Background
The incidence of postoperative complications after radical gastrectomy remains high [
1‐
4], and the estimated incidence is 12.8 to 14% [
5‐
7]. In addition to undermining the short-term survival, postoperative complications may also be correlated with long term prognosis. Currently, increasing numbers of observational studies have investigated the correlation between postoperative complications and long-term prognosis after radical gastrectomy. Although some reports have negative findings [
8‐
12], other studies have demonstrated that overall postoperative complications, infectious complications, and gastrointestinal leakages are all correlated with poor overall survival (OS) and/or recurrence-free survival (RFS) [
13‐
23]. Additionally, the correlations between postoperative complications and long-term prognosis in different stages are controversial and are based on subgroup analyses with small sample sizes [
13,
18‐
20].
Given the prevalence of postoperative complications after radical gastrectomy, it is important to determine whether a correlation exists between postoperative complications and poor prognosis. The existence of that correlation may not only lead to a consideration of shortening follow-up interval and enforcing adjuvant chemotherapy in patient who have developed postoperative complications, but may also underline the necessity of neoadjuvant chemotherapy and stress control management in patients with high risk of developing postoperative complications to reduce the hazard for long term prognosis [
9,
11,
21]. In the meta-analysis, the correlations between postoperative complications and prognosis after radical gastrectomy were assessed.
Methods
Search strategy and eligibility criteria
The PubMed, EMBASE, and Cochrane Library databases were searched from inception until February 24, 2019, for studies that assessed the relationship between postoperative complications and prognosis after radical gastrectomy. The following medical subject heading (MeSH) terms and keywords were used: “Stomach Neoplasms”, “Gastrectomy”, “Postoperative Complications”, and “Prognosis”. The search was restricted to studies on humans and to those that were published in the English language. The titles and abstracts were screened by two authors independently. The inclusion criterion was as follows: any study that compared the long-term prognosis between patients with and without postoperative complications after radical gastrectomy for gastric cancer. The exclusion criteria were as follows: (1) data of other neoplasms other than gastric cancer were included in the survival analysis; (2) data of palliative surgery were included in the survival analysis; (3) studies that describe the same patient population; (4) hazard ratio (HR) cannot be estimated; (5) describing complications without precise definitions; (6) letters, comments, or conference abstracts. When multiple studies describing the same patient population were identified, the most recent publication was used unless additional data were provided in the earlier work.
The following data were extracted: first author, year of publication, study design, number of subjects, adjuvant chemotherapy, tumor stage, types of complications, incidences of complications, HR of any postoperative complications, and 5-year OS and 5-year RFS for patients with and without postoperative complications, as well as whether in-hospital deaths were excluded in the survival analysis. Unreported data were requested through e-mail from corresponding authors of the included studies. If there was no response to the e-mails, the missing data were estimated from the figures in the published literatures using Engauge Digitizer 4.1 (Mark Mitchell, Baurzhan Muftakhidinov, and Tobias Winchen et al., “Engauge Digitizer Software.” Webpage:
http://markummitchell.github.io/engauge-digitizer) and the HRs were estimated using the method of Tierney et al. [
24].
Study quality assessment
The methodological quality of each observational study was assessed by the Newcastle-Ottawa Scale (NOS, ranging 0–9) [
25]. In brief, each study was assessed for the following aspects: selection, comparability, and outcome or exposure. The comparability was primarily assessed for pathological stage and was also assessed for aspects of adjuvant chemotherapy and in-hospital death disposition in the survival analysis.
Statistical analysis
Statistical analysis was performed with RevMan (version 5.3.5.; Cochrane Collaboration). HRs and their 95% confidence intervals (CIs) were used to evaluate the association between postoperative complications and prognosis (OS and/or RFS). Subgroup analyses were performed to investigate the correlations between infectious complication, gastrointestinal leakage, and prognosis. Furthermore, correlations were investigated for each pathological stage when possible. Statistical heterogeneities among studies were assessed by the I2 statistic. The random effects model and the fixed effects model were used. If I2 was less than 40% (cutoff point), we used the fixed effect model, while if I2 was more than 40%, the random effects model was chosen. Sensitivity analysis, in which one study was removed at a time, was performed to evaluate the stability of the results. Descriptive techniques were used when clinical heterogeneity existed or when no data could be used in the pooling analysis. The assessment of publication bias was evaluated using the funnel plot.
We followed both the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Statement [
26], and the guidelines for Meta-analysis of Observational Studies in Epidemiology (MOOSE) in reporting this study [
27]. All analyses were based on previously published studies, thus no ethical approval and patient consent are required.
Discussion
The present study undertook a comprehensive review and meta-analysis of the literatures to assess the relationship between postoperative complications and patient prognosis. The results demonstrated that, although the correlation was not found by several studies, the pooled results showed that postoperative complications correlated with poor prognosis.
Several reasons may contribute to the divergences. First, the negative findings in some studies may be ascribed to the interfered application of adjuvant chemotherapy. Jin et al. demonstrated lower proportion of adjuvant chemotherapy in the complication group (47% vs. 61%), and the combination of postoperative complications and receiving no adjuvant therapy significantly increased the hazard of death and recurrence. Furthermore, decreased OS and RFS were not observed in patients who experienced complications but received adjuvant therapy [
21]. Another study demonstrated that the adjuvant chemotherapy was postponed in patients with intra-abdominal complications (55.3 ± 34.7 vs. 26.6 ± 11.9 days) [
22], and the postponed chemotherapy is correlated with poorer survival in patients with gastric cancer [
28]. Second, the application of prophylactic neoadjuvant chemotherapy may abolish the poor prognosis induced by postoperative complications. In a cohort with 101 patients who underwent curative gastrectomy after receiving neoadjuvant chemotherapy, Eto et al. demonstrated a comparable RFS between patients with and without postoperative complications, and the 5-year RFS was 41.7% and 43.9%, respectively [
11]. Third, the varied perioperative stress level may be an additional reason for the negative finding. Saito et al. demonstrated that the postoperative inflammation degree (reflected by the CRP level), rather than the postoperative complication itself, is related to the recurrence and poor prognosis [
9]. Besides, Watanabe et al. also demonstrated comparable prognosis between patients with and without postoperative complications [
12]. Their patients underwent total gastrectomy with splenectomy for the treatment of proximal advanced gastric cancer. The extensive resection might lead to an excessive surgical stress in both groups and that may lead to a deteriorated prognosis in patients without postoperative complications [
29].
Accordingly, the results of the present study may highlight the importance of both adjuvant and neoadjuvant chemotherapy in patients with postoperative complications or with a high risk of developing postoperative complications. The results of the present study may have also highlighted the stress control management during the perioperative period. However, whether a decreased stress level will result to an improved prognosis remains to be determined. Additionally, any other methods that decrease the postoperative complications may also indirectly improve the prognosis. The intraoperative manipulation, such as the choice of reconstruction or the less invasive approach, may play a role in decreasing the postoperative complications and thereby improve the prognosis indirectly. For instance, recent studies demonstrated that BI reconstruction method significantly reduced the postoperative complications after laparoscopic distal gastrectomy [
30,
31]. Therefore, patients may benefit more from that approach with low risk of postoperative complications.
In the analysis of the relationship between infectious complications or gastrointestinal leakages and OS, the study from Kim et al. demonstrated high heterogeneity. Kim et al. found that gastrointestinal leakage was not associated with decreased survival. There are some possible reasons for the negative results [
8]. First, the effect of leakage may be diluted by the effect of other complications occurred in the control group. That is to say, other complications other than gastrointestinal leakage may also contribute to the poor prognosis and that may cause an underestimated effect of leakage on prognosis. Second, the sample size may not be adequate to detect the significant correlation because their Kaplan-Meier curve demonstrated a trend of poor OS in the leakage group (
p = 0.076) [
8].
The present study had some limitations. First, five of the included studies did not exclude in-hospital death in the survival analysis [
15‐
18,
22]. It is well acknowledged that in-hospital mortality would be higher in patients with postoperative complications and would decrease the OS accordingly. Therefore, a subgroup analysis with the eight reports that excluded in-hospital death or have no in-hospital death was performed and a similar result was found (HR 1.40, 95% CI 1.06–1.84). Second, more preoperative comorbidity, a higher ASA or ECOG score, and older age were frequently observed in the complication group, as shown in Additional file
1: Table S1, and such characteristics are correlated with a shorter life expectancy after surgery. As a result, we analyzed the data from seven studies that reported RFS [
9‐
12,
18,
20,
21], and the HR demonstrated a positive correlation between postoperative complications and reduced RFS (HR 1. 28, 95% CI 1.10–1.49). The correlation between postoperative complications and poor RFS still exist after the in-hospital mortality were excluded (HR 1.33, 95% CI 1.09–1.63). Third, patients in the complication group frequently had more advanced disease. Eleven of the studies demonstrated the proportion of each stage, and six of the studies reported comparable stages between the two groups [
10‐
12,
15,
22,
23]. Such a bias may cause an overestimated correlation of postoperative complications with long-term prognosis. To avoid the influence of unbalanced tumor stages, the correlations between complications and prognosis were analyzed in separate stages based on the data from four studies [
13,
18‐
20]. In addition to the correlation between postoperative complications and decreased OS and RFS in stage II and III patients, attention should be paid to stage I patients with postoperative complications because of the undetermined result (Fig.
5). If such a correlation did exist, the application of adjuvant chemotherapy might be expanded to stage I patients who have developed postoperative complications. However, a limited number of studies were included in the subgroup analysis of separated pathological stages and the confounders cannot be avoided in the subgroup analysis. More solid evidence from studies with larger sample sizes is warranted, and RFS analysis should also be considered in further studies.