Using probit regression to disclose the analytical performance of qualitative and semi-quantitative tests
Authors:
- 6 July 2016
- journal article
- research article
- Published by Taylor & Francis in Scandinavian Journal of Clinical and Laboratory Investigation
- Vol. 76 (7) , 515-519
Abstract
Background: The analytical performance of qualitative and semi-quantitative tests is usually studied by calculating the fraction of positive results after replicate testing of a few specimens with known concentrations of the analyte. We propose using probit regression to model the probability of positive results as a function of the analyte concentration, based on testing many specimens once with a qualitative and a quantitative test. Methods: We collected laboratory data where urine specimens had been analyzed by both a urine albumin (‘protein’) dipstick test (Combur-Test strips) and a quantitative test (BN ProSpec System). For each dipstick cut-off level probit regression was used to estimate the probability of positive results as a function of urine albumin concentration. We also used probit regression to estimate the standard deviation of the continuous measurement signal that lies behind the binary test response. Finally, we used probit regression to estimate the probability of reading a specific semi-quantitative dipstick result as a function of urine albumin concentration. Results: Based on analyses of 3259 specimens, the concentration of urine albumin with a 0.5 (50%) probability of positive result was 57 mg/L at the lowest possible cut-off limit, and 246 and 750 mg/L at the next (higher) levels. The corresponding standard deviations were 29, 83, and 217 mg/L, respectively. Semi-quantitatively, the maximum probability of these three readings occurred at a u-albumin of 117, 420, and 1200 mg/L, respectively. Conclusions: Probit regression is a useful tool to study the analytical performance of qualitative and semi-quantitative tests.Keywords
This publication has 6 references indexed in Scilit:
- Comment on the article entitled “How to Deal with Dichotomous Tests? Application of a Rankit Ordinal Scale Model with examples from the Nordic Ordinal Scale Project on screening tests” by Petersen et al.Scandinavian Journal of Clinical and Laboratory Investigation, 2009
- How to deal with semi-quantitative tests? Application of an ordinal scale model to measurements of urine glucoseScandinavian Journal of Clinical and Laboratory Investigation, 2009
- How to deal with dichotomous tests? Application of a rankit ordinal scale model with examples from the Nordic ordinal scale project on screening testsScandinavian Journal of Clinical and Laboratory Investigation, 2008
- Characterising the performance of qualitative analytical methods: Statistics and terminologyTrAC Trends in Analytical Chemistry, 2005
- A Model for Setting Analytical Quality Specifications and Design of Control for Measurements on the Ordinal ScaleClinical Chemistry and Laboratory Medicine (CCLM), 2000
- Interpreting Probability Models1994