Criteria for considering studies for this review
Types of studies
We do not expect to identify any randomised controlled trials (RCTs) in this area. Instead, we will consider prospective and retrospective cohort studies and case-control studies. Case series, case reports and in vitro and animal studies will be excluded.
Types of participants
Women 18 years of age or older, with existing endometrium/uterine body. Women with preexisting cancer diagnoses of any type will be excluded, along with women who have undergone fertility preservation treatment after receiving a cancer diagnosis.
Types of interventions
Any of the following regimens, offered alone or in combination, will be considered as the exposure: clomiphene citrate (CC), gonadotrophins, hCG and GnRH agonists/antagonists.
Outcomes in subfertile women treated with these agents will be compared with those of subfertile women who received no intervention and with those of control groups of women who had no fertility problems.
Types of outcome measures
Incidence of endometrial (uterine) cancer, clinically or histologically confirmed at any time following treatment for subfertility.
Incidence of endometrial hyperplasia (complex, simple atypical and complex atypical).
Search methods for identification of studies
We will search CENTRAL (current issue), MEDLINE via Ovid (1960 to date) and EMBASE via Ovid (1980 to date). We will search the CENTRAL database for reasons of completeness because, although this review will be based on non-randomised studies (NRSs), CENTRAL contains controlled clinical trials (CCTs), interrupted time series and controlled before and after series, in addition to RCTs.
The search terms will include a combination of thesaurus-based and free-text terms. See Appendix 1 for the MEDLINE search strategy, which will be adapted accordingly for searches of the other databases.
We impose no restriction on language and publication status. All databases will be searched from 1960 onwards, as the interventions sought were not available before that date.
Searching other resources
Reference lists of included studies and any relevant systematic reviews identified will also be searched to identify eligible studies for inclusion. The review authors will try to identify the relevant grey literature by looking at the following.
'Open grey', a system for grey literature produced in Europe, such as research reports, doctoral dissertations and conference papers (http://www.opengrey.eu/).
ProQuest dissertation and thesis databases (http://www.proquest.com/en-US/catalogs/databases/detail/pqdt.shtml).
Published or ongoing trials in the trial registers for ongoing and registered trials: 'ClinicalTrials.gov', a service of the US National Institutes of Health (http://clinicaltrials.gov/ct2/home) and http://www.controlled-trials.com, as well as the World Health Organization International Trials Registry Platform search portal (http://www.who.int/trialsearch/Default.aspx) and Physicians Data Query (http://www.nci.nih.gov).
Conference proceedings and abstracts through ZETOC (http://zetoc.mimas.ac.uk) and WorldCat Dissertations.
Reports of conferences in the following: Gynecologic Oncology (Annual Meeting of the American Society of Gynecologic Oncologists), International Journal of Gynecological Cancer (Annual Meeting of the International Gynecologic Cancer Society), British Journal of Cancer, British Cancer Research Meeting, Annual Meeting of the European Society of Medical Oncology (ESMO) and Annual Meeting of the American Society of Clinical Oncology (ASCO).
Personal communication with experts in the field who are/have been conducting/have led research in the field and on the specific hypothesis of this review.
Data collection and analysis
Selection of studies
Search results will be transferred to a special processing platform developed by P. Kanavidis, a full description of which can be found in our recent publication (Siristatidis 2013).
Once duplicates have been removed, all review coauthors will be involved in selecting studies for eligibility. Review coauthors working in pairs will assess an allocation of titles and abstracts, so that each allocation portion will be independently assessed by two review coauthors. We will not be blinded to authors’ names and institutions, journal of publication or study results while assessing studies for potential inclusion.
Studies that clearly do not meet the inclusion criteria will be excluded. For 'potentially relevant' studies, the full text will be obtained for further assessment. Letters will be sent to study authors to ask for clarification about 'potentially relevant' studies. All disagreements will be resolved by consensus.
Data extraction and management
The review authors will extract data using a pre-designed Excel file before copying the data into Review Manager 2011 (RevMan 5) for analysis. The data extraction form (Appendix 2) was piloted previously and was subsequently used in our recently published report (Siristatidis 2013), which, however, focused solely on IVF. Authors will extract the data independently while working in pairs. Disagreements will be resolved by team consensus.
Data to be extracted include general information (title, author, year, journal, geographical setting and clinical setting), study characteristics (study period, number of participants per exposed/unexposed or case/control group, design, follow-up, ascertainment of exposure and outcome and matching factors), participant characteristics (inclusion/exclusion criteria, age, race, gynaecological and reproductive history, definition and causes of subfertility, gravidity, parity and histological subtype of cancer), interventions (type and agent of fertility treatment, dosage of treatment, number of treatment cycles, age at first use, years since first use, reference population for the comparison and general population or subfertile women) and risk of bias assessment data (cf. below, relevant sections).
In addition, the following results will be extracted, when available.
Maximally adjusted odds ratio (OR) and associated 95% confidence interval (CI), as defined by the study authors.
Maximally adjusted relative risk (RR) and associated 95% CI, as defined by the study authors.
Maximally adjusted hazard ratio (HR) and associated 95% CI, based on the number of events (cases) and with consideration of the time to event.
Standardised incidence ratio (SIR) and associated 95% CI, estimated as the ratio of observed over expected number of cases for the exposed group of women.
Incidence rate ratio (IRR) and associated 95% CI, estimated from the number of cases per person-years for exposed and unexposed women.
Associated raw data for recalculation (data checking) or de novo estimation of missing measures.
Assessment of risk of bias in included studies
No RCTs are anticipated, hence assessment of risk of bias will pertain exclusively to non-randomised studies (NRSs).
An extension to the risk of bias tool for randomised studies is currently under development by a multi-disciplinary working group of experts, with the ultimate aim of creating a risk of bias tool that can be used in systematic reviews of non-randomised studies. This work has advanced enough for the NRS Working Group to consider selecting appropriate systematic reviews to pilot the tool. Our review seems to provide an ideal piloting platform, and the NRS Working Group is considering it as a candidate. This review will be performed with the input of the NRS Working Group; should a timely piloting version of the tool become available, we will use it to assess the risk of bias in included studies. We will not, however, delay completion of this review until developmental work on the tool is completed.
If the tool is not available for pilot use in time, the risk of bias will be assessed in accordance with relevant sections of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011), as well as in keeping with the rationale adopted in the most recent Cochrane review examining the association between ovarian cancer and ovary-stimulation drugs for subfertility (Rizzuto 2013). As suggested by the Cochrane Handbook for Systematic Reviews of Interventions (Section 22.214.171.124), items included in the Newcastle-Ottawa scale (Wells 2011) will be customised for inclusion in the detailed item-to-item list below.
The assessment of risk of bias will encompass the examination of selection bias (comparability of groups and confounding/adjustment), performance bias, detection bias, attrition bias and reporting bias. The qualifications 'low risk', 'high risk' and 'unclear risk' will be adopted for each risk of bias domain, in accordance with the guidelines published by the Newcastle-Ottawa scale (Wells 2011). The 'critical risk category' in the risk of bias tool will also be taken into account.
The following features of the study design will be assessed for selection bias.
Comparability of groups
Consecutive recruitment cases (case-control studies).
Population-based controls derived from the same population as cases (case-control studies).
Non-exposed women drawn from the same population as the exposed cohort (cohort studies).
For all studies, the following factors will be evaluated as potential confounders, given that they represent known risk factors for endometrial cancer (Adami 2008).
As mandated by the Cochrane Handbook for Systematic Reviews of Interventions, an additional table will be created that will list the prestated confounders as columns and the studies as rows. This additional table will indicate whether each study:
restricted participant selection so that all groups had the same value for the confounder;
demonstrated balance between groups for the confounder;
matched on the confounder; or
adjusted for the confounder in statistical analyses to quantify the effect size.
The following features of the study design will be evaluated.
Exposure to ovary-stimulation drugs was ascertained by a secure source, namely, medical records or structured interviews (all studies).
In cases in which a structured interview was performed, interviewers assessing exposure to fertility treatment were blinded to the presence of endometrial cancer (all studies).
The same method was used to ascertain exposure to fertility drugs for both cases and controls (case-control studies).
The following feature will be assessed.
With respect to attrition bias, the following features will be examined.
Length of follow-up was at least 10 years for the exposed group (Siristatidis 2013), as endometrial cancer reaches its peak incidence after the age of 55 years, whereas IVF exposure occurs mostly during the late reproductive years (cohort studies).
At least 80% of women in all groups were included in the final analysis (all studies).
Measures of treatment effect
Both primary and secondary outcome measures will be expressed as odds ratios (ORs), relative risks (RRs), hazard ratios (HRs), standardised incidence ratios (SIRs) or incidence rate ratios (IRRs). Ideally, separate analyses by type of effect estimate will be performed. However, should only a small number of studies be available, we will attempt to transform ORs, RRs and HRs into a single metric to reduce heterogeneity and to provide more robust estimates. SIRs and IRRs will be analysed separately. The 95% confidence interval (CI) for log(SIR) will be reconstructed via the term ± 1.96/(square root (O)), where 'O' represents the observed number of events (Alder 2006).
Unit of analysis issues
None expected. The unit of analysis will always be the participant.
Dealing with missing data
The corresponding authors of 'potentially relevant' and eligible studies will be contacted by email when the need arises to obtain potentially useable missing data, to ask for additional information or to request methodological clarification.
Assessment of heterogeneity
It is generally expected that non-randomised studies will be more heterogeneous than randomised studies (Cochrane Handbook for Systematic Reviews of Interventions, Section 13.6.1 (Higgins 2011)), hence heterogeneity tolerance levels will be adjusted accordingly. Inconsistency among studies will be quantified by estimating I2 (Higgins 2011). When considerable heterogeneity is noted (I2 > 80%), the pooled estimate will be suppressed in the forest plot, and results will be reported as narrative text or in descriptive tables. For levels of I2 between 50% and 80%, heterogeneity will be considered as moderate, and pooled analysis will be attempted by using a random-effects model to allow for heterogeneity. Heterogeneity will also be explored by means of a priori agreed subgroup analyses.
Assessment of reporting biases
If more than 10 studies are available, publication bias will be assessed using Egger’s formal statistical test (Egger 1997) at the 90% level, and a funnel plot will be constructed. An Egger's modified test (Harbord's test) will be used to assess possible small-study effect biases (Harbord 2006).
To our knowledge, no RCTs will be available for inclusion in the analysis, and all included studies are expected to be case-control or cohort studies. If RCTs are found and included, meta-analyses will be carried out separately for RCTs in accordance with the intention-to-treat principle and for any non-randomised studies (cohort and case-control studies).
Random-effects models will be used to calculate separate pooled effect estimates for each of the risk ratio (RR) measures (OR, HR, SIR, IRR) (DerSimonian 1986). As the absolute risk for cancer is rather small, the four relative measures are expected to yield similar estimates (Adami 2008; Larsson 2007). Yet, as described earlier, if feasible, separate analyses by type of effect estimate will be performed. Should only a small number of studies be available, we will attempt to transform ORs, RRs and HRs into a single metric to reduce heterogeneity and to provide more robust estimates. SIRs and IRRs will be analysed separately. Data permitting, the proposed data synthesis strategy will be supported by sensitivity analysis by type of measure. Separate analyses will be carried out for the different reference populations available (general population of women and subfertile women who did not receive fertility treatment).
A comparison of fixed-effect and random-effects summary effects would be used to explore small-study effect biases.
Subgroup analysis and investigation of heterogeneity
Should data be available, the various therapeutic agents will be evaluated by drug subtype (SERMs, gonadotrophins, GnRH agonists and antagonists, hCG) and as individual drugs. If sufficient studies are available, subgroup analyses will be performed for each ovary-stimulation agent by:
type of effect measures;
gravidity and/or parity;
causes of subfertility;
histological type of cancer;
dosage and/or number of cycles; and
studies including or excluding events during the first year of follow-up (Siristatidis 2013).
Although separate subgroup analyses will be attempted, it is well known that some of these factors may interact with each other; therefore, inferences will not be based solely on subgroup analyses, but meta-regression analysis will be additionally performed to adjust for mutual confounding.
As mentioned earlier (under Data synthesis) sensitivity analysis for each type of effect measure will be employed if sufficient numbers of studies are available. Sensitivity analyses based on the risk of bias assessment will also be carried out for the selection, performance and attrition bias domains.