INR & Regression Guidelines
INR Derivation
INR values are computed automatically from each pair of PT measurements. Each method uses its own ISI and Normal Patient Mean (established via the Geometric Mean & VRI module for that reagent lot):
INR — Reference Method (X)
INR_X = (PT_X / NPM_X)^ISI_X
Where PT_X is the reference method PT value in seconds, NPM_X is the Normal Patient Mean established for the reference method's current reagent lot, and ISI_X is the ISI from the reference reagent's package insert.
INR — Test Method (Y)
INR_Y = (PT_Y / NPM_Y)^ISI_Y
Where PT_Y is the test method PT value in seconds, NPM_Y is the Normal Patient Mean established for the test method's current reagent lot, and ISI_Y is the ISI from the test reagent's package insert.
This means that each method's INR is anchored to its own NPM — differences in INR results between methods can arise from differences in raw PT values, differences in ISI, or differences in NPM. The PT analysis isolates the raw instrument/reagent performance, while the INR analysis reflects the clinically reported values including all calibration factors.
Regression Analysis
Three regression methods are computed for both PT and INR. All three are always calculated and displayed; the regression type selected in the configuration is used for Medical Decision Point calculations when MDPs are enabled.
Deming Regression (recommended)
Y = slope × X + intercept (with error in both X and Y)
Accounts for measurement error in both methods — the most statistically appropriate regression for comparing two imprecise measurement procedures. Uses jackknife resampling to compute confidence intervals for slope and intercept. Recommended by CLSI EP09C as the primary regression method when both methods have measurable imprecision.
Passing-Bablok Regression
Y = slope × X + intercept (non-parametric)
A non-parametric regression that makes no distributional assumptions. Robust to outliers and to non-normal data. Uses bootstrap resampling for confidence intervals. Useful when the data contain outliers or when the error structure is unknown. Slope and intercept estimates are based on the median of all pairwise slopes.
Ordinary Least Squares (OLS)
Y = slope × X + intercept (error in Y only)
The classical linear regression that assumes all measurement error is in Y (the test method) and X is measured without error. Appropriate only when the reference method (X) has negligible imprecision relative to the test method. Generally not recommended for method comparison unless the reference method is a true gold standard with known negligible error.
The Analysis Summary table shows slope, intercept, and 95% confidence intervals for each, along with Standard Error of the Estimate (SEE), sample count (N), and Correlation Coefficient (r) — reported for both PT and INR side by side.