A predictive model for optimizing kappa free light chain dilution in cerebrospinal fluid samples.

INTRODUCTION

Accurate measurement of kappa free light chains (KFLC) in cerebrospinal fluid (CSF) requires appropriate sample dilution. This study aimed to develop and validate a decision tree model using CSF albumin and immunoglobulin G (IgG) concentrations to predict initial KFLC dilution requirements and reduce repeat determinations.

MATERIALS AND

METHODS

A classification decision tree was developed using 589 retrospective CSF samples analyzed by nephelometry (Atellica NEPH 630, Siemens Healthineers, Germany). The model was developed in JASP (version 0.96.0) to assign samples to three dilution categories: undiluted, 1:5, and 1:20. The dataset was divided into training (80%) and validation (20%) subsets, followed by independent prospective evaluation in 107 samples. Model performance and laboratory resource utilization were assessed, including repeat determinations, estimated direct costs, and technologist hands-on time.

RESULTS

The classification accuracy was 76.6% (95% confidence interval [CI]: 68.9-84.3%) in the validation dataset and 78.5% (95% CI: 69.8-85.2%) in the prospective dataset. Model-guided dilution selection reduced repeat determinations from 56 to 24 (57.1%). Estimated direct reagent and consumable savings were approximately €290 per 100 CSF samples. Exploratory workflow assessment showed that manual dilution selection required an average of 40.4 s per sample.

CONCLUSIONS

A decision tree based on CSF albumin and IgG concentrations may support initial KFLC dilution selection and reduce repeat determinations and associated reagent consumption. However, its net operational benefit requires further evaluation under routine laboratory conditions.