Evaluation of diagnostic test procedures for SARS-CoV-2 using latent class models

Estimating diagnostic accuracy without a gold standard remains a challenge in infectious disease testing. While PCR and antigen tests are widely used for SARS-CoV-2 detection, their real-world performance is often assumed based on lab conditions. Prior research has shown PCR to be highly specific, but its status as a gold standard may introduce bias. No prior work had resolved the independent sensitivity and specificity of these tests in field settings. This gap motivated the use of statistical methods that do not rely on a single reference test. The lack of unbiased estimates affects public health strategies and clinical decision-making. Understanding test performance in real-world conditions is essential for optimizing diagnostic protocols. This study addresses the need for independent evaluation of SARS-CoV-2 testing procedures.

The aim of this study was to estimate the true sensitivity and specificity of PCR and antigen tests for SARS-CoV-2 without assuming PCR as a gold standard. The researchers wanted to evaluate how these tests perform in real-world conditions. They focused on a method that avoids potential biases introduced by using one test to validate another. The motivation came from the need to improve diagnostic accuracy in public health responses. The study sought to provide data that could inform testing strategies during high prevalence periods. The researchers aimed to assess the reliability of antigen tests in detecting SARS-CoV-2. They also wanted to determine whether serial testing is beneficial under different prevalence levels. This approach allows for more accurate interpretation of test results in clinical and public health settings.

The researchers employed latent class models to estimate test performance. These models do not require a gold standard reference test. They used data from Danish national registries to analyze test results. The study population included individuals tested with both PCR and antigen tests. The models accounted for the correlation between test results and disease status. The researchers applied statistical techniques to estimate sensitivity and specificity. They calculated 95% confidence intervals for each estimate. The approach allowed for unbiased evaluation of both testing procedures in the field.

The specificity of both PCR and antigen tests was found to be greater than 99.7%. PCR test sensitivity was estimated at 95.7% (95% CI: 92.8%-98.4%). Antigen test sensitivity was significantly lower at 53.8% (95% CI: 49.8%-57.9%). These results suggest that PCR is more reliable for detecting SARS-CoV-2. The high specificity indicates that both tests rarely produce false positives. The wide confidence interval for antigen test sensitivity reflects variability in real-world conditions. The findings imply that antigen tests may miss a substantial number of cases. The study also found that serial testing may not be beneficial at higher prevalence levels.

The study concludes that PCR and antigen tests have high specificity but differ in sensitivity. The researchers suggest that PCR remains the more accurate test for SARS-CoV-2 detection. The lower sensitivity of antigen tests indicates they may miss a significant proportion of cases. The high specificity supports the use of antigen tests for screening in low-prevalence settings. The findings suggest that confirmatory PCR testing after a positive antigen result may not be necessary in all cases. The study highlights the importance of considering test performance in public health strategies. The results may inform decisions on testing protocols during different stages of an outbreak. The authors propose that test selection should be based on prevalence levels and clinical context.