Association of perioperative allogeneic blood transfusions and long-term outcomes following radical surgery for gastric and colorectal cancers: systematic review and meta-analysis of propensity-adjusted observational studies

Abstract

Background

Previous meta-analyses reporting significant associations between perioperative allogeneic blood transfusions and poor prognosis in gastric cancer or colorectal cancer had a high risk of confounding bias. This meta-analysis explored this issue using observational studies that applied propensity score analysis.

Methods

PubMed, Embase, and the Cochrane Central Register of Controlled Trials were searched for manuscripts published between 2013 and 2022. Studies applying propensity score analysis were included to investigate the association between perioperative allogeneic blood transfusions and prognosis in gastric cancer or colorectal cancer after radical surgery. Pooled HRs for overall survival and disease-free survival were calculated using a fixed-effect model or random-effect model according to heterogeneity.

Results

Twelve retrospective cohort studies with 17 607 patients reported were included. Ten studies applied propensity score matching and two applied inverse probability of treatment weighting using propensity score. A total of 5962 patients were analysed after propensity score adjustment. After propensity score adjustment, perioperative allogeneic blood transfusions did not correlate with disease-free survival in gastric cancer (HR 1.16; 95 per cent c.i. 0.96–1.39; heterogeneity was assessed by the chi-squared test and inconsistency index (I²) = 57 per cent) or colorectal cancer (HR 1.12; 95 per cent c.i. 0.84–1.49; I² = 54 per cent). However, after propensity score adjustment, perioperative allogeneic blood transfusions were significantly associated with worse overall survival in gastric cancer (HR 1.20; 95 per cent c.i. 1.08–1.32; I² = 25 per cent) and colorectal cancer (HR 1.40; 95 per cent c.i. 1.06–1.85; I² = 52 per cent). Subgroup analyses showed that perioperative allogeneic blood transfusions did not correlate with overall survival in colorectal cancer when major postoperative complications were balanced after propensity score.

Conclusion

Perioperative allogeneic blood transfusion is not correlated with recurrence of gastric cancer and colorectal cancer. Perioperative allogeneic blood transfusions are significantly associated with worse overall survival in gastric cancer and colorectal cancer, which may be attributable to unbalanced major postoperative complications after propensity score adjustment.

In the present study, we performed a systematic review and meta-analysis to investigate the influence of perioperative allogeneic blood transfusions on the recurrence and overall survival in gastric cancer and colorectal cancer after radical surgery using observational studies that applied propensity score analysis. We found that perioperative allogeneic blood transfusions were not associated with recurrence but were associated with worse overall survival in gastric cancer and colorectal cancer.

Introduction

Allogeneic blood transfusion is frequently utilized for patients with gastric cancer (GC) or colorectal cancer (CRC) in the perioperative period because of a high incidence of perioperative anaemia and complex oncologic surgery. Previous meta-analyses of observational studies reported that perioperative allogeneic blood transfusions (PABTs) were significantly associated with recurrence and worse overall survival (OS) in GC and CRC^1–9. As a consequence, PABTs were thought to have a detrimental effect on prognosis, for which transfusion-associated immunomodulation was thought to be the primary cause¹⁰. However, in recent years, growing support has developed for the opposite view that the worse prognosis in patients receiving PABTs is caused by the clinical circumstances necessitating PABTs, such as advanced tumour stage, preoperative anaemia, difficulty and duration of the procedure, old age and co-morbidities, and is not due to PABTs^11–13. In other words, the significant associations of PABTs with recurrence and worse OS may result from uncontrolled confounding bias. Consequently, it has been argued that restricting blood transfusions due to concerns about prognosis is not justified. Unfortunately, without effectively controlling confounding bias, previous meta-analyses failed to resolve the issue of whether the significant association between PABTs and worse prognosis is caused by PABTs or by other poor prognostic factors associated with PABTs. Owing to the lack of compelling evidence, surgeons are unsure when making decisions regarding transfusions for patients with GC and CRC during the perioperative period. Thus, determining the true relationship between PABTs and prognosis by adequately controlling confounding bias is particularly important and performing an RCT by comparing administering PABTs with withholding PABTs is impossible for ethical reasons. Alternatives include the application of other statistical methods, such as propensity score (PS) analysis, instrumental variable analysis, and graphical causal model (Directed Acyclic Graph, DAG). Similar to randomization, PS analysis, which includes propensity score matching (PSM), stratification on the PS and inverse probability of treatment weighting (IPTW) using the PS, reduces confounding bias by comparing outcomes in treated and untreated subjects who have similar distributions of measured baseline covariates^14,15. In recent years, several observational studies have applied PS analysis to investigate the association between PABTs and prognosis in GC or CRC, but their findings are highly inconsistent^16–27. Most of these studies involved a single-centre cohort with a small sample size. The one study involving multiple centres analysed only 312 patients after PSM²⁴. In addition, the PS analysis methods differed among these studies. Consequently, there is still a significant knowledge gap as to the association between PABTs and prognosis in GC and CRC.

In the present study, a systematic review and meta-analysis were performed to investigate the influence of PABTs on recurrence and OS of GC and CRC after radical surgery using observational studies that applied PS analysis.

Methods

The present systematic review and meta-analysis were performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement²⁸. Supplementary File shows the completed PRISMA checklist. The protocol for this systematic review and meta-analysis was defined a priori and registered at PROSPERO, the international prospective register of systematic reviews, under registration number CRD42022355942. The review protocol is available in full at https://www.crd.york.ac.uk/PROSPERO/.

All stages of the literature search, study selection, data extraction and quality assessment were carried out independently by two reviewers. Disagreements were resolved by consensus or based on independent assessment by a third reviewer.

Eligibility criteria

Patients (P): patients with GC or CRC undergoing radical surgery; Intervention (I): the use of PABTs; Comparison (C): patients with GC or CRC undergoing radical surgery who did not receive PABTs; Outcomes (O): OS, disease-free survival (DFS) and disease-specific survival (DSS). OS was defined as the time from the day of surgery to that of all-cause death. DFS was defined as the time from the day of surgery to that of tumour recurrence. DSS was defined as the time from the day of surgery to that of death from cancer. All of the included studies were required to have reported at least one of the above-mentioned primary outcomes. The studies that did not provide hazard ratios (HRs) and associated 95 per cent confidence intervals (c.i.) for at least one of the above-mentioned primary outcomes and did not provide sufficient data to calculate them by the method reported by Tierney et al.²⁹ were excluded. Study design (S): peer-reviewed prospective and retrospective studies applying PS analysis. Review articles, editorials, commentaries, and conference abstracts were excluded. Studies that investigated the prognostic role of PABTs without PS analysis were also excluded; timing: all studies published up to 27 July, 2022 and language: English.

In cases in which multiple reports had been published from the same institution with identical or overlapping patient cohorts, only the most informative paper was included.

Literature search

PubMed, Embase and the Cochrane Central Register of Controlled Trials were searched from their inception to 27 July, 2022. The detailed search strategies are reported in the Supplementary material. No restrictions in terms of language or publication status were used. Reference lists of included studies were also screened to identify additional eligible studies.

Study selection

Reference management software (EndNote 19.1) was used to merge results from different electronic databases and remove duplicate studies. After the removal of duplicates, the titles and abstracts of the searched articles were assessed for potentially eligible articles. Thereafter, the full texts of the potentially eligible articles were assessed according to the eligibility criteria. Reasons for exclusion during the full-text screening were recorded.

Data extraction and quality assessment

Data extraction was undertaken using a standardized Microsoft Excel extraction form. All extracted variables were crosschecked to ensure their reliability. The following data from the included studies were extracted: first author, publication year, country, recruitment period, study design, sample size, single centre or multiple centres, follow-up, age, sex, rate of PABTs, definition of perioperative period, blood products, transfusion criteria, tumour lymph node metastasis (TNM) stage, PS analysis method, factors used in the PS analysis and unbalanced factors after PS adjustment. Subsequently, the HRs and 95 per cent c.i. of PABTs associated with OS, DSS and DFS before and after PS adjustment were extracted. If an HR or 95 per cent c.i. was not given, they were calculated from other data provided in the article as previously reported²⁹. Major postoperative complications (MPCs; Clavien–Dindo grade III or higher complications) were extracted if reported. When important outcome data were missing, the authors of the article were contacted by e-mail to obtain such data.

The Newcastle–Ottawa Scale (NOS) was used for quality assessment³⁰. Studies were rated on a scale with a maximum score of nine. High quality, moderate quality and low quality were defined as total scores of 7–9, 4–6 and <4 respectively. The assessment was performed at the study level because the assessment results of OS were similar to those of DFS in all included studies.

Statistical analysis

The primary outcomes of this meta-analysis were OS, DFS and DSS. The HR was used to measure the effects on the primary outcomes. The secondary outcome of this meta-analysis was MPCs, the effects on which were measured using the risk ratio (RR). An HR or RR greater than 1 suggested a worse outcome in the PABT group and was considered statistically significant if the 95 per cent c.i. did not overlap with 1, with P <0.050. Study estimates, along with pooled estimates, are presented as forest plots. Heterogeneity was assessed by the chi-squared test and inconsistency index (I²). The results are expressed as chi-squared test P values (P value <0.100: significant) and I² values (I² < 25 per cent: no heterogeneity, 25 per cent < I² < 50 per cent: low heterogeneity, 50 per cent < I² < 75 per cent: moderate heterogeneity, I² > 75 per cent: high heterogeneity). To calculate pooled HRs and RRs, a fixed-effect model (inverse variance method for HRs and Mantel–Haenszel method for RRs) was used if no significant heterogeneity was observed. Otherwise, a random-effect model (DerSimonian and Laird method) was used. The effect of publication bias on the reported outcomes was assessed graphically using funnel plots. Different subgroup analyses were performed on studies reporting that there was no difference in MPCs between the PABT group and non-PABT group after PS adjustment, and on studies reporting that there was a significant difference in MPCs between the PABT group and the non-PABT group after PS adjustment. Sensitivity analyses were conducted that considered high-quality studies, studies investigating the prognostic role of red blood cell (RBC) transfusion and studies applying PSM. A sensitivity analysis was also conducted by removing the study investigating the prognostic role of intraoperative blood transfusion. All analyses were performed using Review Manager Version 5.4 software.

Results

Study selection

The search strategy retrieved 5685 articles. After removing duplicates and screening the title and abstract, 94 full texts were reviewed, of which 82 articles were excluded for various reasons. No additional eligible studies were identified by screening the reference lists of the included studies. As a result, 12 studies were included^16–27. A flow diagram of the study selection process is shown in Fig. 1. A list of the excluded studies and the reasons for their exclusion are detailed in the Supplementary material.

Fig. 1.

Flow diagram of the study selection process

*Reasons described in Supplementary material.

Study and patient characteristics

Of the 12 included studies, five studies investigated the prognostic role of PABTs in GC^16–20 and seven studies investigated the prognostic role of PABTs in CRC^21–27. A total of 17 607 patients were included in the present systematic review and meta-analysis, of whom 3449 patients (19.59 per cent) received PABTs. The transfusion rate in the included studies ranged from 1.93 per cent to 54.11 per cent. All studies were retrospective cohort studies and their publication dates ranged from 2013 to 2022. Of the 12 included studies, eight studies^{16–20,22,23,27} investigated the role of RBC transfusion, of which two studies^20,22 stated that only leucocyte-depleted RBCs were transfused, and the other six studies^{16–19,23,27} did not state whether RBCs were leucocyte depleted. The remaining four studies^21,24–26 did not report which types of blood products were transfused. Of the 12 included studies, six studies originated from Asia^{16–19,25,27}, five from Europe^20–23,26, and one from the USA²⁴. All studies except one²⁴ were performed at a single centre. All studies reported data on OS, nine studies reported data on DFS^{16–18,20,21,24–27} and only one study reported data on DSS²³. The characteristics of the study and patients from the included studies are summarized in Tables 1 and 2.

Table 1.

Study characteristics of included studies

Study	Sample size	Country	Recruitment period	Study design	Multicentre	Follow-up (months) median (range)	Outcome
GC
Hsu 202116	569	China	2009–2014	RC	No	59.8 (i.q.r. 22.8–90.7)	OS DFS
Song 202217	2905	Korea	2006–2015	RC	No	62 (1–153)	OS DFS MPCs*
Xiao 201818	1020	China	2010–2015	RC	No	33 (3–86)	OS DFS
Cui 201619	1150	China	2003–2008	RC	No	40†	OS
Reim 201620	610	Germany	2001–2013	RC	No	41 (0.1–153)	OS DFS
CRC
Warschkow 201421	401	Switzerland	1996–2008	RC	No	34.2 (2–161)	OS DFS
McSorley 202022	544	UK	2012–2017	RC	No	43 (12–77)	OS MPCs*
Turri 202223	895	Italy	2005–2017	RC	No	Minimum 24†	OS DSS
Hanna 202124	924	USA	2010–2018	RC	Yes	Mean(s.d.) 34.8(25.2)	OS DFS
Kang 202225	4250	China	2011–2020	RC	No	37 (1–114)	OS DFS MPCs*
Tarantino 201326	309	Switzerland	1996–2008	RC	No	Mean(s.d.) 47(38)	OS DFS
Wu 201827	4030	China	2005–2014	RC	No	46.1 (i.q.r. 24.7–73.1)	OS DFS

GC, gastric cancer; CRC, colorectal cancer; RC, retrospective cohort; i.q.r., interquartile range; s.d., standard deviation; OS, overall survival; DFS, disease-free survival; DSS, disease-specific survival; MPCs, major postoperative complications. *Major postoperative complications were defined as Clavien–Dindo grade III or higher complications. †Range missing.

Table 2.

Patient characteristics of included studies

Study	Age (years, mean(s.d.))		Male (%)		Rate of PABTs (%)	Definition of perioperative to receive PABTsa	Blood products	Transfusion criteria	pTNM stage
	T	NT	T	NT
GC
Hsu 202116	>70: 63.13% (proportion)	>70: 44.25% (proportion)	68.75	62.35	28.12	From 14 days before to 7 days after surgery	RBC	NR	I–III
Song 202217	Median 65 (i.q.r. 55–73)	Median 58 (i.q.r. 49–68)	62.98	64.99	18.69	Within 30 days before and after surgery	RBC	Hb < 80 g/L, massive IBL, or hemodynamic instability	II, III
Xiao 201818	56.7(11.8)	54.1(10.3)	62.77	69.07	22.65	During perioperative hospitalization	RBC	Hb < 80 g/L, hemodynamic instability	II, III
Cui 201619	62.0 (Range 23–89)	60.5 (Range 26–93)	71.24	70.98	26	During perioperative hospitalization	RBC	HB < 70 g/L or massive IBL	I–III
Reim 201620	≤65: 50.18%	≤65: 57.91%	67.27	71.64	45.08	3 days before and after surgery	RBCb	NR	0–III
CRC
Warschkow 201421	67.0(10.7)	62.6(12.6)	60.37	67.93	54.11	NR	NR	NR	I–III
McSorley 202022	<65: 30.23% (proportion)	<65: 36.68% (proportion)	47.67	55.24	15.81	From 30 days before to 30 days after surgery	RBCb	NR	0–III
Turri 202223	Median 75.1 (i.q.r. 64.1–81.6)	Median 66.5 (i.q.r. 58.3–74.7)	55.13	58.86	29.39	From 30 days before to 90 days after surgery	RBC	Hb < 80 g/L	I–III
Hanna 202124	62(14)	59(12)	56	61	14.61	From the start of the surgery to discharge from the hospital	NR	NR	0–III
Kang 202225	66.8(14.9)	62.8(12.1)	56.1	58.78	1.93	During surgery	NR	NR	I–III
Tarantino 201326	72.3(10.6)	67.5(11.6)	66.89	60.87	47.90	From 3 days before to 7 days after surgery	NR	NR	I–III
Wu 201827	73(12)	66(13)	62.57	61.16	25.06	Within 7 days of surgery	RBC	NR	I–III

^aDefinition of perioperative period to receive PABTs: perioperative allogeneic blood transfusions

^bLeukocyte-depleted RBC

GC, gastric cancer; CRC, colorectal cancer; SD, standard deviation; IQR, interquartile range; T, transfusion; NT, non-transfusion; NR, not reported; RBC, red blood cell; Hb, hemoglobin; IBL, intraoperative blood loss; TNM, tumor-lymph node-metastasis.

Concerning the methods of PS analysis, 10 studies applied PSM^{17–19,21–27} and two studies applied IPTW^16,20. A total of 5962 patients were analysed to investigate the prognostic role of PABTs after PS adjustment. Concerning factors used in the PS analysis, all included studies used clinical characteristics (for example age, sex, BMI) and tumour characteristics (for example TNM stage, tumour size, lymphovascular invasion). In addition, 11 studies used co-morbidities and/or American Society of Anesthesiologists (ASA) class^{16–18–19,21–26–27}, 11 studies used surgery-related factors^{16–23–24,26,27}, seven studies used perioperative anaemia or perioperative haemoglobin level^{16,21–23–24,26,27}, seven studies used adjuvant therapy and/or neoadjuvant therapy^{16,19–22,26,27}, and two studies used postoperative complications^20,23. After PS adjustment, three studies stated that all confounders were well balanced^20,21,26, and the remaining nine studies stated that a few factors still differed between the PABT group and the non-PABT group^{16–19,22–25,27}. The characteristics of PS analysis of the included studies are summarized in Tables 3 and 4.

Table 3.

Characteristics of propensity score analysis of included studies (gastric cancer)

Study	Sample size* (T/NT)	Method	Factors used in the PS analysis	Unbalanced factors after PS adjustment
Hsu 202116	1018 (505/513)	IPTW	Sex, BMI, CCI, Hb, albumin, CEA, CA19-9, anaesthesia time, IBL, epidural analgesia, previous abdominal surgery, operation year, gastrectomy type, laparoscopic surgery, TNM stage, tumour size, histologic differentiation, LVI, positive margin, AT	Sex, histologic differentiation
Song 202217	996 (498/498)	PSM	Age, sex, BMI, ASA class, operation year, surgeon, laparoscopic surgery, gastrectomy type, number of retrieved LNs, operation duration, IBL, combined operation, PNI, tumour size, histologic differentiation, LVI, TNM stage	MPCs†
Xiao 201818	410 (205/205)	PSM	ASA class, age, BMI, co-morbidity, gastrectomy type, combined resection, tumour size, tumour location	Hb, IBL, OPCs
Cui 201619	458 (229/229)	PSM	Age, sex, BMI, WBCs, platelets, albumin, total protein, creatinine, urea, tumour size, histologic differentiation, gastrectomy type, lymphadenectomy extent, chemotherapy, TNM stage, operative factors (excluding IBL and Hb)	Hb, IBL
Reim 201620	610 (275/335)	IPTW	TNM stage, grading, gastrectomy type, Lauren histotype, MPCs†, sex, age, NAT, AT, LVI, splenectomy, surgical extension, number of retrieved LNs, tumour localization	No

T, transfusion; NT, non-transfusion; PS, propensity score; IPTW, inverse probability of treatment weighting; PSM, propensity score matching; BMI, body mass index; CCI, Charlson co-morbidity index; Hb, haemoglobin; CEA, carcinoembryonic antigen; CA19-9, carbohydrate antigen 19-9; IBL, intraoperative blood loss; TNM, tumour lymph node metastasis; MPCs, major postoperative complications; LVI, lymphovascular invasion; AT, adjuvant treatment; ASA, American Society of Anesthesiologists; LNs, lymph nodes; PNI, prognostic nutritional index; WBCs, white blood cells; NAT, neoadjuvant treatment; LOS, length of stay; mGPS, modified Glasgow Prognostic Score; CACI, Charlson-Age Co-morbidity Index; CHF, chronic heart failure; T2DM, type 2 diabetes mellitus; INR, international normalized ratio; CRP, C-reactive protein; OPCs, overall postoperative complications. *After propensity score adjustment. †Major postoperative complications were defined as Clavien–Dindo grade III or higher complications.

Table 4.

Characteristics of propensity score analysis of included studies (colorectal cancer)

Study	Sample size* (T/NT)	Method	Factors used in the PS analysis	Unbalanced factors after PS adjustment
Warschkow 201421	336 (185/151)	PSM	Age, sex, BMI, ASA class, TNM stage, distance from anal verge, NAT, NAT regimes, operation type, mesorectal excision extent, surgeon, number of retrieved LNs, distal resection margin, operation duration, IBL, LOS, Hb, AT	No
McSorley 202022	116 (58/58)	PSM	Age, sex, BMI, ASA class, mGPS, tumour site, TNM stage, NAT, laparoscopic surgery, intraoperative dexamethasone, anaemia	Hb, postoperative CRP, albumin, mGPS, MPCs†, LOS
Turri 202223	330 (165/165)	PSM	Operation year, age, sex, tumour location, CACI score, Hb, TNM stage, MPCs†	BMI, LOS
Hanna 202124	312 (100/212)	PSM	Age, sex, race, BMI, ASA class, diabetes, previous cardiac event, CHF, renal failure, dialysis, anaemia, operation type, operation duration, TNM stage, differentiation, resection status, IBL	IBL, operation duration, LOS
Kang 202225	164 (82/82)	PSM	Age, sex, BMI, smoking, drinking, hypertension, T2DM, tumour location, TNM stage	Operation duration, IBL, LOS, OPCs‡
Tarantino 201326	240 (118/122)	PSM	Age, sex, ASA class, BMI, TNM stage, operation type, operation duration, IBL, Hb, AT	No
Wu 201827	972 (486/486)	PSM	Hb, platelets, INR, age, sex, ASA class, diabetes, coronary arterial disease, heart failure, stroke, chronic kidney disease, CEA, tumour location, epidural block, anaesthesia time, TNM stage, differentiation, LVI, perineural invasion, AT, NAT	Sex, anaesthesia time, laparoscopic surgery

T, transfusion; NT, non-transfusion; PS, propensity score; PSM, propensity score matching; BMI, body mass index; CCI, Charlson co-morbidity index; Hb, haemoglobin; CEA, carcinoembryonic antigen; CA19-9, carbohydrate antigen 19-9; IBL, intraoperative blood loss; TNM, tumour lymph node metastasis; MPCs, major postoperative complications; LVI, lymphovascular invasion; AT, adjuvant treatment; ASA, American Society of Anesthesiologists; LNs, lymph nodes; PNI, prognostic nutritional index; WBCs, white blood cells; NAT, neoadjuvant treatment; LOS, length of stay; mGPS, modified Glasgow Prognostic Score; CACI, Charlson-age co-morbidity Index; CHF, chronic heart failure; T2DM, type 2 diabetes mellitus; INR, international normalized ratio; CRP, C-reactive protein; OPCs, overall postoperative complications. *After propensity score adjustment. †Major postoperative complications were defined as Clavien–Dindo grade III or higher complications. ‡Major postoperative complications were well balanced after propensity score adjustment.

Quality assessment

A summary of quality assessment for the included studies is shown in Table 5. Of the 12 included studies, 10^{17–21,23,25–27} and 2^22,24 studies were identified as being of high quality and moderate quality respectively.

Table 5.

Quality assessment for included studies

Study	Selection				Comparability	Outcome			Quality
	Domain 1	Domain 2	Domain 3	Domain 4	Domain 5	Domain 6	Domain 7	Domain 8
GC
Hsu 202116									High
Song 202217									High
Xiao 201818									High
Cui 201619									High
Reim 201620									High
CRC
Warschkow 201421									High
McSorley 202022									Moderate
Turri 202223									High
Hanna 202124									Moderate
Kang 202225									High
Tarantino 201326									High
Wu 201827									High

Green means one point. Red means zero point.

Domain 1: representativeness of the exposed cohort. Domain 2: selection of the non-exposed cohort. Domain 3: ascertainment of exposure. Domain 4: demonstration that outcome of interest was not present at start of study. Domain 5: comparability of cohorts on the basis of the design or analysis. Domain 6: assessment of outcome. Domain 7: was follow-up long enough for outcomes to occur? Domain 8: adequacy of follow up of cohorts.

Prognostic role of PABTs after PS adjustment

The pooled analyses of HR were performed for OS and DFS. The analyses were not performed for DSS because only one included study reported DSS.

For GC, the pooled analysis of HR for DFS after PS adjustment showed that PABTs were not associated with poor survival, with moderate heterogeneity (HR 1.16; 95 per cent c.i. 0.96–1.39; I² = 57 per cent; four studies with 3034 patients; Fig. 2). Meanwhile, the pooled analysis of HR for OS after PS adjustment showed that PABTs were significantly associated with poor survival, with low heterogeneity (HR 1.20; 95 per cent c.i. 1.08–1.32; I² = 25 per cent; five studies with 3492 patients; Fig. 2).

Fig. 2.

Forest plot and funnel plot of the meta-analysis of hazard ratio for survival a Disease-free survival in gastric cancer. b Overall survival in gastric cancer. c Disease-free survival in colorectal cancer. d Overall survival in colorectal cancer. CRC, colorectal cancer; DFS, disease-free survival; GC, gastric cancer; OS, overall survival; PABTs, perioperative allogeneic blood transfusions; SE, standard error.

For CRC, the pooled analysis of HR for DFS after PS adjustment showed that PABTs were not associated with poor survival, with moderate heterogeneity (HR 1.12; 95 per cent c.i. 0.84–1.49; I² = 54 per cent; five studies with 2024 patients; Fig. 2). Moreover, the pooled analysis of HR for OS after PS adjustment showed that PABTs were significantly associated with poor survival, with moderate heterogeneity (HR 1.40; 95 per cent c.i. 1.06–1.85; I² = 52 per cent; seven studies with 2470 patients; Fig. 2).

Funnel plots showed no significant publication bias for the reported outcome (Fig. 2).

Subgroup analyses

Two studies^23,25 focused on CRC reported there was no difference in MPCs between the PABT group and the non-PABT group after PS adjustment. Subgroup analysis for these two studies shows that PABTs were not associated with OS, with no heterogeneity (HR 1.26; 95 per cent c.i. 0.91–1.74; I² = 0 per cent; 494 patients; Fig. 3). Subgroup analyses could not be performed for other subgroups because there was only one study in each subgroup.

Fig. 3.

Forest plot of subgroup analyses and sensitivity analyses a Subgroup analysis for studies that reported that there was no difference in major postoperative complications between the perioperative allogeneic blood transfusion group and the no perioperative allogeneic blood transfusion group after propensity score adjustment in colorectal cancer (overall survival). b Sensitivity analyses for red blood cell transfusions in colorectal cancer (overall survival). c and d Sensitivity analyses for propensity score matching in gastric cancer (c, overall survival; d, disease-free survival). e and f Sensitivity analyses by removing intraoperative blood transfusions in colorectal cancer (e, disease-free survival; f, overall survival). g and h Sensitivity analyses for high-quality studies in colorectal cancer (g, overall survival; h, disease-free survival). CRC, colorectal cancer; DFS, disease-free survival; GC, gastric cancer; OS, overall survival; PABTs, perioperative allogeneic blood transfusions; PSM, propensity score matching; RBC, red blood cell; MPCs, major postoperative complications.

Sensitivity analyses

Three studies^22,23,27 focused on CRC investigated the prognostic role of RBC transfusion, and sensitivity analyses of these studies showed that PABTs were significantly associated with poor OS, with moderate heterogeneity (HR 1.64; 95 per cent c.i. 1.08–2.50; I² = 50 per cent; three studies with 1418 patients; Fig. 3). All of the included studies focused on GC investigated the prognostic role of RBC transfusion; therefore, such sensitivity analysis was not needed for GC.

Three studies^17–19 focused on GC investigated the prognostic role of PABTs by applying PSM. Sensitivity analyses of these studies showed that PABTs were marginally associated with poor OS, with low heterogeneity (HR 1.14; 95 per cent c.i. 1.00–1.30; I² = 49 per cent; three studies with 1864 patients; Fig. 3), and were not associated with DFS, with no heterogeneity (HR 1.01; 95 per cent c.i. 0.85–1.20; I² = 0 per cent; two studies with 1406 patients; Fig. 3). The included studies focusing on CRC were all conducted by applying PSM, so such sensitivity analysis was not needed for CRC.

One study²⁵ on CRC investigated the prognostic role of intraoperative blood transfusion. Removing this study, sensitivity analyses showed that PABTs were not associated with poor DFS, with low heterogeneity (HR 1.06; 95 per cent c.i. 0.88–1.29; I² = 46 per cent; four studies with 1860 patients; Fig. 3), and were significantly associated with poor OS, with moderate heterogeneity (HR 1.39; 95 per cent c.i. 1.02–1.91; I² = 60 per cent; six studies with 2306 patients; Fig. 3).

Five studies^{21,23,25–27} on CRC were identified as being of high quality. Sensitivity analyses of these studies showed that PABTs were marginally associated with poor OS, with low heterogeneity (HR 1.23; 95 per cent c.i. 1.00–1.51; I² = 48 per cent; five studies with 2042 patients; Fig. 3), and were not associated with DFS, with moderate heterogeneity (HR 1.13; 95 per cent c.i. 0.80–1.60; I² = 65 per cent; four studies with 1712 patients; Fig. 3). The included studies regarding GC were all identified as being of high quality, so sensitivity analysis was not needed for GC.

MPCs

Two studies^22,25 focused on CRC and one study¹⁷ focused on GC reported the data on MPCs after PS adjustment. Meta-analysis showed that PABTs were not associated with MPCs in CRC and GC, with moderate heterogeneity (RR 2.49; 95 per cent c.i. 0.97–6.36; I² = 61 per cent; three studies with 1276 patients; Fig. 4), and in CRC, with high heterogeneity (RR 2.33; 95 per cent c.i. 0.24–22.25; I² = 78 per cent; two studies with 280 patients; Fig. 4).

Fig. 4.

Forest plot of the meta-analysis of risk ratio for major postoperative complications in gastric cancer and colorectal cancer (a), and in colorectal cancer (b) GC, gastric cancer; MPCs, major postoperative complications; PABTs, perioperative allogeneic blood transfusions; RBC, red blood cell.

Additional analyses

A meta-analysis was also performed for OS and DFS before PS adjustment that were provided by the included studies. For GC, meta-analyses showed significant associations of PABTs with DFS and OS when using unadjusted data obtained from univariate analysis, and when using adjusted data obtained from multivariate Cox regression without PS adjustment. For CRC, meta-analyses showed similar significant associations, Table 6.

Table 6.

Meta-analysis of survivals

Outcome	Studies	Patients	HR (95% c.i.)	Heterogeneity (I2)
GC
DFS (univariate analysis)	4	5104	1.59 (1.17–2.17)	88%
DFS (multivariate Cox regression)	3	2199	1.39 (1.16–1.65)	13%
OS (univariate analysis)	5	6254	1.75 (1.42–2.15)	82%
OS (multivariate Cox regression)	4	3349	1.45 (1.19–1.75)	58%
CRC
DFS (univariate analysis)	4	8990	1.79 (1.59–2.03)	0%
DFS (multivariate Cox regression)	4	8990	1.35 (1.17–1.54)	0%
OS (univariate analysis)	6	10 429	2.51 (2.06–3.05)	52%
OS (multivariate Cox regression)	5	9885	1.71 (1.48–1.99)	2%

GC, gastric cancer; CRC, colorectal cancer; DFS, disease-free survival; OS, overall survival; HR, hazard ratio; c.i., confidence interval.

Discussion

This systematic review and meta-analysis investigated the association between PABTs and prognosis in GC and CRC based on observational studies that applied PS analysis. The present meta-analysis suggested that PABTs are not associated with DFS in GC and CRC when confounding bias is minimized after PS adjustment. These findings support the view that the greater number of recurrences observed in patients receiving PABTs is not due to the PABTs but rather to the worse clinical, pathological and surgical characteristics that necessitate PABTs. In other words, PABTs themselves may not lead to increased recurrence but are just a surrogate marker for patients with a high risk of recurrence. Additional evidence from RCTs found that perioperative autologous transfusions or leucocyte-depleted blood transfusions do not result in decreased recurrence compared with allogeneic transfusions or blood transfusions without leucocyte depletion in CRC, which disproved the hypothesis that transfusion-associated immunomodulation promotes recurrence^31,32. Nevertheless, patient blood management programmes have been proven to reduce transfusion rates and postoperative infections, and postoperative infections may favour tumour recurrence³³; however, the optimization of patient perioperative blood management is warranted.

In contrast to the findings for DFS, the present meta-analysis found that PABTs were significantly associated with worse OS after PS adjustment in GC and CRC. It has been reported that tumour-related factors (for example TNM stage), treatment-related factors (for example intraoperative blood loss, adjuvant therapy), patient-related factors (for example age, anaemia), postoperative recurrence, and MPCs affect OS in GC and CRC^34,35. As described above, recurrence was not associated with PABTs, and tumour-related factors, treatment-related factors and patient-related factors were reported to be balanced in most included studies. Consequently, the significant association between worse OS and PABTs may be attributable to MPCs, and subgroup analysis was performed for studies reporting that there was no difference in MPCs between the PABT group and the non-PABT group after PS adjustment in CRC. Interestingly, the subgroup analysis found that the significant association disappeared and there was no heterogeneity when MPCs were balanced. However, it should be noted that only two included studies were available for the subgroup analysis. Considering that MPCs are postoperative outcomes just like OS and DFS, whether PABTs cause more MPCs is undetermined. The present meta-analysis found that PABTs are not associated with more MPCs when other confounders are reduced by PS analysis in GC and CRC. Nevertheless, the results should be interpreted with caution because only three included studies were eligible for the analysis, and a wide 95 per cent c.i. and high heterogeneity were identified. Two RCTs found that autologous transfusion or leucocyte-depleted blood transfusions reduced postoperative infectious complications in patients with CRC compared with allogeneic transfusions or blood transfusions without leucocyte depletion, suggesting that transfusion-associated immunomodulation has detrimental effects on MPCs^36,37. Conversely, another RCT reported similar rates of postoperative infectious complications between a leucocyte-depleted blood transfusion group and a non-leucocyte-depleted blood transfusion group in CRC³². Thus, the effect of PABTs on MPCs in GC and CRC is still debatable, and further studies are needed to explore this issue.

It is commonly agreed that RCTs are the ‘standard’ for treatment evaluation³⁸. However, performing an RCT by comparing administering PABTs with withholding PABTs is impossible for ethical reasons. RCTs have also often been criticized for limited external validity, which results from the enrolment of highly selected patient groups^38,39. In contrast, observational studies are conducted in the general population, which are expected to have larger external validity. Nonetheless, observational studies exhibit a significant risk of confounding bias, which results in limited internal validity⁴⁰. PS analyses are sound statistical techniques for dealing with confounding bias in observational studies, which reduces confounding bias by comparing outcomes in treated and untreated subjects who have similar distributions of measured baseline covariates^14,15. Therefore, observational studies that apply PS analysis are thought to have both large external validity and large internal validity. Since 2013, several observational studies have applied PS analysis to investigate the associations of PABTs with recurrence and/or OS in GC or CRC, but their findings are highly inconsistent^16–27. The conflicting results are mainly due to small sample sizes, single-centre approaches and differences in the PS analysis methods used in these studies. As a consequence, the present meta-analysis was performed to obtain more convincing results and to provide insight into the association between PABTs and prognosis. The findings of the present study are important to aid decision-making regarding blood transfusions when managing patients with GC and CRC during the perioperative period.

Several limitations exist in the present meta-analysis. First, despite only studies applying PS analysis being included in the analysis and more than 80 per cent of the included studies being graded as high quality, confounding bias still affected this study. This is because PS analysis could not balance factors that have not been recorded or are unknown. Second, four included studies did not report which types of blood products (RBCs, fresh frozen plasma, platelets, whole blood) were transfused. Different blood products may cause different degrees of immunomodulation and show various associations with prognosis. Nevertheless, the subgroup analysis shows that RBC transfusions significantly correlated with OS in GC and CRC, and did not correlate with recurrence of GC. Third, significant heterogeneity was identified in several pooled analyses. Indeed, the studies included in the present meta-analysis differed in terms of rate of transfusion, type of blood product, transfusion criteria, definition of ‘perioperative,’ methods of PS analysis, variables included in the PS model, sample size, follow-up and recruitment period, which may have caused the heterogeneity. However, it is impossible to perform subgroup analysis or meta-regression to meticulously explore potential sources of heterogeneity in the present meta-analysis due to the limited number of included studies and the limited information provided by these studies. Despite this, in the present meta-analysis, the results of sensitivity analyses were always consistent with those of the primary analyses, which proves the robustness of these findings. Fourth, only nine studies and one study reported data on DFS and DSS respectively, and most of these studies did not provide detailed diagnostic criteria for tumour recurrence. The lack of association between PABT and tumour recurrence may be partly due to the study limitations.

In conclusion, PABTs are not associated with recurrence but with worse OS in patients with GC and CRC undergoing radical surgery. The significant association between PABTs and worse OS may be attributable to unbalanced MPCs after PS adjustment. The findings of the present study are important to aid decision-making regarding transfusions when managing patients with GC and CRC during the perioperative period. However, considering the limitations in the present meta-analysis, these findings need to be confirmed.

Funding

This work was supported by the National Natural Science Foundation of China (No. 81602522).

Disclosure

The authors declare no conflict of interest.

Supplementary material

Supplementary material is available at BJS Open online.

Data availability

The template data collection forms, data extracted from included studies, and data used for all analyses are available from the corresponding author (H.J.) upon reasonable request.

Author contributions

Weilan Zhang (Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Resources, Software, Writing—original draft), Huimian Xu (Conceptualization, Supervision, Validation, Visualization, Writing—review & editing), Baojun Huang (Data curation, Formal analysis, Investigation, Methodology, Project administration, Software), Yan Xu (Data curation, Formal analysis, Investigation, Methodology, Project administration), and Jinyu Huang (Conceptualization, Funding acquisition, Supervision, Validation, Visualization, Writing—review & editing).