Objective: To replicate and to critique a recently published meta-analysis[1] of the incidence of nonpreventable serious and fatal adverse drug reactions (ADRs) in hospitalized patients, to better understand its results and conclusions.
Methods: The published methods described in the meta-analysis of Lazarou and colleagues were followed.[1] This meta-analysis reviewed 30 original publications describing 39 prospective studies. In each study, the numbers of patients with nonpreventable ADRs, probably or definitely related to drugs, were sought to allow calculation of the incidence of "all-severities," serious and fatal, ADRs. In the original meta-analysis, these ADR incidences were then pooled to provide estimates of the incidence in all hospitalized patients. In our analysis, the original studies were examined by 2 investigators for consistency with the study search and inclusion criteria of the meta-analysis by Lazarou and colleagues, as well as accuracy and appropriateness of data extraction, meta-analysis, and conclusions.
Results: Multiple sources of heterogeneity among studies and data were found and include important differences in populations and hospital wards monitored, surveillance techniques, ADR definitions, determination of preventability of ADRs, distinguishing relationship to drugs, and in formats of reporting ADRs (by numbers of events or by patients). Imputations of event numerators made by the authors of the original meta-analysis were questionable and may overestimate the results of any individual study. With regard to fatal ADRs, the problem of small numbers of events is likely to introduce large errors in incidence estimates. Simple pooling of fatal event frequencies from only those studies specifically reporting the number of fatal ADRs, as was done in the meta-analysis of Lazarou and colleagues, is likely to dramatically overestimate the death rate.
Conclusion: Meta-analysis was invalid because of heterogeneity of the studies. Most of these studies did not report the data needed for incidence calculations. The methodology used was seriously flawed, and no conclusions regarding ADR incidence rates in the hospitalized population in the United States should be made on the basis of the original meta-analysis.