SIPMeL

Login

094 - A simple and effective bayesian calculator for daily practice of Laboratory Medicine

Autor(s): D. Giavarina, G. Pistolato, R.M. Dorizzi

Issue: RIMeL - IJLaM, Vol. 3, N. 2, 2007 (MAF Servizi srl ed.)

Page(s): 94-98

The extension of Evidence Based Medicine (EBM) principles to diagnostics has been widely advocated and the assessment of accuracy is essential for an “evidence based” evaluation of diagnostic procedures. The practice of Evidence Based Laboratory Medicine (EBLM) requires not only the calculation of likelihood ratios, predictive values and number needed to diagnose but also the estimation of their imprecision. However, the calculation of these parameters is not very easy nor common and the inadequacy of the available tools limits EBLM adoption in general practice and specialized environment. Information technology is making more and more available accessible easy-to-use, accurate and fast tools for these needs. Here we summarize the Bayesian theory for the probabilistic evaluation of a diagnostic test and briefly review several Web available Bayesian calculators. A simple calculator written in JAVA is presented that allows entering the data in a single page in Italian and English, prepared on behalf of the Evidence Based Laboratory Medicine group of the Italian Society of Laboratory Medicine. The diagnostic performance of BNP assay at different cut-off values reported in a classical paper are used in order to clarify the possible use of the proposed calculator. The calculator instantaneously shows the very poor performance of the cut-off of 17.9 ng/L (number needed to diagnose = 4.72, i.e. almost five assays are needed for a single diagnosis) and the much better performance of the cut-off of 76 ng/L (number needed to diagnose = 1.91).

Article in PDF format

Back to current issue