A two-level copula joint model for joint analysis of longitudinal and competing risks data.

In this article, we propose a two-level copula joint model to analyze clinical data with multiple disparate continuous longitudinal outcomes and multiple event-times in the presence of competing risks. At the first level, we use a copula to model the dependence between competing latent event-times, in the process constructing the submodel for the observed event-time, and employ the Gaussian copula to construct the submodel for the longitudinal outcomes that accounts for their conditional dependence; these submodels are glued together at the second level via the Gaussian copula to construct a joint model that incorporates conditional dependence between the observed event-time and the longitudinal outcomes. To have the flexibility to accommodate skewed data and examine possibly different covariate effects on quantiles of a non-Gaussian outcome, we propose linear quantile mixed models for the continuous longitudinal data. We adopt a Bayesian framework for model estimation and inference via Markov Chain Monte Carlo sampling. We examine the performance of the copula joint model through a simulation study and show that our proposed method outperforms the conventional approach assuming conditional independence with smaller biases and better coverage probabilities of the Bayesian credible intervals. Finally, we carry out an analysis of clinical data on renal transplantation for illustration.

Estimating sensitivity and specificity of diagnostic tests using latent class models that account for conditional dependence between tests: a simulation study.

BACKGROUND: Latent class models are increasingly used to estimate the sensitivity and specificity of diagnostic tests in the absence of a gold standard, and are commonly fitted using Bayesian methods. These models allow us to account for 'conditional dependence' between two or more diagnostic tests, meaning that the results from tests are correlated even after conditioning on the person's true disease status. The challenge is that it is not always clear to researchers whether conditional dependence exists between tests and whether it exists in all or just some latent classes. Despite the increasingly widespread use of latent class models to estimate diagnostic test accuracy, the impact of the conditional dependence structure chosen on the estimates of sensitivity and specificity remains poorly investigated. METHODS: A simulation study and a reanalysis of a published case study are used to highlight the impact of the conditional dependence structure chosen on estimates of sensitivity and specificity. We describe and implement three latent class random-effect models with differing conditional dependence structures, as well as a conditional independence model and a model that assumes perfect test accuracy. We assess the bias and coverage of each model in estimating sensitivity and specificity across different data generating mechanisms. RESULTS: The findings highlight that assuming conditional independence between tests within a latent class, where conditional dependence exists, results in biased estimates of sensitivity and specificity and poor coverage. The simulations also reiterate the substantial bias in estimates of sensitivity and specificity when incorrectly assuming a reference test is perfect. The motivating example of tests for Melioidosis highlights these biases in practice with important differences found in estimated test accuracy under different model choices. CONCLUSIONS: We have illustrated that misspecification of the conditional dependence structure leads to biased estimates of sensitivity and specificity when there is a correlation between tests. Due to the minimal loss in precision seen by using a more general model, we recommend accounting for conditional dependence even if researchers are unsure of its presence or it is only expected at minimal levels.

Evaluation using latent class models of the diagnostic performances of three ELISA tests commercialized for the serological diagnosis of Coxiella burnetii infection in domestic ruminants.

ELISA methods are the diagnostic tools recommended for the serological diagnosis of Coxiella burnetii infection in ruminants but their respective diagnostic performances are difficult to assess because of the absence of a gold standard. This study focused on three commercial ELISA tests with the following objectives (1) assess their sensitivity and specificity in sheep, goats and cattle, (2) assess the between- and within-herd seroprevalence distribution in these species, accounting for diagnostic errors, and (3) estimate optimal sample sizes considering sensitivity and specificity at herd level. We comparatively tested 1413 cattle, 1474 goat and 1432 sheep serum samples collected in France. We analyzed the cross-classified test results with a hierarchical zero-inflated beta-binomial latent class model considering each herd as a population and conditional dependence as a fixed effect. Potential biases and coverage probabilities of the model were assessed by simulation. Conditional dependence for truly seropositive animals was high in all species for two of the three ELISA methods. Specificity estimates were high, ranging from 94.8% [92.1; 97.8] to 99.2% [98.5; 99.7], whereas sensitivity estimates were generally low, ranging from 39.3 [30.7; 47.0] to 90.5% [83.3; 93.8]. Between- and within-herd seroprevalence estimates varied greatly among geographic areas and herds. Overall, goats showed higher within-herd seroprevalence levels than sheep and cattle. The optimal sample size maximizing both herd sensitivity and herd specificity varied from 3 to at least 20 animals depending on the test and ruminant species. This study provides better interpretation of three widely used commercial ELISA tests and will make it possible to optimize their implementation in future studies. The methodology developed may likewise be applied to other human or animal diseases.

Memory and Markov Blankets.

In theoretical biology, we are often interested in random dynamical systems-like the brain-that appear to model their environments. This can be formalized by appealing to the existence of a (possibly non-equilibrium) steady state, whose density preserves a conditional independence between a biological entity and its surroundings. From this perspective, the conditioning set, or Markov blanket, induces a form of vicarious synchrony between creature and world-as if one were modelling the other. However, this results in an apparent paradox. If all conditional dependencies between a system and its surroundings depend upon the blanket, how do we account for the mnemonic capacity of living systems? It might appear that any shared dependence upon past blanket states violates the independence condition, as the variables on either side of the blanket now share information not available from the current blanket state. This paper aims to resolve this paradox, and to demonstrate that conditional independence does not preclude memory. Our argument rests upon drawing a distinction between the dependencies implied by a steady state density, and the density dynamics of the system conditioned upon its configuration at a previous time. The interesting question then becomes: What determines the length of time required for a stochastic system to 'forget' its initial conditions? We explore this question for an example system, whose steady state density possesses a Markov blanket, through simple numerical analyses. We conclude with a discussion of the relevance for memory in cognitive systems like us.

Inferring directional relationships in microbial communities using signed Bayesian networks.

BACKGROUND: Microbe-microbe and host-microbe interactions in a microbiome play a vital role in both health and disease. However, the structure of the microbial community and the colonization patterns are highly complex to infer even under controlled wet laboratory conditions. In this study, we investigate what information, if any, can be provided by a Bayesian Network (BN) about a microbial community. Unlike the previously proposed Co-occurrence Networks (CoNs), BNs are based on conditional dependencies and can help in revealing complex associations. RESULTS: In this paper, we propose a way of combining a BN and a CoN to construct a signed Bayesian Network (sBN). We report a surprising association between directed edges in signed BNs and known colonization orders. CONCLUSIONS: BNs are powerful tools for community analysis and extracting influences and colonization patterns, even though the analysis only uses an abundance matrix with no temporal information. We conclude that directed edges in sBNs when combined with negative correlations are consistent with and strongly suggestive of colonization order.

A Projection-based Conditional Dependence Measure with Applications to High-dimensional Undirected Graphical Models.

Measuring conditional dependence is an important topic in econometrics with broad applications including graphical models. Under a factor model setting, a new conditional dependence measure based on projection is proposed. The corresponding conditional independence test is developed with the asymptotic null distribution unveiled where the number of factors could be high-dimensional. It is also shown that the new test has control over the asymptotic type I error and can be calculated efficiently. A generic method for building dependency graphs without Gaussian assumption using the new test is elaborated. We show the superiority of the new method, implemented in the R package pgraph, through simulation and real data studies.

A note on the interpretation of tree-based regression models.

Tree-based models are a popular tool for predicting a response given a set of explanatory variables when the regression function is characterized by a certain degree of complexity. Sometimes, they are also used to identify important variables and for variable selection. We show that if the generating model contains chains of direct and indirect effects, then the typical variable importance measures suggest selecting as important mainly the background variables, which have a strong indirect effect, disregarding the variables that directly influence the response. This is attributable mainly to the variable choice in the first steps of the algorithm selecting the splitting variable and to the greedy nature of such search. This pitfall could be relevant when using tree-based algorithms for understanding the underlying generating process, for population segmentation and for causal inference.

Bayesian Network Resource for Meta-Analysis: Cellular Toxicity of Quantum Dots.

A web-based resource for meta-analysis of nanomaterials toxicity is developed whereby the utility of Bayesian networks (BNs) is illustrated for exploring the cellular toxicity of Cd-containing quantum dots (QDs). BN models are developed based on a dataset compiled from 517 publications comprising 3028 cell viability data samples and 837 IC50 values. BN QD toxicity (BN-QDTox) models are developed using both continuous (i.e., numerical) and categorical attributes. Using these models, the most relevant attributes identified for correlating IC50 are: QD diameter, exposure time, surface ligand, shell, assay type, surface modification, and surface charge, with the addition of QD concentration for the cell viability analysis. Data exploration via BN models further enables identification of possible association rules for QDs cellular toxicity. The BN models as web-based applications can be used for rapid intelligent query of the available body of evidence for a given nanomaterial and can be readily updated as the body of knowledge expands.

Social Complexity and Brain Evolution: Comparative Analysis of Modularity and Integration in Ant Brain Organization.

The behavioral demands of living in social groups have been linked to the evolution of brain size and structure, but how social organization shapes investment and connectivity within and among functionally specialized brain regions remains unclear. To understand the influence of sociality on brain evolution in ants, a premier clade of eusocial insects, we statistically analyzed patterns of brain region size covariation as a proxy for brain region connectivity. We investigated brain structure covariance in young and old workers of two formicine ants, the Australasian weaver ant Oecophylla smaragdina, a pinnacle of social complexity in insects, and its socially basic sister clade Formica subsericea. As previously identified in other ant species, we predicted that our analysis would recognize in both species an olfaction-related brain module underpinning social information processing in the brain, and a second neuroanatomical cluster involved in nonolfactory sensorimotor processes, thus reflecting conservation of compartmental connectivity. Furthermore, we hypothesized that covariance patterns would reflect divergence in social organization and life histories either within this species pair or compared to other ant species. Contrary to our predictions, our covariance analyses revealed a weakly defined visual, rather than olfactory, sensory processing cluster in both species. This pattern may be linked to the reliance on vision for worker behavioral performance outside of the nest and the correlated expansion of the optic lobes to meet navigational demands in both species. Additionally, we found that colony size and social organization, key measures of social complexity, were only weakly correlated with brain modularity in these formicine ants. Worker age also contributed to variance in brain organization, though in different ways in each species. These findings suggest that brain organization may be shaped by the divergent life histories of the two study species. We compare our findings with patterns of brain organization of other eusocial insects.

Estimation of diagnostic accuracy of a combination of continuous biomarkers allowing for conditional dependence between the biomarkers and the imperfect reference-test.

Estimating biomarker-index accuracy when only imperfect reference-test information is available is usually performed under the assumption of conditional independence between the biomarker and imperfect reference-test values. We propose to define a latent normally-distributed tolerance-variable underlying the observed dichotomous imperfect reference-test results. Subsequently, we construct a Bayesian latent-class model based on the joint multivariate normal distribution of the latent tolerance and biomarker values, conditional on latent true disease status, which allows accounting for conditional dependence. The accuracy of the continuous biomarker-index is quantified by the AUC of the optimal linear biomarker-combination. Model performance is evaluated by using a simulation study and two sets of data of Alzheimer's disease patients (one from the memory-clinic-based Amsterdam Dementia Cohort and one from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database). Simulation results indicate adequate model performance and bias in estimates of the diagnostic-accuracy measures when the assumption of conditional independence is used when, in fact, it is incorrect. In the considered case studies, conditional dependence between some of the biomarkers and the imperfect reference-test is detected. However, making the conditional independence assumption does not lead to any marked differences in the estimates of diagnostic accuracy.

Understanding the effects of conditional dependence in research studies involving imperfect diagnostic tests.

When two imperfect diagnostic tests are carried out on the same subject, their results may be correlated even after conditioning on the true disease status. While past work has focused on the consequences of ignoring conditional dependence, the degree to which conditional dependence can be induced has not been systematically studied. We examine this issue in detail by introducing a hypothetical missing covariate that affects the sensitivities of two imperfect dichotomous tests. We consider four forms for this covariate, normal, uniform, dichotomous and trichotomous. In the case of a dichotomous covariate, we derive an expression showing that the conditional covariance is a function of the product of the changes in test sensitivities (or specificities) between the subgroups defined by the covariate. The maximum possible covariance is induced by a dichotomous covariate with a very strong effect on both tests. Through simulations, we evaluate the extent to which fitting a latent class model ignoring each type of covariate but including a general covariance term can adjust for the correlation induced by the covariate. We compare the results to when the conditional dependence is ignored. We find that the bias because of ignoring conditional dependence is generally small even for moderate covariate effects, and when bias is present, a model including a covariance term works well. We illustrate our methods by analyzing data from a childhood tuberculosis study. Copyright © 2016 John Wiley & Sons, Ltd.

Modeling conditional dependence among multiple diagnostic tests.

When multiple imperfect dichotomous diagnostic tests are applied to an individual, it is possible that some or all of their results remain dependent even after conditioning on the true disease status. The estimates could be biased if this conditional dependence is ignored when using the test results to infer about the prevalence of a disease or the accuracies of the diagnostic tests. However, statistical methods correcting for this bias by modelling higher-order conditional dependence terms between multiple diagnostic tests are not well addressed in the literature. This paper extends a Bayesian fixed effects model for 2 diagnostic tests with pairwise correlation to cases with 3 or more diagnostic tests with higher order correlations. Simulation results show that the proposed fixed effects model works well both in the case when the tests are highly correlated and in the case when the tests are truly conditionally independent, provided adequate external information is available in the form of fixed constraints or prior distributions. A data set on the diagnosis of childhood pulmonary tuberculosis is used to illustrate the proposed model.

Bias due to composite reference standards in diagnostic accuracy studies.

Composite reference standards (CRSs) have been advocated in diagnostic accuracy studies in the absence of a perfect reference standard. The rationale is that combining results of multiple imperfect tests leads to a more accurate reference than any one test in isolation. Focusing on a CRS that classifies subjects as disease positive if at least one component test is positive, we derive algebraic expressions for sensitivity and specificity of this CRS, sensitivity and specificity of a new (index) test compared with this CRS, as well as the CRS-based prevalence. We use as a motivating example the problem of evaluating a new test for Chlamydia trachomatis, an asymptomatic disease for which no gold-standard test exists. As the number of component tests increases, sensitivity of this CRS increases at the expense specificity, unless all tests have perfect specificity. Therefore, such a CRS can lead to significantly biased accuracy estimates of the index test. The bias depends on disease prevalence and accuracy of the CRS. Further, conditional dependence between the CRS and index test can lead to over-estimation of index test accuracy estimates. This commonly-used CRS combines results from multiple imperfect tests in a way that ignores information and therefore is not guaranteed to improve over a single imperfect reference unless each component test has perfect specificity, and the CRS is conditionally independent of the index test. When these conditions are not met, as in the case of C. trachomatis testing, more realistic statistical models should be researched instead of relying on such CRSs.

A Recent Development of a Network Approach to Assessment Data: Latent Space Item Response Modeling for Intelligence Studies.

This article aims to provide an overview of the potential advantages and utilities of the recently proposed Latent Space Item Response Model (LSIRM) in the context of intelligence studies. The LSIRM integrates the traditional Rasch IRT model for psychometric data with the latent space model for network data. The model has person-wise latent abilities and item difficulty parameters, capturing the main person and item effects, akin to the Rasch model. However, it additionally assumes that persons and items can be mapped onto the same metric space called a latent space and distances between persons and items represent further decreases in response accuracy uncaptured by the main model parameters. In this way, the model can account for conditional dependence or interactions between persons and items unexplained by the Rasch model. With two empirical datasets, we illustrate that (1) the latent space can provide information on respondents and items that cannot be captured by the Rasch model, (2) the LSIRM can quantify and visualize potential between-person variations in item difficulty, (3) latent dimensions/clusters of persons and items can be detected or extracted based on their latent positions on the map, and (4) personalized feedback can be generated from person-item distances. We conclude with discussions related to the latent space modeling integrated with other psychometric models and potential future directions.

Conditional Dependence across Slow and Fast Item Responses: With a Latent Space Item Response Modeling Approach.

There recently have been many studies examining conditional dependence between response accuracy and response times in cognitive tests. While most previous research has focused on revealing a general pattern of conditional dependence for all respondents and items, it is plausible that the pattern may vary across respondents and items. In this paper, we attend to its potential heterogeneity and examine the item and person specificities involved in the conditional dependence between item responses and response times. To this end, we use a latent space item response theory (LSIRT) approach with an interaction map that visualizes conditional dependence in response data in the form of item-respondent interactions. We incorporate response time information into the interaction map by applying LSIRT models to slow and fast item responses. Through empirical illustrations with three cognitive test datasets, we confirm the presence and patterns of conditional dependence between item responses and response times, a result consistent with previous studies. Our results further illustrate the heterogeneity in the conditional dependence across respondents, which provides insights into understanding individuals' underlying item-solving processes in cognitive tests. Some practical implications of the results and the use of interaction maps in cognitive tests are discussed.

Mixed cumulative probit: a multivariate generalization of transition analysis that accommodates variation in the shape, spread and structure of data.

Biological data are frequently nonlinear, heteroscedastic and conditionally dependent, and often researchers deal with missing data. To account for characteristics common in biological data in one algorithm, we developed the mixed cumulative probit (MCP), a novel latent trait model that is a formal generalization of the cumulative probit model usually used in transition analysis. Specifically, the MCP accommodates heteroscedasticity, mixtures of ordinal and continuous variables, missing values, conditional dependence and alternative specifications of the mean response and noise response. Cross-validation selects the best model parameters (mean response and the noise response for simple models, as well as conditional dependence for multivariate models), and the Kullback-Leibler divergence evaluates information gain during posterior inference to quantify mis-specified models (conditionally dependent versus conditionally independent). Two continuous and four ordinal skeletal and dental variables collected from 1296 individuals (aged birth to 22 years) from the Subadult Virtual Anthropology Database are used to introduce and demonstrate the algorithm. In addition to describing the features of the MCP, we provide material to help fit novel datasets using the MCP. The flexible, general formulation with model selection provides a process to robustly identify the modelling assumptions that are best suited for the data at hand.

Factor Tree Copula Models for Item Response Data.

Factor copula models for item response data are more interpretable and fit better than (truncated) vine copula models when dependence can be explained through latent variables, but are not robust to violations of conditional independence. To circumvent these issues, truncated vines and factor copula models for item response data are joined to define a combined model, the so-called factor tree copula model, with individual benefits from each of the two approaches. Rather than adding factors and causing computational problems and difficulties in interpretation and identification, a truncated vine structure is assumed on the residuals conditional on one or two latent variables. This structure can be better explained as a conditional dependence given a few interpretable latent variables. On the one hand, the parsimonious feature of factor models remains intact and any residual dependencies are being taken into account on the other. We discuss estimation along with model selection. In particular, we propose model selection algorithms to choose a plausible factor tree copula model to capture the (residual) dependencies among the item responses. Our general methodology is demonstrated with an extensive simulation study and illustrated by analyzing Post-Traumatic Stress Disorder.

A Latent Space Diffusion Item Response Theory Model to Explore Conditional Dependence between Responses and Response Times.

Traditional measurement models assume that all item responses correlate with each other only through their underlying latent variables. This conditional independence assumption has been extended in joint models of responses and response times (RTs), implying that an item has the same item characteristics fors all respondents regardless of levels of latent ability/trait and speed. However, previous studies have shown that this assumption is violated in various types of tests and questionnaires and there are substantial interactions between respondents and items that cannot be captured by person- and item-effect parameters in psychometric models with the conditional independence assumption. To study the existence and potential cognitive sources of conditional dependence and utilize it to extract diagnostic information for respondents and items, we propose a diffusion item response theory model integrated with the latent space of variations in information processing rate of within-individual measurement processes. Respondents and items are mapped onto the latent space, and their distances represent conditional dependence and unexplained interactions. We provide three empirical applications to illustrate (1) how to use an estimated latent space to inform conditional dependence and its relation to person and item measures, (2) how to derive diagnostic feedback personalized for respondents, and (3) how to validate estimated results with an external measure. We also provide a simulation study to support that the proposed approach can accurately recover its parameters and detect conditional dependence underlying data.

An Iterative, Frequentist Approach for Latent Class Analysis to Evaluate Conditionally Dependent Diagnostic Tests.

Latent class analysis is a well-established method in human and veterinary medicine for evaluating the accuracy of diagnostic tests without a gold standard. An important assumption of this procedure is the conditional independence of the tests. If tests with the same biological principle are used, this assumption is no longer met. Therefore, the model has to be adapted so that the dependencies between the tests can be considered. Our approach extends the traditional latent class model with a term for the conditional dependency of the tests. This extension increases the number of parameters to be estimated and leads to negative degrees of freedom of the model, meaning that not enough information is contained in the existing data to obtain a unique estimate. As a result, there is no clear solution. Hence, an iterative algorithm was developed to keep the number of parameters to be estimated small. Given adequate starting values, our approach first estimates the conditional dependencies and then regards the resulting values as fixed to recalculate the test accuracies and the prevalence with the same method used for independent tests. Subsequently, the new values of the test accuracy and prevalence are used to recalculate the terms for the conditional dependencies. These two steps are repeated until the model converges. We simulated five application scenarios based on diagnostic tests used in veterinary medicine. The results suggest that our method and the Bayesian approach produce similar precise results. However, while the presented approach is able to calculate more accurate results than the Bayesian approach if the test accuracies are initially misjudged, the estimates of the Bayesian method are more precise when incorrect dependencies are assumed. This finding shows that our approach is a useful addition to the existing Bayesian methods, while it has the advantage of allowing simpler and more objective estimations.

Joint Microbial and Metabolomic Network Estimation with the Censored Gaussian Graphical Model.

Joint analysis of microbiome and metabolomic data represents an imperative objective as the field moves beyond basic microbiome association studies and turns towards mechanistic and translational investigations. We present a censored Gaussian graphical model framework, where the metabolomic data are treated as continuous and the microbiome data as censored at zero, to identify direct interactions (defined as conditional dependence relationships) between microbial species and metabolites. Simulated examples show that our method metaMint performs favorably compared to the existing ones. metaMint also provides interpretable microbe-metabolite interactions when applied to a bacterial vaginosis data set. R implementation of metaMint is available on GitHub.