Article type
Year
Abstract
Background: Mortality is frequently used as a primary outcome in critical care trials. It is a patient-orientated variable and robust against information bias. Mortality incidence however needs to be measured at a defined time-point. Practice of meta-analysis in critical care shows that follow-up times of trials in critical care medicine may differ substantially. This has substantial implications on the potential pooling of such mortality estimates.
Objectives: Describe the current practice of mortality follow-up time definitions in a representative sample of published critical care randomised controlled trials (RCTs) and analyze the influence of handling of different follow-up times on pooled effect estimates.
Methods: We searched CENTRAL, EMBASE, MEDLINE, PASCAL Biomed, and PsycINFO for studies published after 2000 using the Cochrane RCT-filter and a critical care-filter. A random sample of 50% was drawn for further title and abstract review. Study characteristics such as sample size, type of intervention, disease spectrum, intensive care unit setting, hospital setting, funding and patient characteristics including age, gender, severity of illness scores were extracted, as well as number and time-points of mortality ascertainment within individual studies.
Meta-regression and multilevel mixed-effects linear regression was used to analyze the influence of follow-up time on deviation of pooled risk ratios from study-baseline, using the aforementioned study-characteristics as co-variables, allowing for clustering on level of study and time-point.
Results: Search resulted in 9246 studies, 4573 of which were chosen as a random sample. After title-, abstract- and full text-review, 106 studies representing 63,713 patients were included; 60 (57%) studies reported only one time-point, 26 (25%) reported two, 14 (13%) reported three, 4 (4%) reported four, and 2 (2%) reported five. No influence of time-point on effect estimates was found using the aforementioned methods.
Conclusions: In a large sample of critical care RCTs, almost half the studies reported more than one mortality time-point. We found no influence of different time-points on pooled effect estimates.
Objectives: Describe the current practice of mortality follow-up time definitions in a representative sample of published critical care randomised controlled trials (RCTs) and analyze the influence of handling of different follow-up times on pooled effect estimates.
Methods: We searched CENTRAL, EMBASE, MEDLINE, PASCAL Biomed, and PsycINFO for studies published after 2000 using the Cochrane RCT-filter and a critical care-filter. A random sample of 50% was drawn for further title and abstract review. Study characteristics such as sample size, type of intervention, disease spectrum, intensive care unit setting, hospital setting, funding and patient characteristics including age, gender, severity of illness scores were extracted, as well as number and time-points of mortality ascertainment within individual studies.
Meta-regression and multilevel mixed-effects linear regression was used to analyze the influence of follow-up time on deviation of pooled risk ratios from study-baseline, using the aforementioned study-characteristics as co-variables, allowing for clustering on level of study and time-point.
Results: Search resulted in 9246 studies, 4573 of which were chosen as a random sample. After title-, abstract- and full text-review, 106 studies representing 63,713 patients were included; 60 (57%) studies reported only one time-point, 26 (25%) reported two, 14 (13%) reported three, 4 (4%) reported four, and 2 (2%) reported five. No influence of time-point on effect estimates was found using the aforementioned methods.
Conclusions: In a large sample of critical care RCTs, almost half the studies reported more than one mortality time-point. We found no influence of different time-points on pooled effect estimates.