A total of 1628 human plasma samples from Cycle 1 of the Canadian Health Measures Survey were assayed for
total 25-hydroxyvitamin D using the DiaSorin RIA method and the Diasorin ”LIAISON 25 OH Vitamin D Total“ method.
Bland-Altman comparison showed an average bias of 4.8 ± 16.7 nmol/L (6.3%: P<0.001) with the LIAISON method giving
higher values. The relationship was investigated using linear and Deming regression. Linear regression gave:
LIAISON = RIA±(0.87 ± 0.02) + (13.3 ± 1.2) (mean ± SE) and weighted Deming regression (constant CV) gave:
LIAISON = RIA±(1.14 ± 0.02) - (4.2 ± 1.2). The significant deviations from a slope of unity and the significant non-zero
intercepts were further investigated using non-linear regression. Quadratic regression gave: LIAISON = RIA2 ± (-0.0025 ±
0.0005) + RIA±(1.211 ± 0.07) + (2.9 ± 2.5). The intercept was not significantly different from 0. The quadratic equation
significantly decreased the residual sum of squares (P<0.0001) indicating this model better described the relationship.
Non-linearity was apparent at RIA 25-hydroxyvitamin D ≥ 110 nmol/L, where the relationship was described by:
LIAISON = 97.5339 + 0.1388±RIA (r2 = 0.0039; N.S.). However, removing points RIA ≥ 110 nmol/L did not substantially
alter the regression parameters. Comparing the analytical imprecision with the total random regression error (Sy/x)
suggested that sample-related effects were not present. It is recommended that cross-over analysis between these two
methods include points from all parts of the range of interest to elucidate the complete nature of the relationship.