Intention-to-diagnose and distinct research foci in diagnostic accuracy studies

Abstract

The intention-to-diagnose principle, an analogue to the intention-to-treat principle in clinical trials, protects the foundation for inference in diagnostic test accuracy studies. This foundation provides for robust control of error rates during hypothesis testing and correct coverage probability during confidence interval estimation of accuracy parameters, in well defined populations for transparent generalisability. The intention-to-diagnose principle requires distinguishing between various non-positive non-negative (NPNN) test results, such as equivocal and invalid results, and appropriate handling of these distinct results during statistical analyses. Pragmatic application accuracy, pragmatic scientific accuracy, and explanatory scientific accuracy are three major research foci in studies of diagnostic test accuracy. Selection of appropriate analysis sets and appropriate handling of NPNN results depend on the specific focus. Selection has important implications regarding preservation of the foundation for statistical inference, generalisability, and comparability with results from other studies. We recommended analyses for diagnostic accuracy studies that include the estimation of accuracy parameters associated with each research foci. A free online tool is available to estimate these accuracy parameters. As with dedicated commitment to the intention-to-treat principle in clinical trials, commitment to application of the intention-to-diagnose principle will lead to studies with high integrity, a comprehensive understanding of the scientific and administration characteristics of the diagnostic test, clarity of interpretation and generalisability, and better clinical decision making.