Application of stratum-specific likelihood ratios in mental health screening.

Abstract

The accuracy of a diagnostic procedure is commonly assessed by measuring sensitivity, specificity and positive and negative predictive values. Likelihood ratios provide an alternative method for describing these results, though they are typically reported only for dichotomized outcomes. However, likelihood ratios can also be applied to ordinal or continuous results. The present paper discusses the application of stratum-specific likelihood ratios in a primary care setting using the General Health Questionnaire (GHQ-12) and the Symptom Check List 90-R (SCL-90-R). A randomly selected sample (n = 408) of adult outpatients from primary care offices in Düsseldorf was screened using the German versions of the GHQ-12 and the SCL-90-R. Logistic regression analysis indicated that stratum-specific or multilevel likelihood ratios preserve more information than a fixed threshold approach with a single cutoff point. For each test, five clinically useful strata with monotonically increasing stratum-specific likelihood ratios were selected. Stratum-specific likelihood ratios have enormous practical value, and they are becoming an important way of expressing and comparing the usefulness of different tests. Stratum-specific likelihood ratios reduce the spectrum bias that might arise if only two categories (cases and non-cases) are chosen. Additionally, multilevel likelihood ratios can be used as bedside information to obtain the post-test probability from the pre-test probability of the disorder.