Multiple method comparison: Statistical model using percentage similarity

• Published in: Cytometry. Part B, Clinical cytometry Clinical cytometry, 2003-07, Vol.54B (1), p.46-53
• Main Authors: Scott, Lesley E., Galpin, Jacky S., Glencross, Deborah K.
• Format: Article
• Language: English
• Subjects: accuracy and precision; Biometry - methods; Bland‐Altman difference plots; CD4 Lymphocyte Count - methods; CD4 Lymphocyte Count - standards; CD4 Lymphocyte Count - statistics & numerical data; coefficient of variation; Humans; Models, Statistical; percentage similarity, histogram (frequency distribution); Reproducibility of Results; Sensitivity and Specificity; Statistical Distributions; statistical multiple method comparison
• ISSN: 1552-4949; 1552-4957
• DOI: 10.1002/cyto.b.10016

BACKGROUND Method comparison typically determines how well two methods agree. This is usually performed using the difference plot model, which measures absolute differences between two methods. This is often not applicable to data with wide ranges of absolute values. An alternative model is introduced that simplifies comparisons specifically for multiple methods compared to a gold standard. METHODS The average between a new method and the gold standard is represented as a percentage of the gold standard. This is interpreted as a percentage similarity value and accommodates wide ranges of data. The representation of the percentage similarity values in a histogram format highlights the accuracy and precision of several compared methods to a gold standard. The calculation of a coefficient of variation further defines agreement between methods. RESULTS Percentage similarity histograms of several new methods can be compared to a gold standard simultaneously, and the comparison easily visualized through use of a single 100% similarity reference line drawn common to all plots. CONCLUSION This simple method of comparison would be particularly useful for multiple method comparison and is especially applicable for centers collating for external quality assessment or assurance programs to demonstrate differences in results between laboratories. Cytometry Part B (Clin. Cytometry) 54B:46–53, 2003. © 2003 Wiley‐Liss, Inc.