2024 年的期刊文章 · 第 14 页

共 1515 篇,按发表时间排,新的在前,每页 40 篇。这一页是 2024-06-15 到 2024-06-04。

按期刊、按主题看 →

第 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 页

SiM Spatial & Time Series · 发表于 2024-06-15

Dynamic hierarchical state space forecasting

Ziyue Liu, Wensheng Guo

摘要

In this paper, we aim to both borrow information from existing units and incorporate the target unit's history data in time series forecasting. We consider a situation when we have time series data from multiple units that share similar patterns when aligned in terms of an internal time. The internal time is defined as an index according to evolving features of interest. When mapped back to the calendar time, these time series can span different time intervals that can include the future calendar time of the targeted unit, over which we can borrow the information from other units in forecasting the targeted unit. We first build a hierarchical state space model for the multiple time series data in terms of the internal time, where the shared components capture the similarities among different units while allowing for unit‐specific deviations. A conditional state space model is then constructed to incorporate the information of existing units as the prior information in forecasting the targeted unit. By running the Kalman filtering based on the conditional state space model on the targeted unit, we incorporate both the information from the other units and the history of the targeted unit. The forecasts are then transformed from internal time back into calendar time for ease of interpretation. A simulation study is conducted to evaluate the finite sample performance. Forecasting state‐level new COVID‐19 cases in United States is used for illustration.

SiM Other · 发表于 2024-06-15

Cumulative incidence of cardiac surgery associated with exposure to benfluorex: A retrospective analysis based on compensation claims data

Paddy Farrington, Solène Lellinger

摘要

Data on retrospective compensation claims for injuries caused by pharmaceutical drugs are prone to selection and reporting biases. Nevertheless, this case study of the antidiabetic drug benfluorex shows that such data can be used to estimate the cumulative incidence of drug‐related injury, and to provide insights into its epidemiology. To this end, we develop a modelling framework for under‐reporting of retrospective claims for compensation arising from drug damage. The model involves a longitudinal component related to attrition of cases over time, and a cross‐sectional component related to incomplete reporting. We apply this model to cardiac valve surgery necessitated by exposure to benfluorex. Benfluorex was marketed in France between 1976 and 2009, when it was withdrawn because it caused valvular heart disease. A scandal erupted in 2010 over the scale of the damage caused by the drug. Since then, no further estimates of cumulative incidence have been published, though thousands of claims for compensation have been processed. The analysis combines compensation claims data and sociological survey data on benfluorex users, together with data on benfluorex sales and duration of treatment. We find a threshold of toxicity at about 6 months' exposure, and that at least 1690 individuals (95% CI 1290 to 2320) needed heart surgery to replace or repair valves damaged by exposure to benfluorex in France: a cumulative incidence of 3.68 per 10,000 (95% CI 2.68 to 5.34) benfluorex users or 3.22 per 10,000 (95% CI 2.48 to 4.39) person‐years at risk above the exposure threshold. While these findings are tentative, they are consistent with those obtained previously using very different methods.

JRSSB High-Dimensional · 发表于 2024-06-15

Correlation adjusted debiased Lasso: debiasing the Lasso with inaccurate covariate model

Michael Celentano, Andrea Montanari

摘要

We consider the problem of estimating a low-dimensional parameter in high-dimensional linear regression. Constructing an approximately unbiased estimate of the parameter of interest is a crucial step towards performing statistical inference. Several authors suggest to orthogonalize both the variable of interest and the outcome with respect to the nuisance variables, and then regress the residual outcome with respect to the residual variable. This is possible if the covariance structure of the regressors is perfectly known, or is sufficiently structured that it can be estimated accurately from data (e.g. the precision matrix is sufficiently sparse). Here we consider a regime in which the covariate model can only be estimated inaccurately, and hence existing debiasing approaches are not guaranteed to work. We propose the correlation adjusted debiased Lasso, which nearly eliminates this bias in some cases, including cases in which the estimation errors are neither negligible nor orthogonal.

Biometrika Networks & Graphs Spatial & Time Series Optimization & Learning Theory · 发表于 2024-06-14

Radial neighbours for provably accurate scalable approximations of Gaussian processes

Yichen Zhu, Michele Peruzzi, Cheng Li, David B Dunson

摘要

In geostatistical problems with massive sample size, Gaussian processes can be approximated using sparse directed acyclic graphs to achieve scalable O(n) computational complexity. In these models, data at each location are typically assumed conditionally dependent on a small set of parents that usually include a subset of the nearest neighbours. These methodologies often exhibit excellent empirical performance, but the lack of theoretical validation leads to unclear guidance in specifying the underlying graphical model and sensitivity to graph choice. We address these issues by introducing radial-neighbour Gaussian processes, a class of Gaussian processes based on directed acyclic graphs in which directed edges connect every location to all of its neighbours within a predetermined radius. We prove that any radial-neighbour Gaussian process can accurately approximate the corresponding unrestricted Gaussian process in the Wasserstein-2 distance, with an error rate determined by the approximation radius, the spatial covariance function and the spatial dispersion of samples. We offer further empirical validation of our approach via applications on simulated and real-world data, showing excellent performance in both prior and posterior approximations to the original Gaussian process.

Biometrika Networks & Graphs High-Dimensional · 发表于 2024-06-14

On some algorithms for estimation in Gaussian graphical models

S Højsgaard, S Lauritzen

摘要

In Gaussian graphical models, the likelihood equations must typically be solved iteratively. This paper investigates two algorithms: a version of iterative proportional scaling, which avoids inversion of large matrices, and an algorithm based on convex duality and operating on the covariance matrix by neighbourhood coordinate descent, which corresponds to the graphical lasso with zero penalty. For large, sparse graphs, the iterative proportional scaling algorithm appears feasible and has simple convergence properties. The algorithm based on neighbourhood coordinate descent is extremely fast and less dependent on sparsity, but needs a positive-definite starting value to converge. We provide an algorithm for finding such a starting value for graphs with low colouring number. As a consequence, we also obtain a simplified proof of existence of the maximum likelihood estimator in such cases.

Biometrika Causal Inference Nonparametric & Semiparametric · 发表于 2024-06-14

On propensity score matching with a diverging number of matches

Yihui He, Fang Han

摘要

This paper re-examines the work of Abadie & Imbens (2016) on propensity score matching for average treatment effect estimation. We explore the asymptotic behaviour of these estimators when the number of nearest neighbours, M, grows with the sample size. It is shown, while not surprising, but technically nontrivial, that the modified estimators can improve upon the original fixed M-estimators in terms of efficiency. Additionally, we demonstrate the potential to attain the semiparametric efficiency lower bound when the propensity score admits some special structures, echoing the insight of Hahn (1998).

JRSSB Testing & Uncertainty Genetics & Omics · 发表于 2024-06-14

Catch me if you can: signal localization with knockoff e-values

Paula Gablenz, Chiara Sabatti

摘要

We consider problems where many, somewhat redundant, hypotheses are tested and we are interested in reporting the most precise rejections, with false discovery rate (FDR) control. This is the case, for example, when researchers are interested both in individual hypotheses as well as group hypotheses corresponding to intersections of sets of the original hypotheses, at several resolution levels. A concrete application is in genome-wide association studies, where, depending on the signal strengths, it might be possible to resolve the influence of individual genetic variants on a phenotype with greater or lower precision. To adapt to the unknown signal strength, analyses are conducted at multiple resolutions and researchers are most interested in the more precise discoveries. Assuring FDR control on the reported findings with these adaptive searches is, however, often impossible. To design a multiple comparison procedure that allows for an adaptive choice of resolution with FDR control, we leverage e-values and linear programming. We adapt this approach to problems where knockoffs and group knockoffs have been successfully applied to test conditional independence hypotheses. We demonstrate its efficacy by analysing data from the UK Biobank.

Biometrika Longitudinal Data Testing & Uncertainty · 发表于 2024-06-14

Bootstrap test procedure for variance components in nonlinear mixed effects models in the presence of nuisance parameters and a singular Fisher information matrix

T Guédon, C Baey, E Kuhn

摘要

Summary: We examine the problem of variance component testing in general mixed effects models using the likelihood ratio test. We account for the presence of nuisance parameters, ie, the fact that some untested variances might also be equal to zero. Two main issues arise in this context, leading to a nonregular setting. First, under the null hypothesis, the true parameter value lies on the boundary of the parameter space. Moreover, due to the presence of nuisance parameters, the exact locations of these boundary points are not known, which prevents the use of classical asymptotic theory of maximum likelihood estimation. Then, in the specific context of nonlinear mixed effects models, the Fisher information matrix is singular at the true parameter value. We address these two points by proposing a shrunk parametric bootstrap procedure, which is straightforward to apply even for nonlinear models. We show that the procedure is consistent, solving both the boundary and the singularity issues, and we provide a verifiable criterion for the applicability of our theoretical results. We show through a simulation study that, compared to the asymptotic approach, our procedure has a better small sample performance and is more robust to the presence of nuisance parameters. A real data application on bird growth rates is also provided.

JASA Other · 发表于 2024-06-13

Generalized Data Thinning Using Sufficient Statistics

Ameer Dharamshi, Anna Neufeld, Keshav Motwani, Lucy L. Gao, Daniela Witten, Jacob Bien

摘要

Our goal is to develop a general strategy to decompose a random variable X into multiple independent random variables, without sacrificing any information about unknown parameters. A recent paper showed that for some well-known natural exponential families, X can be thinned into independent random variables X ( 1 ) , … , X ( K ) , such that X = ∑ k = 1 K X ( k ) . These independent random variables can then be used for various model validation and inference tasks, including in contexts where traditional sample splitting fails. In this paper, we generalize their procedure by relaxing this summation requirement and simply asking that some known function of the independent random variables exactly reconstruct X . This generalization of the procedure serves two purposes. First, it greatly expands the families of distributions for which thinning can be performed. Second, it unifies sample splitting and data thinning, which on the surface seem to be very different, as applications of the same principle. This shared principle is sufficiency. We use this insight to perform generalized thinning operations for a diverse set of families. *(abstract via PubMed)*

JRSSB Networks & Graphs Latent Variables & Mixtures 讨论 / 评论 · 发表于 2024-06-12

Yang Feng and Jiajin Sun’s contribution to the Discussion of ‘Root and community inference on the latent growth process of a network’ by Crane and Xu

Yang Feng, Jiajin Sun

摘要

We congratulate Drs. Crane and Xu for their fascinating piece of work. In the multiple roots PAPER model, one assumption is that different trees share the same affine preferential attachment (APA) growth parameters (α,β)⁠. In practice, the multiple trees in a network may have varying growth mechanisms, and it may be of interest to model each tree to have its own growth parameters. We consider such a heterogeneous affine preferential attachment (HAPA) model. Definition 1 resembles the K roots model in Crane Xu (2023), with two major differences: (i) instead of homogeneous (α,β) parameters, the new model has K pairs of growth parameters, one for each tree and (ii) the likelihood of the HAPA forest depends on not only the degree sequence but also the tree growth history. EM algorithm of estimation of (α,β)=(αk,βk)k=1K under the HPAPER model Gibbs sampling of ordering and forests from the HPAPER model To conduct inference for the model in Definition 1, we need to address two primary tasks: estimating the (αk,βk)k=1K parameters, and sampling from the distribution of time labelled forests. Both tasks bring new challenges compared with the K roots model in Crane and Xu (2023). First, the growth parameters can be estimated in the PAPER model without knowing the class assignments of the nodes, while in the HPAPER model, one must have some knowledge or estimation of the clustering as they are necessary even for defining the growth parameters. Second, some of the smart sampling techniques for the APA model are built upon the property that the forest likelihood depends only on the degree sequences of its nodes, which is no longer the case for the HAPA likelihood (1). Following the notation in Crane and Xu (2023), we write f~n=πfn where f~n denotes the randomly alphabetically labelled forest and π denotes the ordering of the nodes. We describe an Expectation–Maximization (EM) algorithm framework to estimate the parameters (α,β) and a Gibbs sampling framework to sample forest f~n and ordering π in Algorithms 1 and 2, respectively. We plan to give a full treatment to the proposed models and algorithms in a follow-up work. In the future, it is also of interest to establish an ‘HPAPER-SBM’ model, in which the Erdős–Rényi parameter is allowed to be different within and between different communities, and further explore suitable sampling and estimation procedures under that model. Another interesting direction is to incorporate nodal or edge covariates to the model (Huang et al., 2023; Weng & Feng, 2022). *(abstract via OpenAlex)*

SMMR High-Dimensional · 发表于 2024-06-12

Point estimation of the 100p percent lethal dose using a novel penalised likelihood approach

Yilei Ma, Youpeng Su, Peng Wang, Ping Yin

摘要

Estimation of the 100 p percent lethal dose ([Formula: see text]) is of great interest to pharmacologists for assessing the toxicity of certain compounds. However, most existing literature focuses on the interval estimation of [Formula: see text] and little attention has been paid to its point estimation. Currently, the most commonly used method for estimating the [Formula: see text] is the maximum likelihood estimator (MLE), which can be represented as a ratio estimator, with the denominator being the slope estimated from the logistic regression model. However, the MLE can be seriously biased when the sample size is small, a common nature in such studies, or when the dose–response curve is relatively flat (i.e. the slope approaches zero). In this study, we address these issues by developing a novel penalised maximum likelihood estimator (PMLE) that can prevent the denominator of the ratio from being close to zero. Similar to the MLE, the PMLE is computationally simple and thus can be conveniently used in practice. Moreover, with a suitable penalty parameter, we show that the PMLE can (a) reduce the bias to the second order with respect to the sample size and (b) avoid extreme estimates. Through simulation studies and real data applications, we show that the PMLE generally outperforms the existing methods in terms of bias and root mean square error.

JASA Causal Inference · 发表于 2024-06-12

Distribution-Free Prediction Intervals Under Covariate Shift, With an Application to Causal Inference

Jing Qin, Yukun Liu, Moming Li, Chiung-Yu Huang

摘要

Owing to its appealing distribution-free feature, conformal inference has become a popular tool for constructing prediction intervals with a desired coverage rate. In scenarios involving covariate shift, where the shift function needs to be estimated from data, many existing methods resort to data-splitting techniques. However, these approaches often lead to wider intervals and less reliable coverage rates, especially when dealing with finite sample sizes. To address these challenges, we propose methods based on a pivotal quantity derived under a parametric working model and employ a resampling-based framework to approximate its distribution. The resampling-based approach can produce prediction intervals with a desired coverage rate without splitting the data and can be easily applied to causal inference settings where a shift in the covariate distribution can occur between treatment and control arms. Additionally, the proposed approaches enjoy a double robustness property and are adaptable to different prediction tasks. Our extensive numerical experiments demonstrate that, compared to existing methods, the proposed novel approaches can produce substantially shorter conformal prediction intervals with lower variability in the interval lengths while maintaining promising coverage rates and advantages in versatile usage. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work. *(abstract via OpenAlex)*

JASA Machine Learning Latent Variables & Mixtures · 发表于 2024-06-11

Reinforcement Learning in Latent Heterogeneous Environments

Elynn Y. Chen, Rui Song, Michael I. Jordan

摘要

Reinforcement Learning holds great promise for data-driven decision-making in various social contexts, including healthcare, education, and business. However, classical methods that focus on the mean of the total return may yield misleading results when dealing with heterogeneous populations typically found in large-scale datasets. To address this issue, we introduce the K-Value Heterogeneous Markov Decision Process, a framework designed to handle sequential decision problems with latent population heterogeneity. Within this framework, we propose auto-clustered policy evaluation for estimating the value of a given policy and auto-clustered policy iteration for estimating the optimal policy within a parametric policy class. Our auto-clustered algorithms can automatically identify homogeneous subpopulations while simultaneously estimating the action value function and the optimal policy for each subgroup. We establish convergence rates and construct confidence intervals for the estimators. Simulation results support our theoretical findings, and an empirical study conducted on a real medical dataset confirms the presence of value heterogeneity and validates the advantages of our novel approach. *(abstract via OpenAlex)*

SiM Diagnostics & Prediction · 发表于 2024-06-11

Non‐parametric inference on calibration of predicted risks

Mohsen Sadatsafavi, John Petkau

摘要

Moderate calibration, the expected event probability among observations with predicted probability z being equal to z, is a desired property of risk prediction models. Current graphical and numerical techniques for evaluating moderate calibration of risk prediction models are mostly based on smoothing or grouping the data. As well, there is no widely accepted inferential method for the null hypothesis that a model is moderately calibrated. In this work, we discuss recently‐developed, and propose novel, methods for the assessment of moderate calibration for binary responses. The methods are based on the limiting distributions of functions of standardized partial sums of prediction errors converging to the corresponding laws of Brownian motion. The novel method relies on well‐known properties of the Brownian bridge which enables joint inference on mean and moderate calibration, leading to a unified “bridge” test for detecting miscalibration. Simulation studies indicate that the bridge test is more powerful, often substantially, than the alternative test. As a case study we consider a prediction model for short‐term mortality after a heart attack, where we provide suggestions on graphical presentation and the interpretation of results. Moderate calibration can be assessed without requiring arbitrary grouping of data or using methods that require tuning of parameters.

Biometrics Diagnostics & Prediction Causal Inference · 发表于 2024-06-11

Efficient testing of the biomarker positive and negative subgroups in a biomarker-stratified trial

Lang Li, Anastasia Ivanova

摘要

In many randomized placebo-controlled trials with a biomarker defined subgroup, it is believed that this subgroup has the same or higher treatment effect compared with its complement. These subgroups are often referred to as the biomarker positive and negative subgroups. Most biomarker-stratified pivotal trials are aimed at demonstrating a significant treatment effect either in the biomarker positive subgroup or in the overall population. A major shortcoming of this approach is that the treatment can be declared effective in the overall population even though it has no effect in the biomarker negative subgroup. We use the isotonic assumption about the treatment effects in the two subgroups to construct an efficient way to test for a treatment effect in both the biomarker positive and negative subgroups. A substantial reduction in the required sample size for such a trial compared with existing methods makes evaluating the treatment effect in both the biomarker positive and negative subgroups feasible in pivotal trials especially when the prevalence of the biomarker positive subgroup is less than 0.5.

JASA Bayesian Methods Genetics & Omics · 发表于 2024-06-11

Controlled Discovery and Localization of Signals via Bayesian Linear Programming

Asher Spector, Lucas Janson

摘要

Scientists often must simultaneously localize and discover signals. For instance, in genetic fine-mapping, high correlations between nearby genetic variants make it hard to identify the exact locations of causal variants. So the statistical task is to output as many disjoint regions containing a signal as possible, each as small as possible, while controlling false positives. Similar problems arise, for example, when locating stars in astronomical surveys and in changepoint detection. Common Bayesian approaches to these problems involve computing a posterior distribution over signal locations. However, existing procedures to translate these posteriors into credible regions for the signals fail to capture all the information in the posterior, leading to lower power and (sometimes) inflated false discoveries. We introduce Bayesian Linear Programming (BLiP), which can efficiently convert any posterior distribution over signals into credible regions for signals. BLiP overcomes an extremely high-dimensional and nonconvex problem to verifiably nearly maximize expected power while controlling false positives. Applying BLiP to existing state-of-the-art analyses of UK Biobank data (for genetic fine-mapping) and the Sloan Digital Sky Survey (for astronomical point source detection) increased power by 30%–120% in just a few minutes of additional computation. BLiP is implemented in pyblip (Python) and blipr (R). Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work. *(abstract via OpenAlex)*

SiM Bayesian Methods Optimization & Learning Theory Clinical Trials · 发表于 2024-06-10

REDOMA: Bayesian random‐effects dose‐optimization meta‐analysis using spike‐and‐slab priors

Cheng‐Han Yang, Evan Kwiatkowski, J. Jack Lee, Ruitao Lin

摘要

The rise of cutting‐edge precision cancer treatments has led to a growing significance of the optimal biological dose (OBD) in modern oncology trials. These trials now prioritize the consideration of both toxicity and efficacy simultaneously when determining the most desirable dosage for treatment. Traditional approaches in early‐phase oncology trials have conventionally relied on the assumption of a monotone relationship between treatment efficacy and dosage. However, this assumption may not hold valid for novel oncology therapies. In reality, the dose‐efficacy curve of such treatments may reach a plateau at a specific dose, posing challenges for conventional methods in accurately identifying the OBD. Furthermore, achieving reliable identification of the OBD is typically not possible based on a single small‐sample trial. With data from multiple phase I and phase I/II trials, we propose a novel Bayesian random‐effects dose‐optimization meta‐analysis (REDOMA) approach to identify the OBD by synthesizing toxicity and efficacy data from each trial. The REDOMA method can address trials with heterogeneous characteristics. We adopt a curve‐free approach based on a Gamma process prior to model the average dose‐toxicity relationship. In addition, we utilize a Bayesian model selection framework that uses the spike‐and‐slab prior as an automatic variable selection technique to eliminate monotonic constraints on the dose‐efficacy curve. The good performance of the REDOMA method is confirmed by extensive simulation studies.

SiM Survival Analysis Missing Data & Measurement Error · 发表于 2024-06-10

Conditional score approaches to errors‐in‐variables competing risks data in discrete time

Chi‐Chung Wen, Yi‐Hau Chen

摘要

Analysis of competing risks data has been an important topic in survival analysis due to the need to account for the dependence among the competing events. Also, event times are often recorded on discrete time scales, rendering the models tailored for discrete‐time nature useful in the practice of survival analysis. In this work, we focus on regression analysis with discrete‐time competing risks data, and consider the errors‐in‐variables issue where the covariates are prone to measurement errors. Viewing the true covariate value as a parameter, we develop the conditional score methods for various discrete‐time competing risks models, including the cause‐specific and subdistribution hazards models that have been popular in competing risks data analysis. The proposed estimators can be implemented by efficient computation algorithms, and the associated large sample theories can be simply obtained. Simulation results show satisfactory finite sample performances, and the application with the competing risks data from the scleroderma lung study reveals the utility of the proposed methods.

SiM Other · 发表于 2024-06-10

Categorical linkage‐data analysis

Li‐Chun Zhang, Tiziana Tuoto

摘要

Analysis of integrated data often requires record linkage in order to join together the data residing in separate sources. In case linkage errors cannot be avoided, due to the lack a unique identity key that can be used to link the records unequivocally, standard statistical techniques may produce misleading inference if the linked data are treated as if they were true observations. In this paper, we propose methods for categorical data analysis based on linked data that are not prepared by the analyst, such that neither the match‐key variables nor the unlinked records are available. The adjustment is based on the proportion of false links in the linked file and our approach allows the probabilities of correct linkage to vary across the records without requiring that one is able to estimate this probability for each individual record. It accommodates also the general situation where unmatched records that cannot possibly be correctly linked exist in all the sources. The proposed methods are studied by simulation and applied to real data.

SiM Survival Analysis · 发表于 2024-06-09

Familywise error for multiple time‐to‐event endpoints in a group sequential design

Henrik F. Thomsen, Nanna L. Lausvig, Christian B. Pipper, Søren Andersen, Lars H. Damgaard, Scott S. Emerson, Henrik Ravn

摘要

We investigate the familywise error rate (FWER) for time‐to‐event endpoints evaluated using a group sequential design with a hierarchical testing procedure for secondary endpoints. We show that, in this setup, the correlation between the log‐rank test statistics at interim and at end of study is not congruent with the canonical correlation derived for normal‐distributed endpoints. We show, both theoretically and by simulation, that the correlation also depends on the level of censoring, the hazard rates of the endpoints, and the hazard ratio. To optimize operating characteristics in this complex scenario, we propose a simulation‐based method to assess the FWER which, better than the alpha‐spending approach, can inform the choice of critical values for testing secondary endpoints.

SiM Longitudinal Data Bayesian Methods Clinical Trials · 发表于 2024-06-09

Bayesian transition models for ordinal longitudinal outcomes

Maximilian D. Rohde, Benjamin French, Thomas G. Stewart, Frank E. Harrell

摘要

Ordinal longitudinal outcomes are becoming common in clinical research, particularly in the context of COVID‐19 clinical trials. These outcomes are information‐rich and can increase the statistical efficiency of a study when analyzed in a principled manner. We present Bayesian ordinal transition models as a flexible modeling framework to analyze ordinal longitudinal outcomes. We develop the theory from first principles and provide an application using data from the Adaptive COVID‐19 Treatment Trial (ACTT‐1) with code examples in R. We advocate that researchers use ordinal transition models to analyze ordinal longitudinal outcomes when appropriate alongside standard methods such as time‐to‐event modeling.

SiM Bayesian Methods Latent Variables & Mixtures · 发表于 2024-06-09

Bayesian hierarchical profile regression for binary covariates

Jonathan Beall, Hong Li, Bonnie Martin‐Harris, Brian Neelon, Jordan Elm, Evan Graboyes, Elizabeth Hill

摘要

Dysphagia, a common result of other medical conditions, is caused by malfunctions in swallowing physiology resulting in difficulty eating and drinking. The Modified Barium Swallow Study (MBSS), the most commonly used diagnostic tool for evaluating dysphagia, can be assessed using the Modified Barium Swallow Impairment Profile (MBSImP™). The MBSImP assessment tool consists of a hierarchical grouped data structure with multiple domains, a set of components within each domain which characterize specific swallowing physiologies, and a set of tasks scored on a discrete scale within each component. We lack sophisticated approaches to extract patterns of physiologic swallowing impairment from the MBSImP task scores within a component while still recognizing the nested structure of components within a domain. We propose a Bayesian hierarchical profile regression model, which uses a Bayesian profile regression model in conjunction with a hierarchical Dirichlet process mixture model to (1) cluster subjects into impairment profile patterns while respecting the hierarchical grouped data structure of the MBSImP, and (2) simultaneously determine associations between latent profile cluster membership for all components and the outcome of dysphagia severity. We apply our approach to a cohort of patients referred for an MBSS and assessed using the MBSImP. Our research results can be used to inform appropriate intervention strategies, and provide tools for clinicians to make better multidimensional management and treatment decisions for patients with dysphagia.

SiM Clinical Trials · 发表于 2024-06-09

Adding experimental treatment arms to multi‐arm multi‐stage platform trials in progress

Thomas Burnett, Franz König, Thomas Jaki

摘要

Multi‐arm multi‐stage (MAMS) platform trials efficiently compare several treatments with a common control arm. Crucially MAMS designs allow for adjustment for multiplicity if required. If for example, the active treatment arms in a clinical trial relate to different dose levels or different routes of administration of a drug, the strict control of the family‐wise error rate (FWER) is paramount. Suppose a further treatment becomes available, it is desirable to add this to the trial already in progress; to access both the practical and statistical benefits of the MAMS design. In any setting where control of the error rate is required, we must add corresponding hypotheses without compromising the validity of the testing procedure.To strongly control the FWER, MAMS designs use pre‐planned decision rules that determine the recruitment of the next stage of the trial based on the available data. The addition of a treatment arm presents an unplanned change to the design that we must account for in the testing procedure. We demonstrate the use of the conditional error approach to add hypotheses to any testing procedure that strongly controls the FWER. We use this framework to add treatments to a MAMS trial in progress. Simulations illustrate the possible characteristics of such procedures.

SiM High-Dimensional · 发表于 2024-06-07

Model‐based bioequivalence approach for sparse pharmacokinetic bioequivalence studies: Model selection or model averaging?

Morgane Philipp, Adrien Tessier, Mark Donnelly, Lanyan Fang, Kairui Feng, Liang Zhao, Stella Grosser, Guoying Sun, Wanjie Sun, France Mentré, Julie Bertrand

摘要

Conventional pharmacokinetic (PK) bioequivalence (BE) studies aim to compare the rate and extent of drug absorption from a test (T) and reference (R) product using non‐compartmental analysis (NCA) and the two one‐sided test (TOST). Recently published regulatory guidance recommends alternative model‐based (MB) approaches for BE assessment when NCA is challenging, as for long‐acting injectables and products which require sparse PK sampling. However, our previous research on MB‐TOST approaches showed that model misspecification can lead to inflated type I error. The objective of this research was to compare the performance of model selection (MS) on R product arm data and model averaging (MA) from a pool of candidate structural PK models in MBBE studies with sparse sampling. Our simulation study was inspired by a real case BE study using a two‐way crossover design. PK data were simulated using three structural models under the null hypothesis and one model under the alternative hypothesis. MB‐TOST was applied either using each of the five candidate models or following MS and MA with or without the simulated model in the pool. Assuming T and R have the same PK model, our simulation shows that following MS and MA, MB‐TOST controls type I error rates at or below 0.05 and attains similar or even higher power than when using the simulated model. Thus, we propose to use MS prior to MB‐TOST for BE studies with sparse PK sampling and to consider MA when candidate models have similar Akaike information criterion.

JRSSB Networks & Graphs Latent Variables & Mixtures 讨论 / 评论 · 发表于 2024-06-07

Fan Wang, Yi Yu and Alessandro Rinaldo's contribution to the Discussion of ‘Root and community inference on the latent growth process of a network’ by Crane and Xu

Fan Wang, Yi Yu, Alessandro Rinaldo

摘要

We congratulate Prof. Crane and Dr Xu for introducing a very interesting and appealing statistical model for Markovian network growth. The PAPER model enjoys several qualities. In particular, (i) it strikes a rare balance between analytic elegance and tractability, and expressive power; (ii) it allows for practicable algorithms for statistical inference that scale to networks of very large size, and (iii) it is very flexible and can be readily generalized to model a variety of phenomena and features commonly observed in modern networks. As pointed out by the authors, there are many extensions and open problems related to the model that are worth considering. In this note, we suggest possible extensions to change point analysis (CPA) for networks. CPA is a well-studied topic concerned with modelling and detecting abrupt changes in the data-generating distribution in time series data. Developing models, theories and methods for CPA in dynamic and large networks is a relatively new and exciting area of research (e.g. Wang et al., 2021; Yu et al., 2023). We believe the PAPER model provides an excellent reference framework for building powerful and realistic Markovian network models. To provide some details, in the PAPER model, at time point t∈[n]⁠, the newly added node with label t is connected to an existing node wt∈[t−1]⁠, with probability {βDTt−1(wt)+α}/{β2(t−2)+α(t−1)}⁠. The parameters (α,β) are fixed across the whole time course. It would be interesting to consider the scenario where the parameters are instead allowed to change in a piecewise manner at unknown CPs. In the simplest instantiation of the PAPER CP model, there might exist an unknown CP t* such that the values of the parameters α and β change after t*⁠. A possible application could be, in a social network, at the booming stage of a key opinion leader, β, the parameter characterizing the ‘rich gets richer’ phenomenon, is positive and large. As the craze cools down, β should decrease to reflect the fading of fame. In extreme cases, β may even change the sign. To estimate the change time t*⁠, one could consider a likelihood-based ℓ0-penalization (e.g. Wang et al., 2023), with the likelihood stemming from the PAPER model. A second, more sophisticated extension is to consider an APA model with multiple root nodes; at the CP(s), the number of root nodes, along, possibly, with the model parameter, change, thus accounting for the creation, elimination or even merging of tree components. Finally, the CPA tasks just outlined can be analysed in the offline settings, in which the fully grown network at a given time, say n, is observed and then analysed. Lastly, we congratulate Prof. Crane and Dr Xu again for their excellent paper. We anticipate that the ideas and methods of the paper will provide the impetus for further research developments of the PAPER model for years to come! *(abstract via OpenAlex)*

JRSSB Networks & Graphs Latent Variables & Mixtures Bayesian Methods · 发表于 2024-06-07

Authors’ reply to the Discussion of ‘Root and community inference on the latent growth process of a network’

Harry Crane, Min Xu

摘要

We sincerely thank all the discussants for their careful thoughts and insightful contributions. We also appreciate the diversity of topics in the discussions, ranging from applied probability and algorithms to social science and Bayesian modelling—it appropriately reflects the broad thinking necessary to make advancements in network data analysis. In this article, we respond to model related issues in Section 1 and theory related issues in Section 2. In the final section, we respond to various specific points raised by each of the discussants. Many discussants highlighted ways in which the Preferential Attachment Plus Erdös–Rényi (PAPER) model may be unrealistic and suggested potential extensions. We acknowledge that such suggestions could be appropriate in the right context, but emphasize that the context dictates which features are the most salient to model. Except in stylized contexts, none of which exist in networks applications to our knowledge, no model can fully account for all observed properties of the data. And one might argue that no model should even try. Instead, the model ought to be specified to adequately explain the most salient properties of the data for the purpose of a given scientific question. Beyond that, the principle of parsimony suggests that the simplest model that can address the question of interest is often the best. This is the principle we applied in suggesting the PAPER model, as a framework on which more specific and sophisticated models can be tailored to a wide range of applications. For example, the PAPER model may be appropriate if the scientific question is to understand node centrality in a given network. The PAPER model gives a simple way to quantify uncertainty when measuring node centrality; it enables questions like ‘is one node significantly more central than another?’ The multiple roots model also allows one to study ‘community-specific” node centrality instead of global node centrality. In the paper, we discussed the seq-PAPER model and the deletion noise model to illustrate how one could extend the PAPER model and what would be the corresponding modifications needed for the inference algorithm. As Prof. Li noted, inference becomes more computationally involved once we leave the comfort of the PAPER model, but very often it is still tractable on networks with hundreds or thousands of nodes. Yang and Tong, Srakar, and Wyse, Ng, White, and Fop all pointed out that the PAPER notion of community does not seem to take into account the phenomenon of homophily, where nodes with similar characteristics tend to connect with each other, e.g. a person tends to become friends with others of similar cultural backgrounds. Homophily has been the main consideration behind the design of statistical network models for the past two decades. It is a primary motivation for stochastic block models and various latent space models. Homophily is certainly important, but one of the theses of our work is that other features of network data are also important and have been overlooked. Specifically, we believe that the underlying Markovian growth process plays an equally crucial role in forming the topology of a network. Indeed, because the PAPER model is able to obtain good community detection results on real world networks, there is good reason to believe that when estimating communities in a network, one should account for not just homophily but also by the underlying growth processes of the communities. We did not explicitly incorporate homophily in the PAPER model in order to keep the model simple. Just as it is good practice to control as many extraneous variables as possible in a scientific experiment, we believed we can illustrate our ideas most clearly by having the model focus on the growth process and omitting extraneous features. However, we agree that it would be appropriate to incorporate homophily into the PAPER model in many applications. One example is the PAPER-SBM model which we briefly discussed in the paper. We discuss a few other examples here. If we know the characteristics of the nodes that induce homophily, then the extension proposed by Yang and Tong is very sensible and in fact similar to a proposal by Kim and Altmann (2017), who studied the effect of accounting for homophily in the preferential attachment model. To adapt their proposal to the PAPER model, we could, when generating the tree T⁠, have a new node u connect to an existing node v with probability where Λuv is the affinity between node u and node v; for example, we may have Λuv=1 if u,v are in the same community and Λuv0⁠. Remove node 1 and all edges incident on node 1 from G0 to obtain G⁠. Output G without labelling. Once we ‘retire’ the original root node, each direct child of the original root node becomes a new root of a community-tree. Note that in this model, we give each root a base degree of 1 instead of a base degree of 2. This model is a specific example of ‘vertex retirement’ described by Jog and Loh. The sampling procedure for this model is also simple. We can sample π from distribution P(Π=⋅|F~,G~) by sampling an ordering from the history of F~ uniformly at random, equivalent to the cases described in the paper. To sample a forest f~ from distribution P(F~=⋅|Π,G~)⁠, we follow Algorithm 3 in our paper except we choose a new parent wt∈{∅}∪(π1:(t−1)∪Ng~n(πt) for πt with probability proportional to where K is the current number of trees in f⁠. We have implemented this prior and it works as expected in simulation studies. We will study the specification of prior in the random K model more extensively in a future work. The points made by the discussants piqued our own curiosity on a question: how do we interpret frequentist inference results when we know for a fact that our model is misspecified? This question pertains to all of statistics but it is particularly tricky for network data for two reasons. First is that networks edges have strong dependence so we cannot assume we have IID observations; the second related reason is that we typically do not have universality phenomenon such as central limit theorem. How then do we interpret confidence set for the root node? Wyse, Ng, White, and Fop raised a concern very similar to this. We give several answers to this question although we acknowledge that none of them are perfectly satisfactory. our confidence set for the root node has a Bayesian on the that the observed network is the of the PAPER generating we can the root nodes. our procedure has a where it is a centrality which can be as a of centrality Section of our paper. because the root probability of node u is proportional to the of node u the root node, we can interpret nodes with root probability as the root of the of the network model to the PAPER model. the of model can be by a Markovian network to from a graph instead of a root node or several root nodes, as Jog and suggested. There are several works in this direction but a number of One is number of potential graphs in a network with the size of the Another is model with a graph models with interesting question then is how to or estimate the size of the model selection such as do not seem in this setting. As we in the discussion on more research is needed on model selection methods for networks. Jog and this when how one would choose between the of the PAPER model: K random This question can be to include the extension proposed by and and the preferential attachment extension proposed by and How should we choose these One is to and compare the of each of the potential models, but this to the question of how to choose a and computationally One potential is be to the on the degree distribution or higher order and generate by In this section, we respond to Prof. regarding how to the root inference We thank for theory to the which has a of our own inference is a that is to understand but difficult to This is in because the observed data is in an It is not to to a specific node in an For example, in a graph with nodes, we cannot between the two nodes or the two nodes. There are many ways to the root inference as Prof. discussion is also is that these are all equivalent in that the that give the same notion of root In the our goal to give a that is and also in the that the can from the to the without to new This is the main reason we the random it to the inference algorithm. It is not just a way to the node information. In what we out all the clearly and their We leave the to choose the that the most to G be a random graph with n nodes nodes are We of G as a Markovian model nodes are by their although the results apply to any random graph models. We observe where is an In the we that the node of take in to make it that the node of do not to the This is the we in the paper. be a random uniformly For a node we root probability as We that in the paper, we to the As Prof. Rubin-Delanchy pointed we do not to apply the observed graph as This is the that Prof. Rubin-Delanchy that the is uniformly random. as the and as the number of graphs of the same as we in the but we that on the of We these graphs as For a node we which is the set of nodes of from v once the node are We then the root probability of v as This is a that we described in a work The is that we do not even to an in order to the root inference It is to the Indeed, the tree model with and the node graph on the right if it or or or with probability It the graph on if it or or with probability of so that the probability to for a graph as the class The of is as defined in the We that is the set is not a may not be a of we the notion of a v be a node in then If we the and then is the For a node we the root probability as In our work we as the set of nodes of G that are equivalent to This set is the We that is the set We have that The all three of root In the we to and by because these are probability of a of In is the probability of a One of the is that is a for all nodes in the same discussants a number of questions related to on the size of the confidence set which we discuss here. pointed out that to the size of the confidence set when and one a centrality measure the We agree that an alternative centrality measure is needed but what that alternative should be an question. We to use of which is a proposed by et al. to study networks and other related models. network is the seq-PAPER model with and where the noise edge probability is it from the PAPER model in that nodes tend to be more connected each other. of this we could not obtain results the of We also do not how to extend centrality et al., to the setting. For the For the multiple root setting, Yang and Tong what could be community detection Jog and on the size of the confidence We that community is the PAPER model, because it does not seem that we can perfectly estimate the community of the nodes of a community-tree. We do that, in the K setting, the nodes of each of the K communities can be To be more it may be possible to obtain if we use a measure where we each node by the of potential may be to that the confidence set of the K root nodes are to K each of which to a and Tong whether the root can be used to statistics for the number of communities in the network. This is an interesting question. We have not studied how the root distribution when the specified K is or than the Jog and frequentist for the set in the noise setting. This is as in our paper to the noise setting. then the what can be the random K model. There are two to this question. if we assume that the process prior is then it from the that we would also have in that the set all the K root nodes with at However, if we that the graph is according to the PAPER model with K roots but where K is then we do not expect our set to have frequentist Jog and also confidence set for the K root nodes as a set of This is to do with our The reason we did not this in the paper is that the set of may be too when K is to points in the discussion by suggested between the size of our confidence set with by probability analysis. We have these in our paper on the tree and that the confidence are to points in the discussion by and We are for the suggestion of a to illustrate the model and the in the simplest We to our algorithm in and such an We are also for the suggestion of the PAPER model for We believe two examples include graph and to points in the discussion by and We thank and for their interesting of a on the PAPER model. This is similar to the model in et al. et al. consider the preferential attachment tree model and study two one is where the at and the other is where the at for on the degree with the question raised by and that in Markovian networks has properties which to be The proposed by is a to points in the discussion by Ascolani, Lijoi, and Ascolani, Lijoi, and Prünster raised the point of considering multiple networks instead of just an example regarding communities. we our on the network, for which there are In that we that the dynamics of the communities in to infer the network. There are a number of interesting we may consider if we observe multiple networks. When two network have the same set of nodes, we may consider the where of the latent growth that in to points in the discussion by Wyse, Ng, White, and Wyse, Ng, White, and Fop made an insightful regarding the algorithm for the in the PAPER model. In the paper, we made two in the algorithm. we the distribution of the tree degree given the graph by which the dependence between the graph degree of all the nodes. when we the distribution of by We do not have on how significantly these would of the We that the second has a effect because there is of the n distribution of the degree to distribution The may be if the observed graph G does not a PAPER graph at For example, we can that the is if G is a It would be interesting to compare the algorithm with or Bayesian inference where we a prior on the to points in the discussion by Catalano, Fasano, Giordano, and The model proposed by Catalano, Fasano, Giordano, and Rebaudo is If we the PAPER model as an process, then we are not assumptions the the ordering of the nodes would the order of One could consider a model where in each each existing node number of new nodes. is that a network by such a model would more information the root may even of the root We also that is used to process that a which Li to points in the discussion by and We thank and for their interesting discussion on the preferential attachment model. This proposal allows the trees in the to have growth To estimate the K parameter and a this could be even in the PAPER setting. also a way to adapt the sampling procedure to the setting. One is, sampling the ordering π given a forest to a of tree history where is the number of nodes in tree at the tree history on the final forest is The proposed may not of the history with the forest f⁠. methods may be appropriate here. to points in the discussion by that it seems we are a Bayesian The are that the Bayesian set for the root node has frequentist the is and the confidence set has size of an discuss each of these in the paper. to networks which we agree to points in the discussion by Jog and Jog and suggested a number of to the PAPER model. The notion of ‘vertex retirement’ is particularly from practical and if we ‘retire’ the root node in a root PAPER model, we obtain a random K PAPER model where of new trees is by a Pitman–Yor process. discussion in Section to points in the discussion by In to a question which we in Section Rubin-Delanchy that important concern at the that in the applications there would be information on the edges. It would seem to this in We here that our proposed model and are in the in which there is no information other than the In we explicitly assume that information is for the network. no point do we that one should such information if it to If information is then the root inference becomes of In most the information is or so that it could be very to also incorporate the network information. For example, in a network, we may have on the times of of all the to points in the discussion by Li raised a of the PAPER model to a process a In that setting, there is a set of nodes that as a root at an node at random. If we assume that the root node is uniformly at random, then we can also the root However, the set from the root distribution will not have frequentist in does hold when the graph For instance, if the graph is an tree or an to points in the discussion by Jiang and We thank the Jiang and Ke for the PAPER model on a network. In this because the of the edges are we expect the root distribution to higher to that many This seems to be the in the that Jiang and Ke Jiang and Ke that the PAPER model to account for degree to use of the Erdős–Rényi model. We point out that the PAPER model in model degree as a of preferential attachment the preferential attachment of PAPER is the Erdős–Rényi as a noise distribution on of the main we point out that PAPER degree with three the suggested stochastic a parameter for node of the so that the number of on sample is by and *(abstract via OpenAlex)*

JRSSB Networks & Graphs Bayesian Methods Latent Variables & Mixtures 讨论 / 评论 · 发表于 2024-06-07

Andrej Srakar’s contribution to the Discussion of ‘Root and community inference on the latent growth process of a network’ by Crane and Xu

Andrej Srakar

摘要

Paper presented by Harry Crane and Min Xu develops a rather novel area of network archaeology, which builds on findings of network science, probability theory and (in Crane and Xu's article) Bayesian statistics. My comments are directed to possible extensions for future work in this promising area of research. Firstly, in terms of network stochastic processes (authors use a combination of preferential attachment and Erdős–Rényi models) paper does little to explain comparisons with possible alternatives. Briend et al. (2023) discuss network archaeology in a random recursive dags and Cooper–Frieze random networks contexts, and Brandenberger et al. (2022) in the more general context of Bienaymé–Galton–Watson trees. I wonder if topology of the studied network could be incorporated in more detail, for example in today often studied weak-topology context (for example in Gromov-weak or just general Skorokhod topologies). This would need different metric spaces and distance metrics where results for different types of random graphs and trees are wide and could be useful for the development of the area as well for its extensions of studying the asymptotic behaviour and estimation. Combination with graphon perspectives which authors mention in the introduction would also be interesting to explore, as well as addressing the possibilities noted by Brandenberger et al. of extensions to k-ary, Cayley, Motzkin and planted plane trees as well as branching stochastic processes in general. Secondly, it seems to go unexplained why the authors are using Bayesian approach. The Bayesian component in the article seems underexplored and many possibilities could be useful for future developments of the used priors as well as Gibbs sampling procedure (authors themselves mention mixing properties of the sampler). Extensions of the Bayesian part could go in the direction of intractable likelihood perspectives, computational improvements (say, using integrated nested Laplace approximation) or in the selection of priors (parametric, semiparametric, or nonparametric perspectives or even to empirical Bayes possibilities). Study of the Bayesian asymptotic properties here would be very interesting. Finally, I miss more explicit and broad connection to the study of temporal networks and network in general. Not only root vertex and communities could be discovered, but connection to cliques, islands, and homophily as well as many possible types of networks based on characteristics of their ties, such as affiliation, weighted, multi-relational, or multi-layer networks. Possibilities for future research in this area of network science could loom large. The following contributions were received in writing after the meeting: *(abstract via OpenAlex)*

SMMR High-Dimensional · 发表于 2024-06-07

A robust regression model for bounded count health data

Cristian L Bayes, Jorge Luis Bazán, Luis Valdivieso

摘要

Bounded count response data arise naturally in health applications. In general, the well-known beta-binomial regression model form the basis for analyzing this data, specially when we have overdispersed data. Little attention, however, has been given to the literature on the possibility of having extreme observations and overdispersed data. We propose in this work an extension of the beta-binomial regression model, named the beta-2-binomial regression model, which provides a rather flexible approach for fitting a regression model with a wide spectrum of bounded count response data sets under the presence of overdispersion, outliers, or excess of extreme observations. This distribution possesses more skewness and kurtosis than the beta-binomial model but preserves the same mean and variance form of the beta-binomial model. Additional properties of the beta-2-binomial distribution are derived including its behavior on the limits of its parametric space. A penalized maximum likelihood approach is considered to estimate parameters of this model and a residual analysis is included to assess departures from model assumptions as well as to detect outlier observations. Simulation studies, considering the robustness to outliers, are presented confirming that the beta-2-binomial regression model is a better robust alternative, in comparison with the binomial and beta-binomial regression models. We also found that the beta-2-binomial regression model outperformed the binomial and beta-binomial regression models in our applications of predicting liver cancer development in mice and the number of inappropriate days a patient spent in a hospital.

SiM Clinical Trials Survival Analysis Bayesian Methods · 发表于 2024-06-06

Two‐stage randomized clinical trials with a right‐censored endpoint: Comparison of frequentist and Bayesian adaptive designs

Luana Boumendil, Sylvie Chevret, Vincent Lévy, Lucie Biard

摘要

Adaptive randomized clinical trials are of major interest when dealing with a time‐to‐event outcome in a prolonged observation window. No consensus exists either to define stopping boundaries or to combine values or test statistics in the terminal analysis in the case of a frequentist design and sample size adaptation. In a one‐sided setting, we compared three frequentist approaches using stopping boundaries relying on ‐spending functions and a Bayesian monitoring setting with boundaries based on the posterior distribution of the log‐hazard ratio. All designs comprised a single interim analysis with an efficacy stopping rule and the possibility of sample size adaptation at this interim step. Three frequentist approaches were defined based on the terminal analysis: combination of stagewise statistics (Wassmer) or of values (Desseaux), or on patientwise splitting (Jörgens), and we compared the results with those of the Bayesian monitoring approach (Freedman). These different approaches were evaluated in a simulation study and then illustrated on a real dataset from a randomized clinical trial conducted in elderly patients with chronic lymphocytic leukemia. All approaches controlled for the type I error rate, except for the Bayesian monitoring approach, and yielded satisfactory power. It appears that the frequentist approaches are the best in underpowered trials. The power of all the approaches was affected by the violation of the proportional hazards (PH) assumption. For adaptive designs with a survival endpoint and a one‐sided alternative hypothesis, the Wassmer and Jörgens approaches after sample size adaptation should be preferred, unless violation of PH is suspected.

Biometrics Bayesian Methods Networks & Graphs · 发表于 2024-06-06

Introduction to statistical modelling and inference by Murray Aitkin, CRC Press, 2023, ISBN: 978-1032105710, https://www.routledge.com/Introduction-to-Statistical-Modelling-and-Inference/Aitkin/p/book/9781032105710

Chuhsing Kate Hsiao

摘要

This modern introduction to statistical modeling and inference provides beginning students in data sciences and statistics the opportunity to benefit from two schools, the frequentist and Bayesian, in a unified problem-solving manner. Unlike traditional ways of learning frequentist theories first and then Bayesian statistics in a later advanced course, this textbook simultaneously introduces both methodologies to students at the beginning of their studies. Given that many statisticians in the real world embrace and utilize both approaches, this concurrent learning process is practical and relevant. The second distinguishing feature of this book is its inclusion of modern analytical tools commonly used in contemporary data analysis. These tools, such as model assessment, model averaging, regularization in regression, mixture distributions, expectation-maximization (EM) algorithms, Bayesian data augmentation, neural networks, and social networks, are widely employed in scientific communities for modeling purposes. Third, the book incorporates numerous application studies to illustrate the inferential tools. These studies provide concrete examples that enhance the understanding and relevance of the theories behind these tools. The structure and contents are outlined below. While the chapters are not explicitly grouped into specific “parts,” they can be loosely separated, as I have done in the following paragraphs, each of which represents a thematic section. I aim to provide readers with an organized framework while preserving the elements of surprise and discovery as they delve into the content. The first six chapters cover fundamental elements, introducing various studies, research questions, and collected data (in Chapters 3 and 4), which will be used throughout the book. Additionally, the book addresses sample surveys commonly encountered in the real world (Chapter 5) and metrics defined with probabilities (Chapter 6). This chapter on probability discusses the interpretations of relative frequency and degree of belief, laying the groundwork for inference in subsequent chapters. Moreover, the author examines the pitfalls of probability misuse through the tragic case of the Sally Clark trial. The formal introduction to statistical inference begins in Chapter 7. The content is intense. It contains many basic principles in inference, and yet the presentation is easy and exciting to follow. The chapter starts with the highlights of evidence in policymaking. Then, it moves to estimation methods with the survey sampling approach, followed by the introduction of the model-based theories, with emphasis on discrete distributions. The likelihood, frequentist theory, and Bayesian theory are explained in this chapter as well. The rest of this chapter then demonstrates how to apply these two theories and how to interpret the results in various studies. Chapter 8 focuses mainly on analyzing binomial data from randomized clinical trials (RCTs). Again, both frequentist and Bayesian approaches are introduced to be applied in RCTs and the well-known extra-corporeal membrane oxygenation (ECMO) trial is demonstrated. Chapter 9 is a detour for data visualization with histograms, empirical distributions, and survival functions. Inference methods for one-dimensional continuous distributions are considered in Chapter 10 and two-dimensional continuous distributions are covered in Chapter 11. Chapter 10 starts with the exponential distribution, its likelihood function, and derivations of the maximum likelihood estimator (MLE); it then discusses parameter transformation and how it affects the frequentist asymptotic inference but not the Bayesian inference. Inference on the Gaussian model is next investigated, followed by one-parameter frequentist and Bayesian hypothesis testing. Chapter 11 moves to two-parameter continuous distributions, including the Gaussian and three skewed distributions: the lognormal, Weibull, and Gamma. The prior specification (joint prior) and inference (with joint posterior from conditional and marginal posterior distributions) of the two-dimensional parameter are nicely and concisely explained. Chapter 12 assesses the appropriateness of continuous models employed to fit the observed data. The author compares the ordered data with the beta posterior distribution so that multiple fitted models can be evaluated simultaneously and displayed with graphs. The next two chapters, 13 and 14, are unique in textbooks of statistical inference. They contain concise and essential information for modern data scientists. Chapter 13 discusses only the multinomial distribution. The author begins with an explanation of why this chapter is placed after the two-parameter continuous distributions. The beginning of this chapter is very much worth reading. The rest of this chapter contains a brief discussion of frequentist analysis and a more detailed discussion of the Bayesian analysis. The Dirichlet process prior is introduced in this chapter. Chapter 14 examines the model fit with the likelihood ratio test, Bayes factor, frequentist deviance, and Bayesian deviance, when the models are nested and non-nested. This chapter contains many tools considered in model evaluation and concludes with a discussion of model choice and model averaging, making it a good source for students new to this field. Chapters 15–18 are about regression models. Chapter 15 contains most elements one can find in a standard regression textbook, including the design of covariates, interaction terms, and even regularization (ridge, lasso, and elastic net) regression and principal component regression. Chapter 16 discusses incompleteness and missingness, and this chapter then focuses on mixture distributions, EM algorithms, Bayesian analysis, and data augmentation. Many applications are demonstrated to exemplify the analytical tools, so that readers can repeat the analysis by themselves. Chapter 17 is about generalized linear models (GLMs). The theoretical derivations behind GLMs are only briefly explained but many examples are given for demonstrations. Extensions of GLMs are introduced in Chapter 18. Polynomial, piecewise regression, and networks are discussed here. Examples are used to illustrate these extensions with publicly available data, so anyone can recreate the plots and results stated in the book. There are two topics I would love to read about, but are not covered. The first one concerns multivariate data inference, particularly visualization strategies to explore potential relationships among multivariate data that could offer clues in formulating complex models. I understand, however, that this could significantly burden the already heavy volume. Although several related models and analyses, including regularization regression, principal components, and networks, have been discussed in different chapters, a section specifically devoted to this topic, along with other inference methods, could be considered. In addition, it would be nice to read the author’s presentation of the generalized linear mixed effects model (GLMMs) as a new section in Chapter 18 in a future edition. Overall, the book has many features that are personal favorites of mine: a lot of applications that illustrate the methods, a good combination of likelihood and model-based approaches, inclusion of computational operations on methods, and a parallel demonstration of frequentist and Bayesian analyses. Additionally, Appendix 4 provides a short history of statistics from 1890 by introducing four statisticians (Karl Pearson, Ronald Fisher, Jerzy Neyman, and Harold Jeffreys) and describing how they influenced the modern statistics society. This book can be used as a course textbook or for self-learning. Readers can acquire a general understanding of statistical inference and be prepared for any further advanced materials. I highly recommend this book to students who want to pursue a career in data science and statistics. Finally, I want to repeat the final sentence of the Preface: “Bayesians are a diverse group.” This is inspiring. I believe all statisticians are—and have to be—a diverse group, to foster and advance the scientific community. *(abstract via OpenAlex)*

SMMR Survival Analysis Diagnostics & Prediction Nonparametric & Semiparametric · 发表于 2024-06-06

Evaluating prognostic biomarkers for survival outcomes subject to informative censoring

Wei Liu, Danping Liu, Zhiwei Zhang

摘要

Prognostic biomarkers for survival outcomes are widely used in clinical research and practice. Such biomarkers are often evaluated using a C-index as well as quantities based on time-dependent receiver operating characteristic curves. Existing methods for their evaluation generally assume that censoring is uninformative in the sense that the censoring time is independent of the failure time with or without conditioning on the biomarker under evaluation. With focus on the C-index and the area under a particular receiver operating characteristic curve, we describe and compare three estimation methods that account for informative censoring based on observed baseline covariates. Two of them are straightforward extensions of existing plug-in and inverse probability weighting methods for uninformative censoring. By appealing to semiparametric theory, we also develop a doubly robust, locally efficient method that is more robust than the plug-in and inverse probability weighting methods and typically more efficient than the inverse probability weighting method. The methods are evaluated and compared in a simulation study, and applied to real data from studies of breast cancer and heart failure.

SiM Optimization & Learning Theory Clinical Trials · 发表于 2024-06-06

A seamless phase II/III design with dose optimization for oncology drug development

Yuhan Li, Yiding Zhang, Gu Mi, Ji Lin

摘要

The US FDA's Project Optimus initiative that emphasizes dose optimization prior to marketing approval represents a pivotal shift in oncology drug development. It has a ripple effect for rethinking what changes may be made to conventional pivotal trial designs to incorporate a dose optimization component. Aligned with this initiative, we propose a novel seamless phase II/III design with dose optimization (SDDO framework). The proposed design starts with dose optimization in a randomized setting, leading to an interim analysis focused on optimal dose selection, trial continuation decisions, and sample size re‐estimation (SSR). Based on the decision at interim analysis, patient enrollment continues for both the selected dose arm and control arm, and the significance of treatment effects will be determined at final analysis. The SDDO framework offers increased flexibility and cost‐efficiency through sample size adjustment, while stringently controlling the Type I error. This proposed design also facilitates both accelerated approval (AA) and regular approval in a “one‐trial” approach. Extensive simulation studies confirm that our design reliably identifies the optimal dosage and makes preferable decisions with a reduced sample size while retaining statistical power.

Biometrics Testing & Uncertainty · 发表于 2024-06-05

On exact randomization-based covariate-adjusted confidence intervals

Jacob Fiksel

摘要

Randomization-based inference using the Fisher randomization test allows for the computation of Fisher-exact P-values, making it an attractive option for the analysis of small, randomized experiments with non-normal outcomes. Two common test statistics used to perform Fisher randomization tests are the difference-in-means between the treatment and control groups and the covariate-adjusted version of the difference-in-means using analysis of covariance. Modern computing allows for fast computation of the Fisher-exact P-value, but confidence intervals have typically been obtained by inverting the Fisher randomization test over a range of possible effect sizes. The test inversion procedure is computationally expensive, limiting the usage of randomization-based inference in applied work. A recent paper by Zhu and Liu developed a closed form expression for the randomization-based confidence interval using the difference-in-means statistic. We develop an important extension of Zhu and Liu to obtain a closed form expression for the randomization-based covariate-adjusted confidence interval and give practitioners a sufficiency condition that can be checked using observed data and that guarantees that these confidence intervals have correct coverage. Simulations show that our procedure generates randomization-based covariate-adjusted confidence intervals that are robust to non-normality and that can be calculated in nearly the same time as it takes to calculate the Fisher-exact P-value, thus removing the computational barrier to performing randomization-based inference when adjusting for covariates. We also demonstrate our method on a re-analysis of phase I clinical trial data.

JRSSB Networks & Graphs Latent Variables & Mixtures 讨论 / 评论 · 发表于 2024-06-05

Marta Catalano, Augusto Fasano, Matteo Giordano, and Giovanni Rebaudo’s contribution to the Discussion of ‘Root and community inference on the latent growth process of a network’ by Crane and Xu

Marta Catalano, Augusto Fasano, Matteo Giordano, Giovanni Rebaudo

摘要

We congratulate the authors for their methodological and theoretical contribution to the statistical literature on networks. A natural extension of the proposed PAPER model is included, with K communities growing simultaneously and where new nodes are either assigned to an existing community or elected as a new root. The employed assignment rule is of Pólya-urn type, which leads to a logarithmic growth of the number of communities (Korwar & Hollander, 1973) and is known to coincide with the predictive scheme of exchangeable sequences associated with the Dirichlet process. The probability of creating a new community is then independent of the number of past ones, which is a distinctive feature of the Dirichlet process within the class of Gibbs-type priors (De Blasi et al., 2015). An interesting direction would be to allow for more flexible predictive schemes that ensure alternative asymptotics, ranging from power-law behaviours (via the Pitman-Yor process Pitman, 2006) or normalized generalized gamma completely random measures (Lijoi et al., 2007)) to slower than logarithmic growth (via the single-group hierarchical Dirichlet process (Camerlenghi et al., 2018). An important theoretical aspect is the number of communities. While the authors provide an empirical investigation, future research could tackle the question of posterior consistency for the number of communities, along the lines of the existing results for stochastic block models (Geng et al., 2019) and in the growing literature in Bayesian nonparametric mixture models (Ascolani et al., 2023; Miller & Harrison, 2013; Nobile, 1994). Turning to applications, the proposed model lends itself to some natural generalizations suggested by popular epidemiological models, like SIR dynamics, where at each instant the infectious nodes can transmit the disease to their susceptible neighbours with some probability, resulting in multiple new infectious individuals at the next time. Equating new infections in SIR dynamics to added nodes in the PAPER model, a useful extension would be obtained by allowing the addition of multiple nodes at each step: for instance, if computationally feasible, a fraction of the existing nodes at that time, representing an average contact rate, or also a random number, e.g. driven by a nonhomogeneous Poisson process. The SIR analogy further suggests extensions where nodes are active (i.e. accepting newly introduced nodes as neighbours) only for a limited time, representing the period during which an infectious agent can spread the disease. Once more, we commend the authors for an outstanding paper. *(abstract via OpenAlex)*

Biometrics Causal Inference Nonparametric & Semiparametric · 发表于 2024-06-05

Incorporating nonparametric methods for estimating causal excursion effects in mobile health with zero-inflated count outcomes

Xueqing Liu, Tianchen Qian, Lauren Bell, Bibhas Chakraborty

摘要

In mobile health, tailoring interventions for real-time delivery is of paramount importance. Micro-randomized trials have emerged as the “gold-standard” methodology for developing such interventions. Analyzing data from these trials provides insights into the efficacy of interventions and the potential moderation by specific covariates. The “causal excursion effect,” a novel class of causal estimand, addresses these inquiries. Yet, existing research mainly focuses on continuous or binary data, leaving count data largely unexplored. The current work is motivated by the Drink Less micro-randomized trial from the UK, which focuses on a zero-inflated proximal outcome, i.e., the number of screen views in the subsequent hour following the intervention decision point. To be specific, we revisit the concept of causal excursion effect, specifically for zero-inflated count outcomes, and introduce novel estimation approaches that incorporate nonparametric techniques. Bidirectional asymptotics are established for the proposed estimators. Simulation studies are conducted to evaluate the performance of the proposed methods. As an illustration, we also implement these methods to the Drink Less trial data.

SMMR Survival Analysis Clinical Trials · 发表于 2024-06-05

Group sequential methods based on supremum logrank statistics under proportional and nonproportional hazards

Jean Marie Boher, Thomas Filleron, Patrick Sfumato, Pierre Bunouf, Richard J Cook

摘要

Despite the widespread use of Cox regression for modeling treatment effects in clinical trials, in immunotherapy oncology trials and other settings therapeutic benefits are not immediately realized thereby violating the proportional hazards assumption. Weighted logrank tests and the so-called Maxcombo test involving the combination of multiple logrank test statistics have been advocated to increase power for detecting effects in these and other settings where hazards are nonproportional. We describe a testing framework based on supremum logrank statistics created by successively analyzing and excluding early events, or obtained using a moving time window. We then describe how such tests can be conducted in a group sequential trial with interim analyses conducted for potential early stopping of benefit. The crossing boundaries for the interim test statistics are determined using an easy-to-implement Monte Carlo algorithm. Numerical studies illustrate the good frequency properties of the proposed group sequential methods.

SiM Bayesian Methods · 发表于 2024-06-05

A Bayesian method to detect drug‐drug interaction using external information for spontaneous reporting system

Keisuke Tada, Kazushi Maruo, Masahiko Gosho

摘要

Due to the insufficiency of safety assessments of clinical trials for drugs, further assessments are required for post‐marketed drugs. In addition to adverse drug reactions (ADRs) induced by one drug, drug‐drug interaction (DDI)‐induced ADR should also be investigated. The spontaneous reporting system (SRS) is a powerful tool for evaluating the safety of drugs continually. In this study, we propose a novel Bayesian method for detecting potential DDIs in a database collected by the SRS. By applying a power prior, the proposed method can borrow information from similar drugs for a drug assessed DDI to increase sensitivity of detection. The proposed method can also adjust the amount of the information borrowed by tuning the parameters in power prior. In the simulation study, we demonstrate the aforementioned increase in sensitivity. Depending on the scenarios, approximately 20 points of sensitivity of the proposed method increase from an existing method to a maximum. We also indicate the possibility of early detection of potential DDIs by the proposed method through analysis of the database shared by the Food and Drug Administration. In conclusion, the proposed method has a higher sensitivity and a novel criterion to detect potential DDIs early, provided similar drugs have similar observed‐expected ratios to the drug under assessment.

SiM Clinical Trials Longitudinal Data · 发表于 2024-06-04

Robust analysis of stepped wedge trials using composite likelihood models

Emily C. Voldal, Avi Kenny, Fan Xia, Patrick Heagerty, James P. Hughes

摘要

Stepped wedge trials (SWTs) are a type of cluster randomized trial that involve repeated measures on clusters and design‐induced confounding between time and treatment. Although mixed models are commonly used to analyze SWTs, they are susceptible to misspecification particularly for cluster‐longitudinal designs such as SWTs. Mixed model estimation leverages both “horizontal” or within‐cluster information and “vertical” or between‐cluster information. To use horizontal information in a mixed model, both the mean model and correlation structure must be correctly specified or accounted for, since time is confounded with treatment and measurements are likely correlated within clusters. Alternative non‐parametric methods have been proposed that use only vertical information; these are more robust because between‐cluster comparisons in a SWT preserve randomization, but these non‐parametric methods are not very efficient. We propose a composite likelihood method that focuses on vertical information, but has the flexibility to recover efficiency by using additional horizontal information. We compare the properties and performance of various methods, using simulations based on COVID‐19 data and a demonstration of application to the LIRE trial. We found that a vertical composite likelihood model that leverages baseline data is more robust than traditional methods, and more efficient than methods that use only vertical information. We hope that these results demonstrate the potential value of model‐based vertical methods for SWTs with a large number of clusters, and that these new tools are useful to researchers who are concerned about misspecification of traditional models.

JRSSB Networks & Graphs Latent Variables & Mixtures 讨论 / 评论 · 发表于 2024-06-04

Qing Yang and Xin Tong’s contribution to the Discussion of ‘Root and community inference on the latent growth process of a network’ by Crane and Xu

Qing Yang, Xin Tong

摘要

First, we congratulate Dr. Crane and Dr. Xu for this important contribution to statistical network analysis. To address an overlooked fact that most real networks are formed through a growth process, the authors introduce a useful PAPER model and its variations and skilfully construct computationally feasible confidence sets for the root node(s). In the data analysis, the authors also use their models as a new approach to community detection. On this note, our interest lies in the multiple roots setting, as real social networks typically encompass more than one community. Take the well-known political blogs network (Adamic & Glance, 2005) as an example. It has K=2 ground truth communities—the liberal and conservative communities. At time t, the PAPER model attaches the new node t to an existing node wt∈{1,…,t−1} with probability which is somehow proportional to the degree of node wt⁠. In reality, in addition to the node degrees which reflect the popularity, homophily (McPherson et al., 2001) is a well-documented phenomenon that warrants attention. Individuals often exhibit a preference for connecting with others who have similar characteristics. Furthermore, due to either hostility or lack of interest, individuals in the liberal group may even decline connections with specific individuals from the other group. Consequently, an important question arises regarding how to incorporate this type of information into the PAPER model. A nonrigorous intuition in the fixed K roots setting suggests attaching node t to wt with a probability ρwtβDFt−1(wt)+2βI{wt∈S}+α(2β+α)(t−1)⁠, where ρwt∈(0,1) is set to be higher when nodes t and wt share similar characteristics; conversely, reducing the magnitude. Meanwhile, a selection mechanism can be employed to determine whether to retain the current edge choice or not, driven by binary options of ‘Yes’ or ‘No’ when deciding whether to follow an individual on the Internet. Another question pertains to inferring the true number of roots K. The authors propose methodologies to obtain the posterior root distributions, then can these probabilities be utilized to construct test statistics for testing H0:K=K0⁠, i.e. whether the model follows PAPER(α,β,K0,θ) or not? A related query concerns simulation studies that demonstrate promising community recovery performance, particularly with the fixed K model; therefore is it feasible to provide theoretical guarantees? Lastly, we would like to make two suggestions regarding further dissemination. (i) to expand the readerships of this work, the authors might consider writing an R markdown file that only includes the most essential components of the model and one simple example. (ii) We feel that the proposed models and algorithms have potentials to make some real insights in empirical studies beyond what other community detection algorithms can already claim. *(abstract via OpenAlex)*

JRSSB Networks & Graphs Testing & Uncertainty Latent Variables & Mixtures 讨论 / 评论 · 发表于 2024-06-04

Proposers of the vote of thanks to Crane and Xu and contribution to the Discussion of ‘Root and community inference on the latent growth process of a network’

Varun Jog, Po-Ling Loh

摘要

We commend the authors on an illuminating article which significantly extends the state of the art in network archaeology. Notably, whereas most other literature in the area has provided methods for root inference which are theoretically grounded only in the case of random trees, the techniques in this paper have been designed to model nontree-structured networks, as well. This has important practical implications, as many real-world networks contain one or more cycles. The main challenge faced by the probability and statistics community—for the better part of the last decade—was that it was difficult to concoct a model that was simple enough to analyse rigorously, yet complex enough to comprise a sufficiently interesting family of random graph structures. By introducing their PAPER model, Crane and Xu have proposed a model achieving an ideal balance, accompanied by a variety of attractively useful and strikingly intuitive theoretical results. The simulation results are also very impressive, in that the algorithms can be efficiently run on a single laptop even for networks with millions of nodes. The fact that the PAPER model achieves such versatility with only three parameters is a major strength of the work. This makes parameter estimation plausible (the authors suggest several methods for parameter estimation in the appendices, based on EM and Bayesian methods). From our reading, we wondered how one might decide in practice to use a more complicated model for a given network, e.g. sequential PAPER vs. nonsequential PAPER, or fixed-K vs. random K, as the more complicated networks would of course lead to more complicated inference procedures. For instance, is anything known about degree profiles of random-K networks, and are they qualitatively different from the degree profiles of fixed-K networks? (Which would be more reasonable for specific problems such as the coauthor networks studied in this paper?) We are also curious to understand better, in practice, what the confidence sets look like for K>1 (does the algorithm output K connected components?). How might the algorithm be modified if the goal is to output a set of K-tuples such that the K-tuple containing the K root nodes is contained in this set, with a certain high probability? Section 5 of the paper presents theoretical results concerning confidence sets for the root node (or root nodes, when K>1⁠) of a random network. Along the lines of some of our previous work (Jog & Loh, 2018), we wondered how much the theory could extend to the case of constructing a confidence set for the first M>1 nodes (say, for simplicity, in the case when K=1⁠). Also, can reasonable conditions be imposed under which confidence sets are proven to ‘persist’ when the number of nodes n tends to infinity? Finally, if the random network were initialized with a ‘seed graph’ (Lugosi & Pereira, 2019), could confidence sets for the seed graph be constructed by an extension of the techniques in this paper, with corresponding theoretical results? Although several variants of the PAPER model are introduced in Section 2 of the paper, the core theoretical results seem to only cover the vanilla and fixed-K cases. Thus, we wondered whether results such as the fact that credible sets have the correct frequentist coverage (Theorems 7 and 8) also holds for sequential and random-K models. Does the minimality property of confidence sets based on posterior root probabilities (Remark 6/Theorem S5) also hold for K>1? If any of the answers are negative, can the authors comment on the technical difficulties involved in extending their results to these cases? Lastly, we found the results of the paper to be rather thought-provoking in terms of their potential for opening new avenues for studying more complicated random growth models. Some examples include sequential models with a ‘vertex retirement’ feature, e.g. motivated by the study of coauthorship networks. Another idea is to model in deletion noise which is time-dependent, in contrast to the versions of PAPER presented in the paper, which assume that once connections are formed in the sequential model, they remain fixed forever. Other worthwhile models to study in conjunction with PAPER might be the sublinear preferential attachment model, which also exhibits persistence properties analogous to the linear preferential attachment model (Jog & Loh, 2016), and the ‘superstar model’, which purportedly provides a better fit to empirical data in social networks than the standard preferential attachment model (Bhamidi et al., 2015). *(abstract via OpenAlex)*

第 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 页