysis of studies excluded the patients died in early period post operation showed that those patients in the PBT group had lower OS and RFS rates (HR, 1.35; 95% CI, 1.10–1.67; HR, 1.33; 95% CI, 1.16–1.53, respectively), however, subgroup analysis of studies included the patients died in early period post operative showed that patients in both groups had similar OS and RFS.