Unit of analysis errors should be clarified in meta-analyses

doi:10.1136/bmj.326.7385.397

BMJ 2003;326:397 ( 15 February )

Letters

Unit of analysis errors should be clarified in meta-analyses

The first 150 words of the full text of this article appear below.

EDITOR $---$ Weingarten et al present a comprehensive study in what is a complex area of research.¹ We were, however, unclearwhether any of the included primary studies had unit of analysiserrors and how the authors dealt with such studies in their meta-analysis.

Unit of analysis errors occur in cluster randomised trials when individual patients' data are analysed as if there was noclustering in the provider, practice, or units randomised to theintervention groups (patients' data are analysed as independentobservations).² Standard statistical methods that do not accountfor cluster effects in cluster randomised trial data result inthe overestimation of the significance of an intervention (artificiallyextreme P values and overly narrow confidence intervals).²Correspondingly, the inclusion of studies with unit of analysiserrors in a meta-analysis will give greater weight to the resultsof such studies.³

The table of included studies reported by Weingarten et al . . . [Full text of this article]

CiteULike

Complore

Connotea

Del.icio.us Add to Digg

Digg

StumbleUpon Add to Technorati

Technorati What's this?

Relevant Article

Interventions used in disease management programmes for patients with chronic illness $---$ which ones work? Meta-analysis of published reports: Scott R Weingarten, James M Henning, Enkhe Badamgarav, Kevin Knight, Vic Hasselblad, Anacleto Gano, Jr, and Joshua J Ofman
BMJ 2002 325: 925. [Abstract] [Full Text] [PDF]

This article has been cited by other articles:

GILBODY, S., BOWER, P., WHITTY, P. (2006). Costs and consequences of enhanced primary care for depression: Systematic review of randomised economic evaluations. Br. J. Psychiatry 189: 297-308 [Abstract] [Full text]