Latent class models in diagnostic studies when there is no reference standard--a systematic review

Area of Science:

• Biostatistics
• Epidemiology
• Medical Diagnostics

Background:

• Latent class models (LCMs) are statistical tools used to estimate disease prevalence and diagnostic test accuracy when a definitive reference standard is unavailable.
• The application of LCMs in diagnostic accuracy studies has seen a significant rise over the last decade, particularly in infectious disease research.

Purpose of the Study:

• To systematically review the methodology and reporting practices of latent class models in diagnostic accuracy studies.
• To assess the adherence to critical assumptions and the transparency of reporting in studies utilizing LCMs.

Main Methods:

• Conducted a systematic review of 64 diagnostic accuracy studies employing latent class models.
• Analyzed the types of parametric latent variable models used, including Bayesian and frequentist approaches.
• Evaluated the reporting of critical model assumptions, specifically the independence of test observations within classes.

Main Results:

• Latent class models are increasingly prevalent, with 59% of reviewed studies in infectious diseases.
• The independence assumption, crucial for unbiased estimates, was applied in 61% of studies.
• A significant proportion (28%) of studies failed to report information necessary to verify model assumptions or performance, hindering validation.

Conclusions:

• While latent class models are valuable for diagnostic accuracy assessment, inadequate reporting of assumption verification and model fit compromises the interpretability and validity of findings.
• Improved transparency and reporting standards are essential for readers to confidently assess the reliability of LCM-based inferences in diagnostic research.