Calibrating multiplex serology for Helicobacter pylori.

Area of Science:

• Epidemiology
• Infectious Diseases
• Biostatistics

Background:

• Helicobacter pylori (H. pylori) is a primary risk factor for gastric cancer, linked to 89% of non-cardia cases globally.
• Prospective studies are crucial for accurately assessing the H. pylori-gastric cancer association, avoiding false negatives from declining antibody levels.

Purpose of the Study:

• To optimize the translation of continuous multiplex serology measurements into H. pylori infection probabilities.
• To evaluate the performance of various classification algorithms in this calibration process.

Main Methods:

• Utilized a large-scale prospective study from the China Kadoorie Biobank.
• Applied classification algorithms including BART, logistic regression, random forest, and elastic net to multiplex serology data.
• Calibrated models using immunoblot as a reference and compared predictive performance.

Main Results:

• All classification models demonstrated high discriminative ability, with Area Under the Curve (AUC) estimates of at least 95% on both training and test datasets.
• Consistent performance was observed between training and test data, indicating robust model generalization.

Conclusions:

• Classification algorithms effectively calibrate H. pylori multiplex serology tests against the established immunoblot method within the China Kadoorie Biobank cohort.
• This research enhances understanding of applying classification algorithms to serological test data for improved H. pylori diagnostics.