BMJ  2005;330:146 (15 January), doi:10.1136/bmj.330.7483.146-a

Letter

Issues in reporting epidemiological studies

Can't we all agree on confidence intervals?

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

EDITOR—Pocock et al discuss many valuable caveats regarding published epidemiological analyses.¹ The authors, however, perpetuate two misconceptions.

The first is that post hoc power calculations help the reader interpret already collected data. Power calculations are a guide to study design—namely, to consider what the investigators might observe if they do the study. Once they collect the data, the confidence interval contains all the relevant information because it is based on what the investigators actually observed.²

The second misconception is that the P value contains useful information. In almost all cases the confidence interval (along with the point estimate) is the better choice. But not, as many aver, to see whether it crosses some arbitrary null value, such as a relative risk of 1. Rather readers should ask whether the confidence interval excludes important effects. For example, if a reduction of cardiovascular mortality of at least 5% is meaningful, . . . [Full text of this article]

Department of Ambulatory Care and Prevention, Harvard Medical School, Boston, MA 02215, USA Matthew_gillman@hms.harvard.edu

CiteULike    Complore    Connotea    Del.icio.us    Digg    Reddit    StumbleUpon    Technorati    What's this?

Relevant Article

Issues in the reporting of epidemiological studies: a survey of recent practice

Stuart J Pocock, Timothy J Collier, Kimberley J Dandreo, Bianca L de Stavola, Marlene B Goldman, Leslie A Kalish, Linda E Kasten, and Valerie A McCormack
BMJ 2004 329: 883. [Abstract] [Full Text] [PDF]

Online poll

Find out more >>