Intrathecal antibody synthesis - Reiber method revisited

CSF immunoglobulin (IgG) often informs diagnosis in CNS inflammatory disease- qualitatively (oligoclonal bands; OCBs), or quantitatively (CSF: serum IgG: albumin, e.g. IgG index). Most US laboratories estimate excess CSF IgG using linear models, most in Europe, Reiber's hyperbolic model - the latter often also incorporated into CSF pathogen-specific immunoreactivity calculations. We compared Reiber hyperbolic and linear approaches.

Required data were available for 1,872 samples, Lyme-specific data for 1,365. We compared linear and Reiber's hyperbolic models to determine which better describes normal CSF: serum IgG/albumin, assessed correlations with OCBs using ROC analysis and compared Lyme-specific antibody indices (AI) by both methods.

On linear regression (CSF: serum albumin vs. CSF: serum IgG) in 890 normals, r 0.98; 0.86 on polynomial. In 982 abnormals r was 0.95 by both linear and polynomial regression. Linear regression slope was 74% greater in abnormals than normals. We tested the linearity of the Reiber hyperbolic model and found it to be essentially linear in the clinically relevant range. Dichotomizing the full dataset by OCBs, IgG index ROC AUC was 0.81, Reiber 0.74 (p<0.001). Linear models consistently demonstrated stronger associations with OCBs than the Reiber method. IgG index and IgG synthesis cutoffs from ROC analyses differed substantially from, and were more robust than, customary ones. On Lyme AI analysis, Reiber modification was potentially applicable in just 14.7% of samples, possibly changing the diagnostic conclusion in just 0.1%.

Linear models more accurately reflect intra-CNS inflammation, including a novel linear regression model. Notably, we found that, despite its equation's complexity, the Reiber hyperbolic model is also linear in the clinically relevant range. In CNS inflammation, local antibody production and altered BBB/BCB permeability both contribute to excess CSF IgG. ROC-derived cutoffs for IgG index and IgG synthesis more accurately reflect CSF immunoreactivity. Reiber Q_Limit substitution in the Lyme-specific AI denominator did not improve accuracy.