A comparison of Bayesian and maximum likelihood methods to determine the performance of a point of care test for Helicobacter pylori in the office setting

Abstract

Objective: Evaluations of point of care tests (PCT) are often hampered by a lack of appropriate gold standards. This study aimed to compare the results of a Bayesian statistical analysis and a maximum likelihood method to evaluate the performance of a PCT for Helicobacter pylori in primary care.

Methods: The Helisal Rapid Blood Test (Cortecs Diagnostics) was performed in 311 patients from 6 primary care centers, and a concurrent venous sample was taken for 2 enzyme-linked immunosorbent assays (ELISA) performed at the laboratory, blind to the PCT result. The Bayesian analysis was conducted using Markov Chain Monte Carlo methods (WinBUGS). The performance characteristics of the PCT and the 2 ELISA tests were estimated together with 95% credible intervals (95% CIs).

Results: The estimate of prevalence of H. pylori in this population was 64% (95% CI, 59% to 70%), the sensitivity and specificity of the PCT were 89% (84% to 94%) and 84% (77% to 91%), respectively (likelihood ratios positive 5.6, negative 0.13). The equivalent maximum likelihood results were prevalence, 65%; sensitivity, 90%; and specificity, 83%.

Conclusions: The Helisal Rapid Blood Test performed as well as laboratory-based ELISA tests in this cohort of patients. The Bayesian analysis and the maximum likelihood method gave similar results, the Bayesian method also simultaneously estimating 95% CIs.