Use of the likelihood ratio in the management of the young child with fever.

The febrile infant is a common clinical problem for the primary health care provider. This paper employs the example of a young infant with fever to describe an important epidemiologic concept that is useful in the interpretation of diagnostic data--the likelihood ratio. The likelihood ratio expresses the odds of a given diagnostic test result occurring in a patient with (as opposed to without) the target disorder. Likelihood ratios have three properties that are helpful for clinicians: (1) The likelihoods that make up the likelihood ratio are calculated in a manner similar to sensitivity and specificity and therefore show little variation with change in disease prevalence (unlike predictive values, which change dramatically with disease prevalence). (2) Likelihood ratios can be calculated at several levels of a sign, symptom, or laboratory test. (3) Likelihood ratios can be used to shorten the list of diagnostic possibilities because the pretest "odds" X likelihood ratio = post-test "odds" of a disease. Using likelihood ratios in the practice of primary care medicine should reduce the number of patients with false-positive or false-negative results, sparing some patients needless therapy as well as minimizing the number of patients denied efficacious interventions. Support for likelihood ratios within the primary care medical community will hasten their availability in laboratories of clinical medicine.