作者: Keith R. Abrams , Clare L. Gillies , Paul C. Lambert
DOI: 10.1002/SIM.2423
关键词: Meta-analysis 、 Baseline (configuration management) 、 Statistics 、 Computer science 、 Bayesian probability 、 Correlation 、 Intervention effect 、 Systematic review 、 Econometrics 、 Sensitivity (control systems) 、 Physical functioning
摘要: This paper considers the quantitative synthesis of published comparative study results when outcome measures used in individual studies and way which they are reported varies between studies. Whilst former difficulty may be overcome, at least to a limited extent, by use standardized effects, latter is often more problematic. Two potential solutions this problem are; sensitivity analyses fully Bayesian approach, pertinent external information included. Both approaches illustrated using two systematic reviews meta-analyses consider difference mean change systolic blood pressure physical functioning an intervention control group. The examples illustrate that adopting as opposed undertaking assuming fixed values for unknown parameters, overall effect can estimated with greater uncertainty, but assessing choice prior distributions such crucial.