Misclassification and measurement error - planning a study and interpreting results.

Abstract

Background: Measurement error must always be considered when planning a research project and interpreting its results. The accuracy of some data collected during a study can often be confidently assured, but more than one measurement or observer is needed to assess exposure and outcomes status in cases where clinical measurement is prone to measurement error. Little attention is paid in nursing research to misclassification and measurement error. Bias is often discussed in nursing research education, but not its potential consequences or measures that can be taken to improve the study's quality.

Aim: To present examples of random measurement error - misclassification of a binary outcome - in a continuous exposure and outcomes variable, to address this gap in nurses' research training.

Discussion: The article discusses the relationship between exposure and outcome in the absence and presence of measurement error using risk (relative risk) and association using correlation. It provides methods to estimate the true value of these measures of risk and association, when only given the clinical measurements with errors.

Conclusion: If the assumption of random error holds, attenuation of risk or association towards the null will occur.

Implications for practice: Understanding the effect of measurement error including misclassification will enable researchers to interpret the results of their studies, and to take into consideration this potential error when planning and conducting a study.