Article type
Year
Abstract
Background:
Over a third of stroke reviews in the Cochrane Database of Systematic Reviews include a continuous primary outcome; three-quarters contain a continuous secondary outcome. These continuous outcome measures are often of crucial relevance to patients, carers and healthcare professionals (e.g. quality of life measures) or are pivotal in economic evaluations (e.g. hospital length of stay). Many continuous outcomes in this context are not normally distributed, and resulting analysis strategies and reporting vary. Omission of mean or standard deviation (SD) data from clinical trial reports, perhaps due to the skewed distribution of the outcome, can prevent the inclusion of the trial in a standard meta-analysis, potentially causing bias in that analysis.
Objectives:
To identify current approaches and develop improved methods for handling missing summary statistics for continuous outcomes within meta-analysis, enabling best use to be made of available clinical trial evidence.
Methods:
We investigate how best to impute the mean or SD where either of these has not been reported. In certain circumstances these are suitable summary statistics to analyse in a meta-analysis, regardless of the underlying continuous distribution. We report on a comprehensive review of the literature to identify all methods of deriving missing means and standard deviations. We searched electronic resources (including MEDLINE, EMBASE, Web of Science, BioMed Central and The Cochrane Library), relevant journals and grey literature from inception up to March 2014. This updates a previous review (Wiebe 2006) of methods used to determine the SD and extends it to include methods for imputing the mean. Our search focuses on trial-level imputation from non-parametric summaries, but also considers algebraic recalculation from parametric summaries, trial-level imputation from external data sources or another treatment group internally, trial-level imputation of correlation and solutions at the meta-analysis level. For each method, we note whether it calculates the mean or the SD, list which summary statistics are needed to implement it and summarise any assumptions.
Over a third of stroke reviews in the Cochrane Database of Systematic Reviews include a continuous primary outcome; three-quarters contain a continuous secondary outcome. These continuous outcome measures are often of crucial relevance to patients, carers and healthcare professionals (e.g. quality of life measures) or are pivotal in economic evaluations (e.g. hospital length of stay). Many continuous outcomes in this context are not normally distributed, and resulting analysis strategies and reporting vary. Omission of mean or standard deviation (SD) data from clinical trial reports, perhaps due to the skewed distribution of the outcome, can prevent the inclusion of the trial in a standard meta-analysis, potentially causing bias in that analysis.
Objectives:
To identify current approaches and develop improved methods for handling missing summary statistics for continuous outcomes within meta-analysis, enabling best use to be made of available clinical trial evidence.
Methods:
We investigate how best to impute the mean or SD where either of these has not been reported. In certain circumstances these are suitable summary statistics to analyse in a meta-analysis, regardless of the underlying continuous distribution. We report on a comprehensive review of the literature to identify all methods of deriving missing means and standard deviations. We searched electronic resources (including MEDLINE, EMBASE, Web of Science, BioMed Central and The Cochrane Library), relevant journals and grey literature from inception up to March 2014. This updates a previous review (Wiebe 2006) of methods used to determine the SD and extends it to include methods for imputing the mean. Our search focuses on trial-level imputation from non-parametric summaries, but also considers algebraic recalculation from parametric summaries, trial-level imputation from external data sources or another treatment group internally, trial-level imputation of correlation and solutions at the meta-analysis level. For each method, we note whether it calculates the mean or the SD, list which summary statistics are needed to implement it and summarise any assumptions.