Decision threshold and detection capability Ivo Leito ivo
Decision threshold and detection capability Ivo Leito ivo. leito@ut. ee http: //www. ut. ee/ams/ 5/20/2021 http: //www. ut. ee/ams/ 1
Decision threshold (limit) and detection capability • Recently these have become increasingly important • Often required by legislation – E. g. European Commission decision 2002/657/EC concerning the performance of analytical methods and the interpretation of results 5/20/2021 http: //www. ut. ee/ams/ 2
a and b error: fire alarm There is no There is fire Alarm starts No error a error Alarm does not start b error No error 5/20/2021 http: //www. ut. ee/ams/ 3
a and b error: contamination detection There is contamination There is no contamination Contaminant detected No error a error Contaminant not setected b error No error False positive False negative 5/20/2021 http: //www. ut. ee/ams/ 4
Decision threshold (limit), CCa • Answers the question: Starting from which analyte concentration it is possible to state that the probability of false positive is a or lower? or Starting from which analyte concentration it is possible lates millisest tulemuse väärtusest võime öelda, et proovi nõuetele mittevastavuse tõenäosus on üle 1 -a? • According to legislation the a probabilities are normally 0. 01 or 0. 05 depending on analyte 5/20/2021 http: //www. ut. ee/ams/ 5
Detection capability, CCb • Answers the question: Starting from which analyte content we can say the probability of a false negative is b or less? or Starting from which analyte content we can say that the probability of analyte detection is above 1 -b? • According to legislation the b probability is 0. 05 – That is 1 - b = 0. 95 (or 95%) 5/20/2021 http: //www. ut. ee/ams/ 6
Normal Distribution • When we find U then we use for k two-tailed quantiles • k= 1 (68%), 2 (95. 5%), 3 (99. 7%) 68% 95. 5% 99. 7% 5/20/2021 http: //www. ut. ee/ams/ 7
Normal Distribution • Here we use as k single-tailed quantiles: • k= 1. 64 (a = 0. 05), 2. 33 (a = 0. 01) a = 0. 05 (5%) 1 - a = 0. 95 a = 0. 01 (1%) 1 – a = 0. 99 5/20/2021 http: //www. ut. ee/ams/ 8
Decision threshold, if there is no maximum permissible limit Distribution function of a blank Repeatability s of blank Average concentration found for blanks If the result is beyond 2. 33 s level then the probability of the analyte not being present in the sample is less than 1% 5/20/2021 http: //www. ut. ee/ams/ 9
Decision threshold if there is a legal upper limit Distribution function of analyte determination in sample contaminated at the legal limit Standard deviation of sample contaminated at the legal limit Average analyte content found in samples contaminated at the legal limit If the result is beyond the 2. 33 s level then the probability that the content is not above the limit is less than 1% 5/20/2021 CCa http: //www. ut. ee/ams/ CCa 10
Detection capability Distribution function of a sample contaminated at detection capability level Repeatability s of a sample contaminated at the level of decision threshold If the result is below CCa then the sample is considered not contaminated. The probability of contamination is 5%. Decision threshold CCa 5/20/2021 CCb CCa = CCb – http: //www. ut. ee/ams/ 1. 64 ss 11
Assumptions • The measurand is Normally distributed – This means that a sufficient number of parallel measurements must be made • Repeatability is the same at CCa and CCb levels • Repeatability must be determined under constant analyte content • Systematic effects have been taken into account 5/20/2021 http: //www. ut. ee/ams/ 12
- Slides: 12