Time-to-event analyses in meta-analyses: meta-epidemiological assessment of systematic reviews and their included randomized controlled trials (META-TTE)

Autor: Kreuzberger, Nina, Skoetz, Nicole, Bender, Ralf, van Dalen, Elvira, Hemkens, Lars, Monsef, Ina, Trivella, Marialena, Goldkuhle, Marius
Rok vydání: 2022
Předmět:
DOI: 10.17605/osf.io/5qxbd
Popis: Background: Measurements and comparisons of event times in clinical research are based on time-to-event analyses, or survival analyses as they are more generally referred to. For example, in oncology and cardiology where overall mortality, that is evaluating the occurrence of the event ‘death’ irrespective of its cause, is one of the most explored time-to-event outcomes. Various statistical methods to analyze time-to-event outcomes are available, including the non-parametric method proposed by Kaplan and Meier or the semi-parametric Cox proportional hazards model, which allows the calculation of the widely used hazard ratio for between group comparisons. Methods of time-to-event analysis are associated with methodological peculiarities and their results are not always easy to interpret. Challenges include the potential for informative censoring, possible competing events and situations where the underlying assumption of proportional hazards of the Cox model does not hold. For time-to-event meta-analyses based on aggregated data, which are increasingly used in systematic reviews and guidelines, existing instructions for authors are primarily focused on the recovery of data in order to pool effect estimators and leave the confidence in the included estimators largely untouched. Due to the difficulties and requirements of time-to-event analyses, the lack of guidance carries a high risk of limited review quality if existing uncertainties in primary studies are not adequately identified, considered and reported. Aim: To determine how systematic review authors deal with the specifics of time-to-event analyses in meta-analyses of time-to-event outcomes. Proposed methods: Using a meta-epidemiological study approach and two-stage assessment, we will extract information on the handling of time-to-event outcomes from systematic reviews and the respective trials included in them. We will identify 50 Cochrane and 50 non-Cochrane systematic reviews published during the same time-period on the Cochrane Database of Systematic Reviews and Medline. Eligible reviews will be required to include at least one time-to-event meta-analysis of aggregated data over at least two randomized controlled trials. All project steps, including the screening and selection of relevant reviews and trials as well as the extraction of data, will be performed in duplicate. Data will be extracted for each systematic review, meta-analyzed time-to-event review outcome, trial and analyzed trial outcome and will include general (e.g. publication information, PICO questions, sample sizes, general methods, risk of bias and GRADE assessments, etc.) and time-to-event (meta-)analytic information (e.g. outcome definitions, effect measures, follow-up durations, censoring (for non-administrative reasons) and competing events as well as their consideration in reviews (e.g. through sensitivity analyses, risk of bias ratings or discussion)). We will also extract effect estimators both of meta-analyses and trials and will recalculate meta-analyses, should discrepancies be identified in the selection or generation of effect measures from primary trials, in order to perform error analyses. We will use a flexible database to manage and analyze our generated data over multiple levels. Given the exploratory nature of this project, the planned statistical analyses are primarily descriptive and we will use relative effect measures and corresponding confidence intervals for between group comparisons. To allow adequate interpretability and comparisons across and within subgroups of reviews, we plan to analyse our findings respecting relevant clusters (e.g. Cochrane vs. non-Cochrane systematic reviews, different medical fields, different intervention types, different types of outcomes (single event versus composite, primary versus non-primary, competing events possible versus no-competing events possible)).
Databáze: OpenAIRE
Externí odkaz: