ELISA BETTER THAN WE THOUGHT Dr Direk Limmathurotsakul, MD MSc PhD.

ELISA BETTER THAN WE THOUGHT

Dr Direk Limmathurotsakul, MD MSc PhD

Introduction: problems and solutions of imperfect Gold Standard

Introduction: culture is an imperfect gold standard for melioidosis

Introduction: culture is an imperfect gold standard for melioidosis

Parameters

Culture as a gold standard

Final Bayesian LCM

Prevalence 37 % 62 %

Culture

Sensitivity

100 % 60 %

Specificity

100 % 100 %

ELISA

Sensitivity

82 % 76 %

Specificity

73 % 98 %Limmathurotsakul et al (2010) PLoS ONE 5(8) e12485

Next Step: What is the proper cutoff for ELISA

• LCM estimate that true Se and Sp of ELISA were 76% and 98%

• Cut-off used was previously determined by conventional ROC

Next Step: What is the proper cutoff for ELISA

METHOD

• Bayesian latent class models (LCM) were applied to all possible cut-off values

• Sensitivity and specificity estimated from each cut-off value was used to plot unbiased ROC curves

Parameters

Bayesian LCM using biased cut-

off

Bayesian LCM using optimal

cut-off

ELISA

Se 76 % 81 %

Sp 98 % 96 %

RESULT

SUMMARY

• Cut-off determined by conventional method was biased towards misclassification of imperfect gold standard

• LCMs should be used to determine optimal cut-off

• ELISA could be furthered developed for use in the clinical setting with a reasonably high degree of accuracy

• Further evaluation of new diagnostic tests for melioidosis should be done with a carefully selected set of diagnostic tests and appropriate statistical models

How can general researchers apply Bayesian LCM to their own datasets

http://mice.tropmedres.ac

More information: PPI-37 (WMC2013)

END