Regression analysis in interlaboratory surveys: a case study with cholesterol and triglycerides

1. A new interlaboratory survey design, that uses regression analysis to compare results from each laboratory with target values, was tested using cholesterol and triglyceride analyses. The fifty New Zealand laboratories involved showed considerable interlaboratory variation (CV = 8% to 27% for cholesterol, 13% to 113% for triglycerides), 30% and 40% of which was associated with systematic differences between laboratories. 2. End-of-period summaries using regression analysis confirmed the presence of systematic errors. These were either simple types caused apparently by incorrect standardisation (regression slope, B not equal to 1.0) or inappropriate blank correction (intercept, A not equal to zero) or complex types presumably due to nonlinearity or nonspecificity. Graphical display of results from each laboratory aided fault diagnosis and allowed the detection of between-run standardisation differences. 3. Method comparison studies were made: the only highly significant result being lower precision achieved by enzymatic cholesterol methods compared with other colorimetric methods.