arXiv subjects

Adrian O'Hagan

Publications and source records attributed to Adrian O'Hagan.

11 recordsLinked to original sources

arXiv2025

A variational Bayes latent class approach for EHR-based patient phenotyping in R

The VBphenoR package for R provides a closed-form variational Bayes approach to patient phenotyping using Electronic Health Records (EHR) data. We implement a variational Bayes Gaussian Mixture Model (GMM) algorithm using closed-form coordinate ascent variational inference (CAVI) to determine the patient phenotype latent class. We then implement a variational Bayes logistic regression, where we determine the probability of the phenotype in the supplied EHR cohort, the shift in biomarkers for patients with the phenotype of interest versus a healthy population and evaluate predictive performance of binary indicator clinical codes and medication codes. The logistic model likelihood applies the latent class from the GMM step to inform the conditional.

stat.CO↗

arXiv2025

A Dirichlet stochastic block model for composition-weighted networks

Network data are observed in various applications where the individual entities of the system interact with or are connected to each other, and often these interactions are defined by their associated strength or importance. Clustering is a common task in network analysis that involves finding groups of nodes displaying similarities in the way they interact with the rest of the network. However, most clustering methods use the strengths of connections between entities in their original form, ignoring the possible differences in the capacities of individual nodes to send or receive edges. This often leads to clustering solutions that are heavily influenced by the nodes' capacities. One way to overcome this is to analyse the strengths of connections in relative rather than absolute terms, expressing each edge weight as a proportion of the sending (or receiving) capacity of the respective node. This, however, induces additional modelling constraints that most existing clustering methods are not designed to handle. In this work we propose a stochastic block model for composition-weighted networks based on direct modelling of compositional weight vectors using a Dirichlet mixture, with the parameters determined by the cluster labels of the sender and the receiver nodes. Inference is implemented via an extension of the classification expectation-maximisation algorithm that uses a working independence assumption, expressing the complete data likelihood of each node of the network as a function of fixed cluster labels of the remaining nodes. A model selection criterion is derived to aid the choice of the number of clusters. The model is validated using simulation studies, and showcased on network data from the Erasmus exchange program and a bike sharing network for the city of London.

stat.ME↗

arXiv2024

Multiplex Dirichlet stochastic block model for clustering multidimensional compositional networks

Network data often represent multiple types of relations, which can also denote exchanged quantities, and are typically encompassed in a weighted multiplex. Such data frequently exhibit clustering structures, however, traditional clustering methods are not well-suited for multiplex networks. Additionally, standard methods treat edge weights in their raw form, potentially biasing clustering towards a node's total weight capacity rather than reflecting cluster-related interaction patterns. To address this, we propose transforming edge weights into a compositional format, enabling the analysis of connection strengths in relative terms and removing the impact of nodes' total weights. We introduce a multiplex Dirichlet stochastic block model designed for multiplex networks with compositional layers. This model accounts for sparse compositional networks and enables joint clustering across different types of interactions. We validate the model through a simulation study and apply it to the international export data from the Food and Agriculture Organization of the United Nations.

stat.ME↗

arXiv2024

Variational Bayes latent class approach for EHR-based phenotyping with large real-world data

We investigate the performance and characteristics of currently available VB and MCMC software to explore the practicability of available approaches and provide guidance for clinical practitioners. Two case studies are used to fully explore the methods covering a variety of real-world data. First, we use the publicly available Pima Indian diabetes data to comprehensively compare VB implementations of logistic regression. Second, a large real-world data set, Optum(TM) EHR with approximately one million diabetes patients extended the analysis to large, highly unbalanced data containing discrete and continuous variables. A Bayesian patient phenotyping composite model incorporating latent class analysis (LCA) and regression was implemented with the second case study. We find that several data characteristics common in clinical data, such as sparsity, significantly affect the posterior accuracy of automatic VB methods compared with conditionally conjugate mean-field methods. We find that for both models, automatic VB approaches require more effort and technical knowledge to set up for accurate posterior estimation and are very sensitive to stopping time compared with closed-form VB methods. Our results indicate that the patient phenotyping composite Bayes model is more easily usable for real-world studies if Monte Carlo is replaced with VB. It can potentially become a uniquely useful tool for decision support, especially for rare diseases where gold-standard biomarker data is sparse but prior knowledge can be used to assist model diagnosis and may suggest when biomarker tests are warranted.

stat.AP↗

arXiv2023

Joint modelling of the body and tail of bivariate data

In situations where both extreme and non-extreme data are of interest, modelling the whole data set accurately is important. In a univariate framework, modelling the bulk and tail of a distribution has been extensively studied before. However, when more than one variable is of concern, models that aim specifically at capturing both regions correctly are scarce in the literature. A dependence model that blends two copulas with different characteristics over the whole range of the data support is proposed. One copula is tailored to the bulk and the other to the tail, with a dynamic weighting function employed to transition smoothly between them. Tail dependence properties are investigated numerically and simulation is used to confirm that the blended model is sufficiently flexible to capture a wide variety of structures. The model is applied to study the dependence between temperature and ozone concentration at two sites in the UK and compared with a single copula fit. The proposed model provides a better, more flexible, fit to the data, and is also capable of capturing complex dependence structures.

stat.ME↗

arXiv2023

Variational Bayes latent class approach for EHR-based phenotyping with large real-world data

Bayesian approaches to clinical analyses for the purposes of patient phenotyping have been limited by the computational challenges associated with applying the Markov-Chain Monte-Carlo (MCMC) approach to large real-world data. Approximate Bayesian inference via optimization of the variational evidence lower bound, often called Variational Bayes (VB), has been successfully demonstrated for other applications. We investigate the performance and characteristics of currently available R and Python VB software for variational Bayesian Latent Class Analysis (LCA) of realistically large real-world observational data. We used a real-world data set, Optum\textsuperscript{TM} electronic health records (EHR), containing pediatric patients with risk indicators for type 2 diabetes mellitus that is a rare form in pediatric patients. The aim of this work is to validate a Bayesian patient phenotyping model for generality and extensibility and crucially that it can be applied to a realistically large real-world clinical data set. We find currently available automatic VB methods are very sensitive to initial starting conditions, model definition, algorithm hyperparameters and choice of gradient optimiser. The Bayesian LCA model was challenging to implement using VB but we achieved reasonable results with very good computational performance compared to MCMC.

stat.AP↗

arXiv2021

Copula Averaging for Tail Dependence in Insurance Claims Data

Analysing dependent risks is an important task for insurance companies. A dependency is reflected in the fact that information about one random variable provides information about the likely distribution of values of another random variable. Insurance companies in particular must investigate such dependencies between different lines of business and the effects that an extreme loss event, such as an earthquake or hurricane, has across multiple lines of business simultaneously. Copulas provide a popular model-based approach to analysing the dependency between risks, and the coefficient of tail dependence is a measure of dependence for extreme losses. Besides commonly used empirical estimators for estimating the tail dependence coefficient, copula fitting can lead to estimation of such coefficients directly or can verify their existence. Generally, a range of copula models is available to fit a data set well, leading to multiple different tail dependence results; a method based on Bayesian model averaging is designed to obtain a unified estimate of tail dependence. In this article, this model-based coefficient estimation method is illustrated through a variety of copula fitting approaches and results are presented for several simulated data sets and also a real general insurance loss data set.

stat.AP↗

arXiv2019

Investigation of Parameter Uncertainty in Clustering Using a Gaussian Mixture Model Via Jackknife, Bootstrap and Weighted Likelihood Bootstrap

Mixture models are a popular tool in model-based clustering. Such a model is often fitted by a procedure that maximizes the likelihood, such as the EM algorithm. At convergence, the maximum likelihood parameter estimates are typically reported, but in most cases little emphasis is placed on the variability associated with these estimates. In part this may be due to the fact that standard errors are not directly calculated in the model-fitting algorithm, either because they are not required to fit the model, or because they are difficult to compute. The examination of standard errors in model-based clustering is therefore typically neglected. The widely used R package mclust has recently introduced bootstrap and weighted likelihood bootstrap methods to facilitate standard error estimation. This paper provides an empirical comparison of these methods (along with the jackknife method) for producing standard errors and confidence intervals for mixture parameters. These methods are illustrated and contrasted in both a simulation study and in the traditional Old Faithful data set and Thyroid data set.

stat.CO↗

arXiv2019

Bivariate Gamma Mixture of Experts Models for Joint Insurance Claims Modeling

In general insurance, risks from different categories are often modeled independently and their sum is regarded as the total risk the insurer takes on in exchange for a premium. The dependence from multiple risks is generally neglected even when correlation could exist, for example a single car accident may result in claims from multiple risk categories. It is desirable to take the covariance of different categories into consideration in modeling in order to better predict future claims and hence allow greater accuracy in ratemaking. In this work multivariate severity models are investigated using mixture of experts models with bivariate gamma distributions, where the dependence structure is modeled directly using a GLM framework, and covariates can be placed in both gating and expert networks. Furthermore, parsimonious parameterisations are considered, which leads to a family of bivariate gamma mixture of experts models. It can be viewed as a model-based clustering approach that clusters policyholders into sub-groups with different dependencies, and the parameters of the mixture models are dependent on the covariates. Clustering is shown to be important in separating the data into sub-groupings where strong dependence is often present, even if the overall data set exhibits only weak dependence. In doing so, the correlation within different components features prominently in the model. It is shown that, by applying to both simulated data and a real-world Irish GI insurer data set, claim predictions can be improved.

stat.AP↗

arXiv2018

Improved model-based clustering performance using Bayesian initialization averaging

The Expectation-Maximization (EM) algorithm is a commonly used method for finding the maximum likelihood estimates of the parameters in a mixture model via coordinate ascent. A serious pitfall with the algorithm is that in the case of multimodal likelihood functions, it can get trapped at a local maximum. This problem often occurs when sub-optimal starting values are used to initialize the algorithm. Bayesian initialization averaging (BIA) is proposed as an ensemble method to generate high quality starting values for the EM algorithm. Competing sets of trial starting values are combined as a weighted average, which is then used as the starting position for a full EM run. The method can also be extended to variational Bayes (VB) methods, a class of algorithm similar to EM that is based on an approximation of the model posterior. The BIA method is demonstrated on real continuous, categorical and network data sets, and the convergent log-likelihoods and associated clustering solutions presented. These compare favorably with the output produced using competing initialization methods such as random starts, hierarchical clustering and deterministic annealing, with the highest available maximum likelihood estimates obtained in a higher percentage of cases, at reasonable computational cost. The implications of the different clustering solutions obtained by local maxima are also discussed.

stat.CO↗

arXiv2017

Motor Insurance Accidental Damage Claims Modeling with Factor Collapsing and Bayesian Model Averaging

Accidental damage is a typical component of motor insurance claim. Modeling of this nature generally involves analysis of past claim history and different characteristics of the insured objects and the policyholders. Generalized linear models (GLMs) have become the industry's standard approach for pricing and modeling risks of this nature. However, the GLM approach utilizes a single "best" model on which loss predictions are based, which ignores the uncertainty among the competing models and variable selection. An additional characteristic of motor insurance data sets is the presence of many categorical variables, within which the number of levels is high. In particular, not all levels of such variables may be statistically significant and rather some subsets of the levels may be merged to give a smaller overall number of levels for improved model parsimony and interpretability. A method is proposed for assessing the optimal manner in which to collapse a factor with many levels into one with a smaller number of levels, then Bayesian model averaging (BMA) is used to blend model predictions from all reasonable models to account for factor collapsing uncertainty. This method will be computationally intensive due to the number of factors being collapsed as well as the possibly large number of levels within factors. Hence a stochastic optimisation is proposed to quickly find the best collapsing cases across the model space.

stat.AP↗