The protocol was developed and described following the PRISMA-P checklist (Additional file 1) (20, 21). The systematic review and meta-analysis will be reported following Meta-analysis of Observational Studies in Epidemiology (MOOSE) (22) for observational studies and the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) (23) statement for experimental studies. Following the guidelines, our systematic review protocol was registered on Open Science Framework (https://osf.io/frg5z) and was registered with the International Prospective Register of Systematic Reviews (PROSPERO) on March 29, 2021 (registration number CRD42021239348).
Inclusion and Exclusion Criteria
As a systematic review of epidemiological studies, eligibility criteria specified the study design, participants/population, intervention (or exposure for observational studies), comparators, and outcomes.
Study design: We will include original observational and experimental studies that evaluate the associations between diabetes medications and cancer (breast, lung, prostate, colorectal, pancreatic, and liver cancers). Cross-sectional designs will not be considered due to limited ability to confirm temporality. Retrospective and prospective cohort studies, case-control studies within a pre-existing cohort (i.e., nested case-control and case-cohort studies), and randomized controlled trials (RCTs) will be included in this review.
Participants/population: We will include male and female adult participants for cancer of shared anatomic sites (lung, colon and rectum, liver, and pancreas). For breast and prostate cancer, the study population will be restricted to females and males, respectively. People with existing cancer or a previous history of cancer will be excluded from the study. No other restrictions applied to participants based on demographic and medical characteristics.
Intervention/exposure: For cohort studies, ascertainment of exposure to diabetes medication should be determined by written measurements, medical records, pharmacy prescription, or insurance claims database. The same criteria apply to case-control, nested case-control, and case-cohort studies. Diabetes mediations in RCTs will not be restricted to monotherapy.
Comparator(s)/control: Use of active comparator is not required in cohort studies, case-control studies, and variants of the case-control design. For a diabetes medication of interest, people on other diabetes medication(s) are also considered eligible as controls. For RCTs, people receiving no treatment, placebo, or active comparator are considered as controls. Population- and hospital-based controls are sources of control groups in cohort studies, nested case-control studies, case-cohort studies, and classic case-control studies and RCTs.
Outcome: Cancer incidence and cancer mortality should be ascertained through national recording system, cancer registry, or death certificates. Eligible effect measure includes standardized mortality ratio (SMR), standardized incidence ratio (SIR), odds ratio (OR), relative risk (RR), and hazard ratio (HR). External controls from local, regional, or national general populations are needed to obtain SMR or SIR. Computing OR, RR, and HR requires internal controls.
Publication characteristics: To assess and synthesize current evidence, eligible studies will be limited to a publication date from January 2011 onwards. Published journal articles will be considered. No language restrictions will be imposed. After titles/abstracts screening, full-text of selected studies in a language other than English will be translated using Google Translate. Searches will be re-run before the final analysis if necessary.
The research team will carry out the search strategies with the assistance of an expert librarian. We will systematically search MEDLINE (Ovid), Embase (Ovid), Web of Science Core Collection, and Cochrane CENTRAL (Ovid). We will identify additional studies from reference lists of extracted studies and previous systematic reviews. Search terms included controlled vocabulary and keywords terms related to various diabetes medications and cancer at multiple sites. Details of the search strategy for MEDLINE (Ovid) are provided in an additional file (Additional file 2).
All titles and abstracts will be imported to Covidence (www.covidence.org), a web-based software platform. Covidence will streamline the processes of screening, study selection, and identifying discrepancies between two reviewers. The study selection will be conducted in two stages: titles/abstracts screening and full-text screening. First, all titles and abstracts will be screened by reviewers (YC, FM, and SS) for potentially relevant papers in accordance with inclusion and exclusion criteria. Studies declined by two reviewers will be excluded. Second, full texts of all potentially relevant studies will be retrieved and screened independently for eligibility by two reviewers. We will resolve any disagreement through discussion, or if required, we will consult a senior reviewer (RAM). Reasons for exclusion will be documented at the full-text screening. We will identify and collate multiple reports of the same study. Duplicates will be excluded, and each study will be counted once. We will record the selection process in sufficient detail to complete a PRISMA flow diagram and ‘Characteristics of excluded studies’ table.
Two reviewers will extract relevant data independently from all selected studies into a standardized data collection form using Covidence. Before full data extraction, necessary amendments will be made based on a pilot test of twenty studies. Discrepancy respecting extracted data will be resolved by discussion or consensus with a third reviewer. All data in languages other than English will be translated. The lead reviewer will be responsible for primary data extraction and the second reviewer will verify extracted data.
The extracted information will include:
Administrative details: author(s); year of publication, year in which study was conducted;
Details of the study: location(s) of study; study design; duration of follow-up; loss to follow-up (%);
Details of participants: age (range, mean, and standard deviation); sex (% male); sample size;
Details of intervention/exposure: medication name(s); dose, frequency and/or duration of use;
Details of control group: medication name(s), dose, frequency, and/or duration of use if any;
Details of outcomes: cancer identification through national recording system, cancer registry, death certificates, or others.
Details of data analysis: reported numbers for outcome; whether employing an intention-to-treat analysis for RCTs; potential confounders adjusted in observational studies.
The methodological quality of observational studies and RCTs will be examined by two reviewers independently using the Newcastle-Ottawa scale (NOS) (24) and Cochrane Collaboration’s tool for assessing the risk of bias (25). NOS assesses three domains for case-control and cohort studies: selection of study group, comparability of groups, and ascertainment of exposure or outcome of interest. The latter comprises selection, performance, detection, attrition, reporting, and other bias.
We will conduct a narrative synthesis first to summarize the findings of selected studies following the Centre for Reviews and Dissemination (26). Textual description and/or tabulation will compare and contrast potential differences. When two or more studies evaluating the same drug class and cancer site with similar study designs are identified, we will employ a random-effects model to pool the data (27). We will use Cochran’s Q to test statistical heterogeneity across included studies with significance level P = 0.1 for chi-squared (28). Heterogeneity will be quantified using the I2 statistic (29): < 25% and > 75% as low and high heterogeneity, respectively. Possible sources of between-study heterogeneity can be attributed to different diabetes medication types, cancer sites, study designs, participants’ characteristics, etc. Primary analyses will be performed at drug class level for each of the seven broad classes. We will perform subgroup analyses and meta-regression to explore the sources of heterogeneity if possible.
Meta-bias: Funnel plot will be used to visually inspect potential publication bias. If the number of included studies is greater than 10, Egger’s test for asymmetry of the funnel plot may be performed (30).
Confidence in cumulative evidence: To evaluate certainty of the body of evidence, we will use the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) approach (31).
Differences between the protocol and the review: We will document unforeseeable changes and report them in the final manuscript (32).