Sensitivity And Specificity Of Tests For Trachoma In The Absence Of A Gold Standard

Abstract

 

Purpose:

 

Ocular chlamydial infection is difficult to diagnose. WHO treatment protocols are based on a clinical exam but research studies often rely on laboratory tests such as PCR. Sensitivity and specificity of these tests are not well characterized due to the lack of a gold standard test which is 100% sensitive and 100% specific. Here, we characterize multiple diagnostic tests using a Latent Class Analysis (LCA). We also perform a linear regression that uses clinical exam to estimate PCR prevalence.

 

Methods:

 

The Trachoma Elimination Follow-up (TEF) study was a cluster-randomized clinical trial comparing treatment strategies in Ethiopia. Pre-treatment data collection yielded 2115 individuals in 40 villages with WHO simplified clinical exam and DNA-PCR tests. LCA is a method that allows comparisons of each test against a latent gold standard that acts as a composite of all available data. We analyzed TEF data with a clustered LCA to estimate sensitivity and specificity of: clinical exam TF, clinical exam TI, and DNA-PCR (Amplicor, Roche Inc.). Confidence intervals were obtained by 1000 bootstrap resamples at the village level. A linear regression modeling PCR prevalence rate using TF, TI, and TF/TI prevalence rates was fit and underwent model selection using Akaike Information Criterion.

 

Results:

 

Median TF prevalence by village was 73.5% (interquartile range (IQR) 67.2 - 83.3%), median TI prevalence was 29.8% (IQR 21.1 - 42.9%), and median PCR positive rate was 46.9% (IQR 34.1 - 60.5%). Estimates and for sensitivity and specificity are in the table. The linear regression and model selection yielded the formula (PCR) = 0.37(TF) + 0.57(TI), where the terms in parentheses are prevalences.  

 

Conclusions:

 

TF is sensitive but nonspecific, TI lacks sensitivity but is specific, and PCR is close to being a true gold standard. In estimating PCR prevalence, TF and TI appear to be independent predictors. The equation above may be used to estimate PCR prevalence from clinical exam data.

 

Keywords: trachoma • clinical (human) or epidemiologic studies: biostatistics/epidemiology methodology • clinical (human) or epidemiologic studies: prevalence/incidence