Proceedings of the 2013 annual meeting of the Netherlands Epidemiology Society
Volume 1 Issue S1 Abstract 3
M. van Smeden, Julius Center, University Medical Center, Utrecht
J.A.H. de Groot, Julius Center, University Medical Center, Utrecht
C.A. Naaktgeboren, Julius Center, University Medical Center, Utrecht
K.G.M. Moons, Julius Center, University Medical Center, Utrecht
J.B. Reitsma, Julius Center, University Medical Center, Utrecht
Latent class models (LCMs) combine the results of multiple diagnostics tests through a statistical model to obtain estimates of diagnostic accuracy in situations where there is no single, accepted reference test (i.e., no gold standard). To explore the methodology and reporting of LCMs in diagnostic research, we performed a systematic review of such studies. Our review generates insight into the quality of reporting of studies that use LCMs and reveals variation in methodology between studies.
We identified 64 diagnostic studies that reported test accuracy or disease prevalence estimates obtained using LCM parameter estimates. From these studies, information was extracted independently by two reviewers.
The systematic review shows that the use of LCMs in diagnostic studies is increasing over time, notably in studies which evaluate the accuracy of diagnostic tests to detect the presence of an infectious disease (52%). The majority of studies (64%) reported analyses solely based on ‘basic’ 2-class LCMs, for which it is assumed that the tests are independent conditional on a binary target disease status.
Our review revealed several shortcomings in methodology and reporting of studies which use LCMs. The majority of the inferences made were based on ‘basic’ LCMs for which the assumptions are easily violated. Evaluations of the tenability of these assumptions, although thoroughly described in statistical literature, are rarely reported.
Published: 06 Jun, 2013