Likelihood ratios: a real improvement for clinical decision making?

The concept of likelihood ratio has been advocated for several years as one of the better means to evaluate diagnostic tests and as a practical and valuable tool in clinical decision making. In this paper we review the basic concepts underlying the evaluation of diagnostic tests and we explore the properties and usefulness of both positive and negative likelihood ratios compared with sensitivity and specificity. Particular attention is given to the use of likelihood ratios in the clinical setting. Likelihood ratios have three main advantages: they are intuitive, they simplify the predictive value calculation and the overall evaluation of sequential testing. Disadvantages are the non-linearity and the necessity to recalculate probabilities in odds. Although they summarize the information contained in sensitivity and specificity, these characteristics are still necessary for certain clinical decisions. Since likelihood ratios have been promoted among physicians and medical students, we discuss examples of inappropriate use and misunderstandings in the medical literature: the frequent omission of confidence intervals, the choice of cut-off points based on likelihood ratios for positive test results only and the confusion between likelihood ratios for ranges and those for cut-off points.