OBJECTIVE: To demonstrate how the results of a meta-analysis can confuse rather than clarify therapeutic dilemmas if clinical heterogeneity among trials is ignored. Then to further discuss the qualities emergency physicians should expect from published meta-analyses if they are to affect clinical practice. SUBJECTS AND METHODS: The data and results were examined from 23 randomised controlled trials of selective decontamination of the digestive tract (SDD), which have been combined in a previous meta-analysis. These were reviewed to take account of clinical heterogeneity, particularly with regard to severity of patient illness. RESULTS: Severity of patient illness predicts degree of reduction in mortality with SDD in a regression analysis: log odds ratio (OR) of death with SDD = -0.0074 - (0.0035 x control group mortality rate), P = 0.017. This is also true when trials are stratified into more and less severely ill patients: pooled OR (a) for CMR > 41% = 0.69 (0.54 to 0.89), with (b) CMR < 3% = 1.02 (0.86 to 1.21). This difference was not suggested by the original meta-analysis result. CONCLUSIONS: Failure to take account of clinical heterogeneity between trials can mean a meta-analysis result ignores important differences in the effect of a treatment on different groups of patients. The discussion indicates how emergency physicians might guard against basing clinical practice on misleading meta-analysis results.
OBJECTIVE: To demonstrate how the results of a meta-analysis can confuse rather than clarify therapeutic dilemmas if clinical heterogeneity among trials is ignored. Then to further discuss the qualities emergency physicians should expect from published meta-analyses if they are to affect clinical practice. SUBJECTS AND METHODS: The data and results were examined from 23 randomised controlled trials of selective decontamination of the digestive tract (SDD), which have been combined in a previous meta-analysis. These were reviewed to take account of clinical heterogeneity, particularly with regard to severity of patient illness. RESULTS: Severity of patient illness predicts degree of reduction in mortality with SDD in a regression analysis: log odds ratio (OR) of death with SDD = -0.0074 - (0.0035 x control group mortality rate), P = 0.017. This is also true when trials are stratified into more and less severely ill patients: pooled OR (a) for CMR > 41% = 0.69 (0.54 to 0.89), with (b) CMR < 3% = 1.02 (0.86 to 1.21). This difference was not suggested by the original meta-analysis result. CONCLUSIONS: Failure to take account of clinical heterogeneity between trials can mean a meta-analysis result ignores important differences in the effect of a treatment on different groups of patients. The discussion indicates how emergency physicians might guard against basing clinical practice on misleading meta-analysis results.