Data sources and searches
The results of this study are reported according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [17]. We systematically searched PubMed, Embase and Web of Science databases for available literature up to June 30, 2019, without language restrictions. References were assessed for additional articles overlooked during the primary search [18, 19]. We employed a search strategy with the aid of a biomedical information specialist. The following search keywords and terms were used and appropriately translated for the other databases: ((nutritional index) or (albumin) or (lymphocyte)) and ((ovary) or (ovarian)) and ((neoplasms) or (tumour) or (tumor) or (cancer) or (carcinoma)). Searches were undertaken and two reviewers (T-TG and HS) independently assessed the literature for study inclusion.
Study selection
Articles for further assessment were independently selected by two reviewers (T-TG and Q-JW). Articles were included or excluded based on predefined selection criteria before beginning our search. Decisions were based on consensus. Authors were contacted via email in cases where clarification was required. Duplicate reports were removed both automatically and manually using a reference management library. Studies were eligible if they (1) were cohort or randomized controlled trials, (2) defined exposure as a preoperative PNI of 10 × serum albumin value (g/dl) + 0.005 × lymphocyte count (per mm3) in peripheral blood for OC patients, (3) defined outcome as progression-free survival (PFS) or OS of OC, (4) provided appropriate risk estimates (i.e., relative risk or hazard ratio (HR)) of the association between preoperative PNI and survival of OC (if multiple estimates were available, we extracted the estimate that adjusted for the most covariates). Studies were excluded if they (1) were published as letters, editorials, reviews, notes, commentaries, meeting abstracts, case reports, case-control analyses and conducted in animals and (2) reported risk estimates without 95% confidence intervals (CI).
Data abstraction and quality assessment
Data were independently extracted in duplicate by two reviewers using standardized forms (T-TG and J-YZ). Disagreements were resolved by consensus. From each study, the following information was extracted: first author name, publication year, country, patient characteristics, category of exposures and outcomes, and adjustment for confounders.
Quality assessment of the included studies was performed using the Newcastle-Ottawa Scale for observational studies [20] that consists of eight items grouped into three domains (selection, comparability, and outcome). A maximum of nine stars was awarded to any individual study. Studies that achieved a full rating in at least two categories of the three assessments were considered to have low risk of bias [21, 22].
Statistical analysis
For one study that did not use OC patients with high preoperative PNI as a reference group[10], the effective count method proposed by Hamling et al. [23] was applied to recalculate HR and 95% CI. Risk estimates were summarized using a random-effects model, since differences in populations and settings between studies could not easily justify a common effect size. Potential heterogeneity in results across studies was examined using I2 statistics [24]. Cut-off points ≤ 50%, 51–75% and ³ 76% were used to indicate low, moderate and substantial heterogeneity, respectively. The potential for small-study effects, such as publication bias, was assessed using Funnel plot, Egger’s linear regression [25] and Begg’s rank correlation [26] methods. A probability (P) value < 0.05 for the two tests was considered representative of significant publication bias. All statistical analyses were performed using Stata 12.0 software (Stata LLC, College Station, TX, USA).