[LCMATE] Latent class definition question

[Søren Saxmose Nielsen]

Dear Sébastien,

As far as I understand your question, then the latent state (class) your tests identify is the same, irrespective of objective.

The latent state is the "overlap" between "whatever auscultation picks up" and "whatever imaging picks up". 
Because I presume that auscultation includes increased density and fluid in the lungs and imaging does the same, then your latent disease "only" includes bronchopneaumonias with increased density and increased amounts of fluid. 
Any stage of bronchopneumonia that doesn't included increased density and fluid in the bronchia will not be picked up in your system.

So the trick here is that you define what your joint tests can identify. You need to have insight into the pathogenesis to properly define that (and my insight into calf bronchopneumonia is insufficient, so you need to improve my interpretation given above)

This is my take.

Best regards,

Søren S. Nielsen

Søren Saxmose Nielsen
Professor, DVM, PhD, DipECVPH, DVSc

University of Copenhagen
Department of Veterinary and Animal Sciences
Section for Animal Welfare and Disease Control
Grønnegårdsvej 8
1870 Frederiksberg C

DIR +45 35333096
MOB +45 27140858
saxmose at sund.ku.dk
www.ivh.ku.dk


-----Oprindelig meddelelse-----
Fra: lcmate-bounces at lists.uth.gr <lcmate-bounces at lists.uth.gr> På vegne af Buczinski Sébastien
Sendt: 11. februar 2019 12:34
Til: Polychronis Kostoulas <pkost at vet.uth.gr>; lcmate at lists.uth.gr
Emne: [LCMATE] Latent class definition question

Dear Braintrust,
I have several questions and thoughts I want to share with you on interesting questions I get from clinician people and other researchers that have some reluctance with the LCM approach. Most of their concerns can be answered easily but some questions relative to the correct handling of latent status definition are more difficult to be addressed adequately.
Let's imagine I am focusing on bronchopneumonia in calves (one of my research area). I have 2 tests 1 test is for imaging (I), the second test is auscultation of the calves using a computer-aided lung auscultation algorithm (A). I therefore have 2 tests (I and A) that are possibly dependent (or not). These tests have high inter-rater agreement.
Let's imagine I am doing a large cross-sectional survey in feedlot calves where I include 10 000 animals in 1000 different feedlots (huge study).
The definition of I+/I- and A+/A- are fixed. Then I may define 3 different latent classes that are of interest for clinical purpose and rational use of antimicrobial.
1st objective of the study:
Determination of the accuracy of both tests for detection of clinical bronchopneumonia (ie state of disease where infection is active and therefore need to be addressed using antimicrobial) 2nd objective of the study:
Determination of the accuracy to detect any bronchopneumonia lesion (including active lesions, lesions of previous bronchopneumonia (ie still some lung lesion but no active infection/inflammation) 3rd objective of the study:
Determination of the accuracy to detect non-active lesions (ie latent class status 2 -status1) In this hypothetical study using the same test positive/negative definition, my posterior would only be driven by my priors which are including the different definition when gathering information from the literature or from the experts.
However, the data in that hypothetical study would not change (ie I have the same ++ -+ +- and - stratification of my calves). If running a very large study with many data all my models obtained from objective 1, 2 and 3 will converge to the same posterior densities (SeI, Spi, SeA, SpA and prevalence).
How can I therefore be sure that the accuracy results I obtain are for latent class definition 1, 2 or 3? Maybe there is a simple answer but I did not find it until now.
I thank you in advance for any information on that as well as on any specific reference on that fascinating topic.

Thanks a lot

Sébastien