We consulted the Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols (PRISMA-P) guidelines while preparing this protocol and included them as Additional file 1 .
The broad inclusion criteria for this review allow for any study design published in English, including decision analyses, that (i) include women aged 65 years and older in the US, (ii) assess women’s comorbidity (either as a specific condition or a summary score) and/or functional impairments and/or health status, and (iii) report an outcome measure that addresses recent screening mammography utilization. We plan to include any randomized clinical trials and cohort, case–control, quasi-experimental, and cross-sectional studies that meet the above inclusion criteria. Case reports will be excluded. Most, if not all, of the target population are Medicare beneficiaries, with screening mammography covered based on policy changes implemented in 1991 . Women in this age range are near the upper limit of the USPSTF primary screening mammography guidelines (age 74 years) and include age groups where data are currently inconclusive to provide screening recommendations . To account for the Medicare policy effect, we will exclude studies evaluating screening utilization prior to 1991 . Studies must either have complete populations aged 65 years and older or specific subgroups aged 65 years or older to be included in the review.
Information sources and search strategy
The following electronic health databases will be searched: MEDLINE (using the PubMed interface), Excerpta Medica Database (EMBASE), Cumulative Index to Nursing and Allied Health Literature (CINAHL), and PsycINFO. A systematic search of the literature will be conducted between January 1, 1991, and March 1, 2016, to identify all relevant studies in English. Additional studies will be found by searching editorials, letters, overview-type articles, and citations of review articles or by contacting experts in the field regarding any unpublished articles that might be suitable for inclusion.
A search strategy to identify relevant studies will be developed in collaboration with a librarian. We will derive initial terms from Medical Subject Headings (MeSH) of key articles and from other relevant reviews. We will use “breast neoplasms” combined with permutations, variations, and abbreviations of the relevant MeSH keywords and non-MeSH key terms for screening mammography, age and comorbidity, functional limitations, and health status. Comorbidity terms will include Charlson and Elixhauser conditions and any other study comorbidity. An example of a search used for this review is included in the “Search strategy” section of the Appendix.
Outcome of interest
The outcome is the screening mammography utilization in women, defined as screening mammography that occurred within the last 1–5 years. We anticipate that studies will measure screening mammography utilization within the previous 1–5 years and want to ensure that our review will exclude studies that simply measure ever having received screening mammography, since it does not adequately represent recent utilization.
Data management and study selection process
All articles from our MEDLINE and EMBASE searches will be combined, and duplicates will be removed. As a primary screening, titles and abstracts will be reviewed against the eligibility criteria by two members of the review team (JD and DB) in duplicate, who both have expertise in cancer epidemiology and screening. When an article passes primary screening for either member of the review team, the full-text version of the study will be obtained and imported into Mendeley Web (bibliographic database) for easy access among review team members. Reports of the same study will be grouped together for easier data extraction. Using a pre-determined database, we will assess eligibility of each study by reading the full text. The form will specify all of the eligibility criteria and consist of a table with the total list of full-text studies retrieved for our records. Along with these criteria, reviewers will assess the quality of the studies on the same form.
Comments will be included for excluded studies to explain the reasoning behind exclusion. The final list of studies to be included in the review will be added to a database for further data extraction. A PRISMA flowchart will be completed to summarize the entire selection process.
A data extraction form will be used to collect all of the relevant information from the selected studies. The form will include information about study characteristics, such as the type of study, number of participants, length of follow-up, outcome, and quality assessment. Exposures logged in this form will separately include comorbidity scales or specific diseases considered, functional limitation scales used, and measures of health status. The primary outcome is screening mammography utilization, defined as screening mammography that occurred within the last 1–5 years; odds ratios and the corresponding 95 % confidence intervals will serve as our effect estimates, with some studies also showing proportions of utilization. Summary measures from study participants such as measures of socioeconomic status, including education and income, health insurance, and number of physician visits will be extracted. One team member (JD) will extract the data from all studies. One other reviewer will independently assess the data extraction of the other team member for quality assurance. Any discrepancies will be discussed and resolved by the review team.
The team member responsible for data extraction (JD) will also conduct the quality assessment, which will be reviewed by a second team member (DB). As part of the data collection process, we will use the Newcastle-Ottawa Scale to measure the methodological quality of the selected observational studies . Additionally, the Cochrane Collaboration Risk of Bias tool will be used to assess the quality of randomized controlled trials . These scales will provide a summary of the quality ratings of the included studies for descriptive purposes only.
We will conduct a narrative synthesis to describe the findings of included studies, explore associations of interest, and examine the quality of the studies and robustness of the systematic review. Study characteristics, including effect estimates and main findings, will be summarized comprehensively. We will tabulate the full Newcastle-Ottawa and Cochrane risk of bias findings of individual studies for descriptive purposes (see Additional file 2).
Results from each study will be compiled in summary tables for descriptive comparisons of study findings (see Additional file 3). We will evaluate three exposures: comorbidity, functional limitations, and health status. For each exposure, we will aggregate study findings to perform meta-analyses assessing the overall magnitude of the association with recent mammography screening utilization. We acknowledge that there are variations in the strategies for measuring each of our exposures, which will require us to stratify our findings to better account for study heterogeneity.
For the analysis of the association between comorbidity and screening mammography utilization, we will separate studies that measure specific conditions from those evaluating comorbidity using a summary score. Reviewing both individual conditions and comorbidity indices will enable a comprehensive characterization of the most debilitating conditions that could affect screening mammography utilization. Since these are the primary methods of comorbidity measurement, stratified analyses will account for potential sources of heterogeneity.
When analyzing the association between functional limitations and screening mammography utilization, we will group studies that only use ADLs versus IADLs versus both ADLs and IADLs. Finally, for studies analyzing the association between health status and screening mammography utilization, we will separate studies that use a Likert scale health status measure from those using a prognostic index. Within studies using a prognostic index, we will perform subgroup analyses to compare studies that do and do not incorporate ADL or IADL measurements.
The primary outcome in our meta-analysis will be pooled odds ratios for screening utilization with corresponding 95 % confidence intervals. In our analyses, we will assess heterogeneity of studies to determine how structurally different studies are from each other. We will measure study heterogeneity using I2 results and Cochran’s Q from the meta-analysis groupings. If there is no significant heterogeneity within our meta-analysis groupings based on I2 results and Cochran’s Q, we will use the pooled results over the unpooled findings. While we anticipate that our studies have similar sample populations with similar conditions, we will use random-effects modeling to analyze study outcomes because it is a more conservative method to account for inherent variations between our studies under the assumption that the effects being estimated in included studies that are not identical.
Meta-regression controlling for study type, functional limitation, and comorbidity will be used to identify causes of heterogeneity. Our response variable will be the odds ratio of mammography utilization within the last 1–5 years. Along with study type and exposure differences (measures of functional limitation and comorbidity), we will also consider the study year and the minimum age of the study participants as potential covariates. We will account for potential residual heterogeneity and extra variability in our models by using random-effects modeling of our meta-regression.
We will also perform sensitivity analyses to examine potential publication bias including funnel plots, Begg’s test, Egger’s test, trim and fill, and jackknife analyses and report these findings in addition to the primary study findings and subgroup analyses . Moreover, given the differences between the designs that could lead to different findings, we plan to perform a sensitivity analysis that separates randomized clinical trial from observational study findings. The meta-analysis results will also be graphically displayed using forest plots . All analyses will be performed using STATA 13 (Stata, College Station, TX, USA).
Role of the study sponsor
The work represents collaboration between the authors, who represent a group of academic researchers at the University of California, San Francisco. All authors participated in the development of the study design, manuscript preparation, and the decision to submit the protocol for publication. All authors will participate in the data collection and analysis and will be responsible for data completeness and accuracy. The authors intend to publish the results for publication.