Impact of postdiagnostic statin use on ovarian cancer mortality: A systematic review and meta-analysis of observational studies

Data sources and searches

The present systematic review and meta-analysis was performed in accordance with the recommendations of the Meta-Analysis of Observational Studies in Epidemiology group 23. Two individuals independently performed an electronic search of the Medline (https://medlineplus.gov/), Embase (https://www.elsevier.com/solutions/embase) and Web of Science (https://webofknowledge.com/) databases up to 31 December 2017, without language restrictions. The following search keywords and terms were used: ((statin^*) or ((“hydroxymethylglutaryl-CoA reductase”) or (“HMGCoA reductase”) and (inhibitor^*)) or (anticholesteremic) or (simvastatin) or (rosuvastatin^*) or (pravastatin^*) or (atorvastatin^*) or (fluvastatin^*) or (cerivastatin^*) or (pitavastatin^*) or (lovastatin^*)) and ((ovary) or (ovarian)) and ((cancer) or (neoplasms) or (carcinoma) or (tumor)) and ((prognosis) or (outcome) or (survival) or (mortality) or (recurrence) or (progression) or (metastasis)). In addition, the reference lists of published narrative reviews were scanned to identify potential additional publications not retrieved by the manual search 24, 25. In an effort to identify unpublished studies, the abstracts of the annual meetings up to 31 December 2017 of the related associations were also manually searched.

Study selection

NoteExpress Research & Reference Manager software (version 3.0; Aegean Software, Beijing, China) was used to identify and remove duplicate records. Subsequently, two individuals independently scanned each title and abstract. The full text of all studies that advanced beyond this stage were independently reviewed. Once a consensus was reached, the studies that fulfilled the following criteria were included for the meta-analysis: (i) a cohort study or randomized controlled trial; (ii) assessed the relationship of postdiagnostic statin use with survival/mortality of OC patients; and (iii) included hazard ratios (HRs) or relative risk analyses with 95% confidence intervals (CIs), or provided data allowing the calculation of the risk estimates and 95% CIs. Ecological studies, case–control studies, reviews without original data, editorials, commentaries, meeting abstracts and case reports, as well as studies that investigated the association between prediagnostic statin use with survival/mortality of OC patients, and those that reported risk estimates without 95% CIs (e.g. studies that could not be included in the statistical summary) were excluded from analysis.

Data abstraction

Data were extracted independently by two individuals and inconsistencies were resolved through discussion. Data regarding study characteristics (first author, study year, country of origin, design of study and length of follow-up), patient characteristics (number of cases and events, cancer stage and grade), category of exposures and outcomes, and adjustment for confounders [age at diagnosis, International Federation of Gynecology and Obstetrics (FIGO) stage, grade, comorbidity, performance status, residual disease, chemotherapy, and nonstatin drug use] were collected. The primary outcome was OS/all-cause mortality (ACM), although ACM is reported for simplicity. The secondary outcomes were cancer-specific mortality and recurrence-free survival (RFS).

Risk of bias assessment of individual studies

To assess the risk of bias of the included studies, two individuals independently used the Newcastle–Ottawa quality assessment scale for cohort studies 26-30 relying only on the information presented in the reports and making no assumptions. Any disagreements were referred to a third team member. The Newcastle–Ottawa quality assessment scale uses the three quality parameters of selection, comparability, and exposure/outcome assessment. Subsequently, studies that achieved a full rating in at least two categories of these three assessments were considered to have a low risk of bias 28, 31.

Statistical analysis

For one study 21 that did not use the category of never use as a reference to statin use, the effective-count method proposed by Hamling et al. 32 was used to recalculate the HRs and 95% CIs. Overall summary estimates were calculated using inverse variance-weighted random effects meta-analysis. Individual HR and summary estimates are displayed graphically as forest plots. Heterogeneity across the studies was quantified using the I² statistic, which estimates the proportion of variability in the meta-analysis caused by differences between studies rather than sampling error 33. The cut-off points of <30%, 30–50% and >50% were used to indicate low, moderate and substantial levels of heterogeneity. Furthermore, the sequential exclusion strategy proposed by Patsopoulos et al. 34 was used to examine whether the overall estimates were influenced by the substantial heterogeneity observed.

Based on a priori decisions, subgroup analyses for OS/ACM was conducted according to the geographical location (Europe, America and Asia), study quality (low vs. high risk), median number of OC cases (≥ 1000 vs. < 1000), type of cohort (retrospective vs. prospective), FIGO stage (all vs. III–IV), histological type (serous, mucinous, endemetrioid and clear cell), source of exposure (prescription database vs. medical records), statistical analysis (time-dependent vs. not time-dependent), and adjustments made for potential confounders (including age at diagnosis, FIGO stage, grade, comorbidity, residual disease, chemotherapy and nonstatins drug use).

The possibility of small study biases (e.g. publication bias) was explored visually, checking for asymmetry in contoured funnel plots of treatment effects against individual standard errors using formal statistical procedures. Specifically, the Begg et al. 35 and Egger et al. 36 methods were applied to examine possible effects of the inclusion of small studies. The trim-and-fill method was used to reduce the potential influence of publication bias 37.

A series of sensitivity analyses were performed to better understand the sources of statistical heterogeneity between studies, as well as to test the robustness of the findings. First, to examine the effect of individual studies on the summary estimates, influence analysis was conducted, in which the pooled estimates were recalculated by omitting one study at a time. Second, meta-regression analysis was used to assess differences between subgroups. A P value of <0.05 was considered statistically significant. All statistical analyses were performed using Stata 12.0 software (Stata LLC, College Station, TX, USA).

Nomenclature of targets and ligands

Key protein targets and ligands in this article are hyperlinked to corresponding entries in http://www.guidetopharmacology.org, the common portal for data from the IUPHAR/BPS Guide to PHARMACOLOGY 38, and are permanently archived in the Concise Guide to PHARMACOLOGY 2017/18 39.

Search results, study characteristics and quality assessment

The search strategy generated 269 articles, of which 12 were considered of potential value, thus the full texts of these articles were retrieved for detailed evaluation (Figure 1). Four 40-43 of these 12 articles were subsequently excluded for various reasons and a review of the reference lists led to the inclusion of one additional article. Therefore, eight studies 15-22 were included in the present meta-analysis.

Table 1 shows the key characteristics of the included studies, which were published from 2008 to 2017 and included a total of 19 904 OC patients with a range of 60–8159 cases per study. The proportion of patients on statins ranged from 11% to 42.6%. Three of the included studies were conducted in Asia 15, 21, 22, three in Europe 16, 18, 20 and two in America 17, 19. Most of the cohort studies were retrospective 15, 17-19, 21, 22, while two were prospective 16, 20. The median duration of follow-up varied across studies from 28.8 to 54 months and the rate of statin use ranged from 11% to 42.6%. Table 2 demonstrates the adjustment for confounders in the primary analysis of these included studies, all of which were adjusted for age at diagnosis. The majority of studies were adjusted for FIGO stage (n = 7), comorbidity (n = 5), and chemotherapy (n = 5). However, few studies were adjusted for tumour grade (n = 2), residual disease (n = 2) nonstatin drug use (n = 2). Of note, none of the included studies were adjusted for performance status (n = 0).

Study quality scores are summarized in Table 3. Six studies 15, 16, 18-20, 22 were graded as low risk and two 17, 21 as high risk. The most common selection bias was related to the representativeness of the exposed cohort and selection of the unexposed cohort, the most comparability bias was the control for important factor or additional factor, and the most common outcome bias was a sufficient duration of follow-up for outcomes to occur (Table 3).

Postdiagnostic statin use and ACM of OC patients (user vs. nonuser)

Seven studies 15, 17-22 involving 11 745 subjects with OC were used to estimate the summary effects of postdiagnostic statin use on ACM. The summarized HR was 0.74 (95%CI = 0.63–0.87), with moderate heterogeneity (I² = 55%, P = 0.038) (Figure 2). Publication bias was not detected by the Begg's (P = 0.23) and Egger's tests (P = 0.16), and visual inspection of the funnel plot indicated symmetry.

When one study 19 that contributed the largest amount to heterogeneity until I² was <50% was sequentially excluded, the summarized outcomes (HR = 0.82, 95%CI = 0.76–0.90, I² = 45.6%) were similar to the main results. Additionally, sensitivity analyses showed that no study was influential, when one study at a time was omitted. A pooled HR was calculated for the remainder of the studies (Figure 3). The estimated HR in this sensitivity analysis ranged from 0.78 (95%CI = 0.70–0.87, I² = 61.8%) to 0.82 (95%CI = 0.76–0.90, I² = 45.6%).

Table 4 presents the results of subgroup analyses. Notably, the majority of these findings were statistically significant as well as consistent with the main result. More specifically, significant results of subgroup analysis stratified by histology were only observed in patients with serous OC. Meta-regression analyses in consideration of the potential variables mentioned showed no effect of modifiers.

Intensity of postdiagnostic statin use and ACM of OC patients (highest vs. lowest)

For three studies 18-20 involving 11 266 OC cases that investigated the intensity of postdiagnostic statin use and ACM, the summarized HR was 0.84 (95%CI = 0.74–0.96) without heterogeneity (I² = 0%, P = 0.444) (Figure 4).

Postdiagnostic statin use and secondary outcomes of OC patients (user vs. nonuser)

Three studies 15, 17-22 were used to estimate the summary effects of postdiagnostic statin use on cancer-specific mortality and one 20 was used to estimate RFS. The summarized HR was 0.87 (95%CI = 0.80–0.95, I² = 0%, P = 0.411) for cancer-specific mortality and 0.66 (95%CI = 0.40–1.08) for RFS (Figure 5).