Limits...
Hui and Walter's latent-class model extended to estimate diagnostic test properties from surveillance data: a latent model for latent data.

Bermingham ML, Handel IG, Glass EJ, Woolliams JA, de Clare Bronsvoort BM, McBride SH, Skuce RA, Allen AR, McDowell SW, Bishop SC - Sci Rep (2015)

Bottom Line: The model was applied with and without the modelling of conditional dependence between tests.The utility of the extended model was demonstrated through application to bovine tuberculosis surveillance data from Northern and the Republic of Ireland.Our methodology can aid in the interpretation of disease surveillance data, and the results can potentially refine disease control strategies.

View Article: PubMed Central - PubMed

Affiliation: The Roslin Institute and Royal (Dick) School of Veterinary Studies, University of Edinburgh, Easter Bush, Midlothian, EH25 9RG.

ABSTRACT
Diagnostic test sensitivity and specificity are probabilistic estimates with far reaching implications for disease control, management and genetic studies. In the absence of 'gold standard' tests, traditional Bayesian latent class models may be used to assess diagnostic test accuracies through the comparison of two or more tests performed on the same groups of individuals. The aim of this study was to extend such models to estimate diagnostic test parameters and true cohort-specific prevalence, using disease surveillance data. The traditional Hui-Walter latent class methodology was extended to allow for features seen in such data, including (i) unrecorded data (i.e. data for a second test available only on a subset of the sampled population) and (ii) cohort-specific sensitivities and specificities. The model was applied with and without the modelling of conditional dependence between tests. The utility of the extended model was demonstrated through application to bovine tuberculosis surveillance data from Northern and the Republic of Ireland. Simulation coupled with re-sampling techniques, demonstrated that the extended model has good predictive power to estimate the diagnostic parameters and true herd-level prevalence from surveillance data. Our methodology can aid in the interpretation of disease surveillance data, and the results can potentially refine disease control strategies.

No MeSH data available.


Related in: MedlinePlus

Parameter estimates (with 95% Bayesian credibility intervals) of prevalence and accuracy (r) of predicting within cohort prevalence from the simulated data sets from the traditional 4 and extended 6 cell conditional independence and dependence models excluding/including outbreak specific sensitivities (OSS).The simulated data for the conditional independence and dependence models were reconstructed using parameter estimates from the Northern Ireland surveillance data and Clegg et al.22 respectively; with true prevalence ranging from 0.05 to 0.80. The broken perpendicular lines on the upper plots represent the simulation input value for true prevalence, and those on the lower plots depict an accuracy of 1. Modelling OSS had little impact on the precision of dataset and within cohort prevalence estimates. The 6 cell extension improved the accuracy of the estimates of dataset and within cohort prevalence in the absence/presence of conditional dependence between the diagnostic tests.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4493568&req=5

f6: Parameter estimates (with 95% Bayesian credibility intervals) of prevalence and accuracy (r) of predicting within cohort prevalence from the simulated data sets from the traditional 4 and extended 6 cell conditional independence and dependence models excluding/including outbreak specific sensitivities (OSS).The simulated data for the conditional independence and dependence models were reconstructed using parameter estimates from the Northern Ireland surveillance data and Clegg et al.22 respectively; with true prevalence ranging from 0.05 to 0.80. The broken perpendicular lines on the upper plots represent the simulation input value for true prevalence, and those on the lower plots depict an accuracy of 1. Modelling OSS had little impact on the precision of dataset and within cohort prevalence estimates. The 6 cell extension improved the accuracy of the estimates of dataset and within cohort prevalence in the absence/presence of conditional dependence between the diagnostic tests.

Mentions: The simulation study demonstrated that the extended six cell conditional latent class model, with and without conditional dependence between the tests, has good predictive power to estimate diagnostic parameters of the SICTT, abattoir inspection and IFN-γ assay from these surveillance data and for the parameter ranges explored. There was no significant bias observed in the estimated sensitivity and specificity of the tests and the 95% credibility intervals of the simulated diagnostic parameters overlapped widely with the posterior distribution of the diagnostic parameter estimates (Table 7,8). In the simulation study, the four cell conditional independence model generated posterior distributions of SICTT sensitivity and true prevalence that were very similar to those obtained from the surveillance data (Table 3); indicating these parameters were biased upwards, in contrast to those from the six cell model. The posterior distribution of abattoir inspection sensitivity was unbiased (Table 7). The simulation study also demonstrated that the extended six cell conditional independence model had strong predictive power to estimate true herd prevalence. The correlation between the input and output estimated herd-specific prevalence was 0.96 (95% CI 0.93–0.98) and 0.99 (95% CI 0.90–1.00) from the conditional independence and dependence models correspondingly. The four cell latent class model gave marginally lower correlations between the input and output estimated prevalence, viz. 0.93 (95% CI 0.88–0.96) and 0.97(95% CI 0.95–0.98) from the conditional independence and dependence models, respectively. The four cell model was less precise with a root mean square prediction error (RMSE) for herd-specific prevalence estimates of 0.073 and 0.150, compared to 0.036 and 0.024 from the six cell conditional independence and dependence models. Furthermore, the six cell conditional independence (dependence) model had greater accuracy compared to the four cell models, 27% (24%) as opposed to 34% (48%) of the prevalence estimates were one RMSE from the regression line, respectively (Figs 2 and 3). The true prevalence levels in the simulated data had marked effects on the estimated parameter values across the different models (detailed results in Figs 4, 5, 6). Modelling individual out-break sensitivities improved the precision of the diagnostic test parameter estimates to some extent. However it was the six cell extension that removed the dependence of the modelled parameters on the prevalence.


Hui and Walter's latent-class model extended to estimate diagnostic test properties from surveillance data: a latent model for latent data.

Bermingham ML, Handel IG, Glass EJ, Woolliams JA, de Clare Bronsvoort BM, McBride SH, Skuce RA, Allen AR, McDowell SW, Bishop SC - Sci Rep (2015)

Parameter estimates (with 95% Bayesian credibility intervals) of prevalence and accuracy (r) of predicting within cohort prevalence from the simulated data sets from the traditional 4 and extended 6 cell conditional independence and dependence models excluding/including outbreak specific sensitivities (OSS).The simulated data for the conditional independence and dependence models were reconstructed using parameter estimates from the Northern Ireland surveillance data and Clegg et al.22 respectively; with true prevalence ranging from 0.05 to 0.80. The broken perpendicular lines on the upper plots represent the simulation input value for true prevalence, and those on the lower plots depict an accuracy of 1. Modelling OSS had little impact on the precision of dataset and within cohort prevalence estimates. The 6 cell extension improved the accuracy of the estimates of dataset and within cohort prevalence in the absence/presence of conditional dependence between the diagnostic tests.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4493568&req=5

f6: Parameter estimates (with 95% Bayesian credibility intervals) of prevalence and accuracy (r) of predicting within cohort prevalence from the simulated data sets from the traditional 4 and extended 6 cell conditional independence and dependence models excluding/including outbreak specific sensitivities (OSS).The simulated data for the conditional independence and dependence models were reconstructed using parameter estimates from the Northern Ireland surveillance data and Clegg et al.22 respectively; with true prevalence ranging from 0.05 to 0.80. The broken perpendicular lines on the upper plots represent the simulation input value for true prevalence, and those on the lower plots depict an accuracy of 1. Modelling OSS had little impact on the precision of dataset and within cohort prevalence estimates. The 6 cell extension improved the accuracy of the estimates of dataset and within cohort prevalence in the absence/presence of conditional dependence between the diagnostic tests.
Mentions: The simulation study demonstrated that the extended six cell conditional latent class model, with and without conditional dependence between the tests, has good predictive power to estimate diagnostic parameters of the SICTT, abattoir inspection and IFN-γ assay from these surveillance data and for the parameter ranges explored. There was no significant bias observed in the estimated sensitivity and specificity of the tests and the 95% credibility intervals of the simulated diagnostic parameters overlapped widely with the posterior distribution of the diagnostic parameter estimates (Table 7,8). In the simulation study, the four cell conditional independence model generated posterior distributions of SICTT sensitivity and true prevalence that were very similar to those obtained from the surveillance data (Table 3); indicating these parameters were biased upwards, in contrast to those from the six cell model. The posterior distribution of abattoir inspection sensitivity was unbiased (Table 7). The simulation study also demonstrated that the extended six cell conditional independence model had strong predictive power to estimate true herd prevalence. The correlation between the input and output estimated herd-specific prevalence was 0.96 (95% CI 0.93–0.98) and 0.99 (95% CI 0.90–1.00) from the conditional independence and dependence models correspondingly. The four cell latent class model gave marginally lower correlations between the input and output estimated prevalence, viz. 0.93 (95% CI 0.88–0.96) and 0.97(95% CI 0.95–0.98) from the conditional independence and dependence models, respectively. The four cell model was less precise with a root mean square prediction error (RMSE) for herd-specific prevalence estimates of 0.073 and 0.150, compared to 0.036 and 0.024 from the six cell conditional independence and dependence models. Furthermore, the six cell conditional independence (dependence) model had greater accuracy compared to the four cell models, 27% (24%) as opposed to 34% (48%) of the prevalence estimates were one RMSE from the regression line, respectively (Figs 2 and 3). The true prevalence levels in the simulated data had marked effects on the estimated parameter values across the different models (detailed results in Figs 4, 5, 6). Modelling individual out-break sensitivities improved the precision of the diagnostic test parameter estimates to some extent. However it was the six cell extension that removed the dependence of the modelled parameters on the prevalence.

Bottom Line: The model was applied with and without the modelling of conditional dependence between tests.The utility of the extended model was demonstrated through application to bovine tuberculosis surveillance data from Northern and the Republic of Ireland.Our methodology can aid in the interpretation of disease surveillance data, and the results can potentially refine disease control strategies.

View Article: PubMed Central - PubMed

Affiliation: The Roslin Institute and Royal (Dick) School of Veterinary Studies, University of Edinburgh, Easter Bush, Midlothian, EH25 9RG.

ABSTRACT
Diagnostic test sensitivity and specificity are probabilistic estimates with far reaching implications for disease control, management and genetic studies. In the absence of 'gold standard' tests, traditional Bayesian latent class models may be used to assess diagnostic test accuracies through the comparison of two or more tests performed on the same groups of individuals. The aim of this study was to extend such models to estimate diagnostic test parameters and true cohort-specific prevalence, using disease surveillance data. The traditional Hui-Walter latent class methodology was extended to allow for features seen in such data, including (i) unrecorded data (i.e. data for a second test available only on a subset of the sampled population) and (ii) cohort-specific sensitivities and specificities. The model was applied with and without the modelling of conditional dependence between tests. The utility of the extended model was demonstrated through application to bovine tuberculosis surveillance data from Northern and the Republic of Ireland. Simulation coupled with re-sampling techniques, demonstrated that the extended model has good predictive power to estimate the diagnostic parameters and true herd-level prevalence from surveillance data. Our methodology can aid in the interpretation of disease surveillance data, and the results can potentially refine disease control strategies.

No MeSH data available.


Related in: MedlinePlus