- Protocol and registration
This systematic review and meta-analysis were complied with the Preferred Reporting Items.
These items are especially made for Systematic Reviews and Meta-Analyses (PRISMA) guidelines for system reviews [23]. And a review protocol was registered with PROSPERO (Registration NO.CRD42018109341).
- Search strategy and eligibility criteria
Relevant publications were searched in Pubmed, the Cochrane Central Register of Controlled Trials, Scopus, and EMBASE databases up to September 30, 2018. The search terms included “intravenous Anesthesia ”, “inhalational Anesthesia”, “cancer surgery” and “survival”. In addition, we used the specific name of anesthetics to recognize additional studies, such as “propofol”, “sevoflurane”. All the articles were separately screened and identified by two review authors (L G and C D) to ensure sensitivity of the search strategy. Only English language and the full-text papers were eligible. And no time restrictions were applied to the research. Based on a full review of the retrieved articles, study eligibility was assessed according to the PICOS (Populations, Interventions, Comparisons, Outcomes, and Study designs) approach.
1). Populations (P): The sample size of the included studies is more than 50. The population was adult patients undergoing oncologic surgery.
2). Intervention (I) and Comparison (C): compared the long-term outcomes between patients using INHA and TIVA.
3). Outcomes (O): provided important information, including hazard ratios (HRs) and 95% confidence intervals (CIs) for overall survival (OS) or recurrence-free survival (RFS). All the non-adjusted, adjusted by multivariate Cox regression or propensity score (PS) data were obtained from each study. Some more data including the size of baseline samples, the total number of cases, the overall survival rate, and the confounders adjusted in final models, were available. The time of follow-up is at least 3 years.
4). Study design (S): designed as a case-control or cohort study or clinical trial.
Studies were excluded, if they were a protocol, meta-analysis, literature review, letter to the editor or abstract without full text. Studies with insufficient available data to estimate a HR with 95% CIs were removed. Studies with an English abstract but not with an English full text were neither considered.
- Data Extraction and quality assessment
All data and items helpful to assess the validity of the studies were independently extracted by two of the reviewers. Any different opinions between the reviewers were solved through repeated discussion, to finally reach a consensus. To remove any potential systematic biases, other two authors did a comparable study by independently scanning all the eligible publications. After this process we agreed on all the variables assessed.
The following variables were extracted from each paper where the following information was available: authors’ surname, publication year, country or district, sample size before and after propensity score matching of both groups, design type, anesthetic agents, outcomes, adverse effect, median follow-up, survival rate, type of cancer surgery, and hazard ratios (HR) with 95% confidence intervals (CI). An adverse effect study was defined as being the worse long-term survival after surgery in INHA group compared with TIVA. If different groups were divided into by volatile anesthetics as shown in Kim’s study [17], all the relevant intervention groups of the study were combined as a single treatment group. Consequently, both the sample sizes and the quantity of people with intervention events were respectively summed across groups.
The time-to-event data was retrieved from Kaplan-Meier curves, log-rank statistics or multivariate Cox regression analysis results of the individual trials. It was summarized in terms of the HR in logarithm scale as a function of its variance. Non-adjusted (by univariate analysis), multivariable-adjusted, multivariable and PS-adjusted estimates (HR and 95% CI) for OS and RFS in all the studies were separately extracted.
When the outcome data for OS or RFS was unavailable, the Excel spread sheet freely provided by Tierney et al. and the software Engauge Digitizer, version 4.1 (downloaded from http://sourceforge.net), were used to regenerate data points from the Kaplan-Meier curves. Authors of the selected published articles were contacted by email whenever we found that data indispensable for the meta-analysis was not explicitly provided. If the HR and 95% CI of TIVA to INHA for OS and RFS were reported, it would be converted into the HR and 95% CI of INHA to TIVA.
Quality of the included studies was separately assessed by the Newcastle-Ottawa Score (NOS). Low quality was defined as the score less than 8.
- Statistical Analysis
Both Cochran chi-square Q statistics and I-square statistics were used to assess the heterogeneity of the included studies. Heterogeneity was considered as either a P value≤0.10 [26]. It also determined the appropriate use of random-effects (DerSimonian and Laird method) model. And when the I-square statistic was used to measure the consistency of effects across studies, 20-50%, 50-75%, and >75% were considered low, moderate, and high heterogeneity, respectively. Potential predictors of heterogeneity were further explored by sub-group analysis.
Sensitivity analysis was explored through the different types of estimate (unadjusted, multivariable analysis-adjusted or both multivariable and PS-adjusted). Meta-regression analyses were explored the effect of study quality on the results.
The potential publication bias of this meta-analysis was judged visually in a funnel plot of ln [HR] against its standard error (SE ln [HR]), and the degree of asymmetry was examined by the Egger’s and Begg’s test if the included study number is not less than 8. We considered a significant publication bias if a p value <0.1.We further used the trim and fill computation to estimate the effect of publication bias on the explanation of results.
Subgroup analyses were respectively performed on high-quality versus low-quality studies (NOS variable), and Non-Asian versus Asian studies. The pooled HR estimates derived from the two separate analyses were compared with a test of interaction.
Time-to-event outcomes were appropriately analysed and compared graphically using the hazard ratio of non-adjusted effect sizes, adjusted by multivariable analysis, or both multivariable and PS-adjusted. The log HR and its variance were pooled using an inverse variance weighted average, and the results were presented as an HR and 95% CI. All of the statistical analyses in this meta-analysis were conducted with Stata/SE version 14 (Stata Corporation, College Station, TX 77845), using the commands metan, metareg and metabias.