Kai M Eggers1, Fred S Apple2, Lars Lind3, Bertil Lindahl4. 1. Department of Medical Sciences, Uppsala University, SE-751 85 Uppsala, Sweden. Electronic address: kai.eggers@ucr.uu.se. 2. Department of Laboratory Medicine and Pathology, Hennepin County Medical Center, Minneapolis, MN, USA; Minneapolis Heart Institute, Abbott-Northwestern Hospital, Minneapolis, MN, USA. 3. Department of Medical Sciences, Uppsala University, SE-751 85 Uppsala, Sweden. 4. Department of Medical Sciences, Uppsala University, SE-751 85 Uppsala, Sweden; Uppsala Clinical Research Center, Uppsala University, SE-751 85 Uppsala, Sweden.
Abstract
BACKGROUND: Cardiac troponin (cTn) is the biomarker of choice for assessment of patients with acute coronary syndromes. Guidelines recommend the cTn 99th percentile derived from a cardiovascular healthy reference population as decision threshold. The importance of standardized criteria for the composition of such a reference population is well acknowledged. In this analysis, we investigated to which extent different statistical methods might have bearing on the calculated cTnI 99th percentile. METHODS: cTnI (Abbott) 99th percentiles were determined in 521 cardiovascular healthy community-dwelling subjects using the nonparametric method, the Harrell-Davis bootstrap method and the robust method together with different tests to identify potential outliers (Dixon, Tukey, Reed) and different statistical softwares. RESULTS: The cTnI 99th percentiles (nonparametric method) were 37ng/L (total population), 42ng/L (men) and 25ng/L (women). These estimates differed by -7.4% to +5.7% using the Harrell-Davis bootstrap method and were up to 64.1% lower using the robust method. For the robust method, cTnI 99th percentiles varied by 44.2% depending on the applied software. The method of Tukey classified nine subjects as outliers while no outlier was detected using the other methods. Excluding these nine subjects resulted in up to 60.2% lower cTnI 99th percentiles. CONCLUSIONS: Our results emphasize the need of a standardized statistical approach to calculate cTnI 99th percentiles. Our findings support the use of the nonparametric method and a conservative approach to detect outliers. This requires that the assessed population is sufficiently large and well selected on the basis of stringently applied clinical criteria.
BACKGROUND: Cardiac troponin (cTn) is the biomarker of choice for assessment of patients with acute coronary syndromes. Guidelines recommend the cTn 99th percentile derived from a cardiovascular healthy reference population as decision threshold. The importance of standardized criteria for the composition of such a reference population is well acknowledged. In this analysis, we investigated to which extent different statistical methods might have bearing on the calculated cTnI 99th percentile. METHODS:cTnI (Abbott) 99th percentiles were determined in 521 cardiovascular healthy community-dwelling subjects using the nonparametric method, the Harrell-Davis bootstrap method and the robust method together with different tests to identify potential outliers (Dixon, Tukey, Reed) and different statistical softwares. RESULTS: The cTnI 99th percentiles (nonparametric method) were 37ng/L (total population), 42ng/L (men) and 25ng/L (women). These estimates differed by -7.4% to +5.7% using the Harrell-Davis bootstrap method and were up to 64.1% lower using the robust method. For the robust method, cTnI 99th percentiles varied by 44.2% depending on the applied software. The method of Tukey classified nine subjects as outliers while no outlier was detected using the other methods. Excluding these nine subjects resulted in up to 60.2% lower cTnI 99th percentiles. CONCLUSIONS: Our results emphasize the need of a standardized statistical approach to calculate cTnI 99th percentiles. Our findings support the use of the nonparametric method and a conservative approach to detect outliers. This requires that the assessed population is sufficiently large and well selected on the basis of stringently applied clinical criteria.
Authors: Paul Welsh; David Preiss; Anoop S V Shah; David McAllister; Andrew Briggs; Charles Boachie; Alex McConnachie; Caroline Hayward; Sandosh Padmanabhan; Claire Welsh; Mark Woodward; Archie Campbell; David Porteous; Nicholas L Mills; Naveed Sattar Journal: Clin Chem Date: 2018-08-20 Impact factor: 8.327