Calibration of the Periotron 8000® and 6000® by polynomial regression

This paper reports the detailed calibration of the new Periotron 8000 with different fluids and uses the method of least squares to derive polynomial regression equations up to the 6th order, to investigate the most accurate descriptor of the resulting calibration lines. The use of a 4th order polynomial regression equation (recommended by the manufacturer) provided better coefficients of determination (R: 0.999) and root mean square errors (RMSE = 1.6) than either linear regression (R: 0.986, RMSE = 10.9) or quadratic models (R: 0.998, RMSE = 3.2). Data derived using the manufacturer's MLCONVERT software program lacked accuracy and incurred large errors for volumes > 0.5 μl. Calibrations performed on one day could be used with accuracy to derive volumes > 0.1 μl collected on subsequent days, when using the same machine (s.d. for residuals plot =2.49 Periotron units), but this was not the case for different machines (s.d. = 9.57 Periotron units). Varying serum protein concentration by up to 500% had a negligible effect on calculated volumes above 0.1 μl. We conclude that the Periotron 8000® is at least as reliable a machine as the Periotron 6000®, and that the calibration lines for both machines are best described using 4th order polynomial regression equations and "look-up" tables, rather than quadratic (Periotron 6000®) or the manufacturer's software (Periotron 8000®). Serum seems to be an acceptable GCF substitute for calibrations, which can be performed 1 day, and used on subsequent days for a given machine and for volumes above 0.1 μl. polynomial regression; GCF volume; calibration; reliability; Periotron 8000®, Periotron 6000®.