A Review on Metaanalysis of Biomarkers: Promises and Pitfalls

Abstract

BACKGROUND: The last 30 years have seen an exponential increase in metaanalyses. By combining multiple studies, metaanalysis can provide an overview of the totality of evidence on a particular question and the statistical power needed to reduce random error and produce precise estimates of even modest effect sizes. This capability is of particular value when many small studies address similar questions [such as in the investigation of novel cardiovascular disease (CVD) biomarkers]. To provide reliable evidence, however, metaanalyses must be undertaken robustly. CONTENT: In this review, we describe the major issues to consider when designing and conducting metaanalyses, including the design of constituent studies, selection criteria, assessment of exposures and disease outcomes, and control of bias and confounding. Some of the potential challenges and pitfalls associated with metaanalysis are examined, and their consequences are considered. We use 2 examples of novel biomarkers for CVD-homocysteine and triglycerides-to illustrate how metaanalyses of observational studies have contributed to, and on occasion hindered, our understanding; and how subsequent work has built upon these findings. SUMMARY: Metaanalyses of observational studies, particularly metaanalyses of individual-participant data, have the power to provide robust evidence to support our understanding of the role of novel biomarkers for disease. The characteristic limitations and challenges of these studies, including their inability to detect causal associations, must be considered, however, and additional evidence from randomized controlled trials and genetic studies is frequently required to elucidate fully the role of novel biomarkers in predicting cardiovascular risk.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 15 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.