2023 年的期刊文章 · 第 5 页

共 1511 篇,按发表时间排,新的在前,每页 40 篇。这一页是 2023-11-03 到 2023-10-15。

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

SMMR Latent Variables & Mixtures Genetics & Omics Bayesian Methods · 发表于 2023-11-03

Clustering minimal inhibitory concentration data through Bayesian mixture models: An application to detect Mycobacterium tuberculosis resistance mutations

Clara Grazian

摘要

Antimicrobial resistance is becoming a major threat to public health throughout the world. Researchers are attempting to contrast it by developing both new antibiotics and patient-specific treatments. In the second case, whole-genome sequencing has had a huge impact in two ways: first, it is becoming cheaper and faster to perform whole-genome sequencing, and this makes it competitive with respect to standard phenotypic tests; second, it is possible to statistically associate the phenotypic patterns of resistance to specific mutations in the genome. Therefore, it is now possible to develop catalogues of genomic variants associated with resistance to specific antibiotics, in order to improve prediction of resistance and suggest treatments. It is essential to have robust methods for identifying mutations associated to resistance and continuously updating the available catalogues. This work proposes a general method to study minimal inhibitory concentration distributions and to identify clusters of strains showing different levels of resistance to antimicrobials. Once the clusters are identified and strains allocated to each of them, it is possible to perform regression method to identify with high statistical power the mutations associated with resistance. The method is applied to a new 96-well microtiter plate used for testing Mycobacterium tuberculosis.

SMMR Clinical Trials Spatial & Time Series Longitudinal Data · 发表于 2023-11-02

Sample sizes for estimating the sensitivity of a monitoring system that generates repeated binary outcomes with autocorrelation

Albert E Parker, James W Arbogast

摘要

Sample size formulas are provided to determine how many events and how many patient care units are needed to estimate the sensitivity of a monitoring system. The monitoring systems we consider generate time series binary data that are autocorrelated and clustered by patient care units. Our application of interest is an automated hand hygiene monitoring system that assesses whether healthcare workers perform hand hygiene when they should. We apply an autoregressive order 1 mixed effects logistic regression model to determine sample sizes that allow the sensitivity of the monitoring system to be estimated at a specified confidence level and margin of error. This model overcomes a major limitation of simpler approaches that fail to provide confidence intervals with the specified levels of confidence when the sensitivity of the monitoring system is above 90%.

Biometrika High-Dimensional Testing & Uncertainty · 发表于 2023-11-02

On inference in high-dimensional logistic regression models with separated data

R M Lewis, H S Battey

摘要

Summary: Direct use of the likelihood function typically produces severely biased estimates when the dimension of the parameter vector is large relative to the effective sample size. With linearly separable data generated from a logistic regression model, the loglikelihood function asymptotes and the maximum likelihood estimator does not exist. We show that an exact analysis for each regression coefficient produces half-infinite confidence sets for some parameters when the data are separable. Such conclusions are not vacuous, but an honest portrayal of the limitations of the data. Finite confidence sets are only achievable when additional, perhaps implicit, assumptions are made. Under a notional double-asymptotic regime in which the dimension of the logistic coefficient vector increases with the sample size, the present paper considers the implications of enforcing a natural constraint on the vector of logistic transformed probabilities. We derive a relationship between the logistic coefficients and a notional parameter obtained as a probability limit of an ordinary least-squares estimator. The latter exists even when the data are separable. Consistency is ascertained under weak conditions on the design matrix.

SiM Clinical Trials High-Dimensional · 发表于 2023-11-02

A frequentist design for basket trials using adaptive lasso

Lauren Kanapka, Anastasia Ivanova

摘要

A basket trial aims to expedite the drug development process by evaluating a new therapy in multiple populations within the same clinical trial. Each population, referred to as a “basket”, can be defined by disease type, biomarkers, or other patient characteristics. The objective of a basket trial is to identify the subset of baskets for which the new therapy shows promise. The conventional approach would be to analyze each of the baskets independently. Alternatively, several Bayesian dynamic borrowing methods have been proposed that share data across baskets when responses appear similar. These methods can achieve higher power than independent testing in exchange for a risk of some inflation in the type 1 error rate. In this paper we propose a frequentist approach to dynamic borrowing for basket trials using adaptive lasso. Through simulation studies we demonstrate adaptive lasso can achieve similar power and type 1 error to the existing Bayesian methods. The proposed approach has the benefit of being easier to implement and faster than existing methods. In addition, the adaptive lasso approach is very flexible: it can be extended to basket trials with any number of treatment arms and any type of endpoint.

SiM Survival Analysis Causal Inference Nonparametric & Semiparametric · 发表于 2023-10-30

Unified semicompeting risks analysis of hepatitis natural history through mediation modeling

Jih‐Chang Yu, Yen‐Tsung Huang

摘要

Natural history of hepatitis B or C is comprised of multiple milestones such as liver cirrhosis and liver cancer. To fully characterize its natural course, semicompeting risks represent a common problem where liver cirrhosis and liver cancer are both of interest, but only the former may be censored by the latter. Copula, frailty and multistate models serve as well‐established analytics for semicompeting risks. Here, we cast the semicompeting risks in a mediation framework, with liver cirrhosis as a mediator and liver cancer as an outcome. We define the indirect and direct effects as the effects of an exposure on the liver cancer incidence mediated and not mediated through liver cirrhosis, respectively. With the estimands derived as conditional probabilities, we derive respective expressions under the copula, frailty, and multistate models. Next, we propose estimators based on nonparametric maximum likelihood orU‐statistics and establish their asymptotic results. Numerical studies demonstrate that the efficiency of copula models leads to potential bias due to model misspecification. Moreover, the robustness of frailty models is accompanied by a loss in efficiency, and multistate models balance the efficiency and robustness. We demonstrate the utility of the proposed methods by a hepatitis study, showing that hepatitis B and C lead to a higher incidence of liver cancer by increasing liver cirrhosis incidence. Thus, mediation modeling provides a unified framework that accommodates various semicompeting risks models.

SiM Causal Inference · 发表于 2023-10-30

A guide to regression discontinuity designs in medical applications

Matias D. Cattaneo, Luke Keele, Rocío Titiunik

摘要

We present a practical guide for the analysis of regression discontinuity (RD) designs in biomedical contexts. We begin by introducing key concepts, assumptions, and estimands within both the continuity‐based framework and the local randomization framework. We then discuss modern estimation and inference methods within both frameworks, including approaches for bandwidth or local neighborhood selection, optimal treatment effect point estimation, and robust bias‐corrected inference methods for uncertainty quantification. We also overview empirical falsification tests that can be used to support key assumptions. Our discussion focuses on two particular features that are relevant in biomedical research: (i) fuzzy RD designs, which often arise when therapeutic treatments are based on clinical guidelines, but patients with scores near the cutoff are treated contrary to the assignment rule; and (ii) RD designs with discrete scores, which are ubiquitous in biomedical applications. We illustrate our discussion with three empirical applications: the effect CD4 guidelines for anti‐retroviral therapy on retention of HIV patients in South Africa, the effect of genetic guidelines for chemotherapy on breast cancer recurrence in the United States, and the effects of age‐based patient cost‐sharing on healthcare utilization in Taiwan. Complete replication materials employing publicly available data and statistical software in Python , R and Stata are provided, offering researchers all necessary tools to conduct an RD analysis.

SiM Survival Analysis · 发表于 2023-10-28

Win‐loss parameters for right‐censored event data, with application to recurrent events

Erik T. Parner, Morten Overgaard

摘要

The win ratio has become a popular method for comparing multiple event data between two groups in clinical cohort studies. The win ratio compares the event data in prioritized order, where the first prioritized event is death and a typical example for the second prioritized event is hospitalization. Literature is sparse on inference for win and loss parameters, including the win ratio, for censored event data. Inference for two prioritized censored event times has been developed for independent right‐censoring. Many clinical studies include recurrent event data such as hospitalizations. In this article, we suggest inference for win‐loss parameters for death and a recurrent event outcome under independent right‐censoring. The small sample properties of the proposed method are studied in a simulation study showing that the variance formula is accurate even for small samples. The method is applied on a data set from a randomized clinical trial.

SMMR Survival Analysis Nonparametric & Semiparametric · 发表于 2023-10-27

Regression analysis of a future state entry time distribution conditional on a past state occupation in a progressive multistate model

Yuting Yang, Samuel Wu, Somnath Datta

摘要

We present a nonparametric method for estimating the conditional future state entry probabilities and distributions of state entry time conditional on a past state visit when data are subject to dependent censorings in a progressive multistate model where Markovianity of the system is not assumed. These estimators are constructed using the competing risk techniques with risk sets consisting of fractional observations and inverse probability of censoring weights. The fractional observations correspond to estimates of the number of persons who ultimately enter a state from which the future state in question can be reached in one step. We then address the corresponding regression problem by combining these marginal estimators with the pseudo-value approach. The performance of our regression scheme is studied using a comprehensive simulation study. An analysis of existing data on graft-versus-host disease for bone marrow transplant individuals is presented using our novel methodology. A second analysis of another well-known data set on burn patients is also included.

SMMR Survival Analysis Longitudinal Data Diagnostics & Prediction · 发表于 2023-10-27

Random survival forests with multivariate longitudinal endogenous covariates

Anthony Devaux, Catherine Helmer, Robin Genuer, Cécile Proust-Lima

摘要

Predicting the individual risk of clinical events using the complete patient history is a major challenge in personalized medicine. Analytical methods have to account for a possibly large number of time-dependent predictors, which are often characterized by irregular and error-prone measurements, and are truncated early by the event. In this work, we extended the competing-risk random survival forests to handle such endogenous longitudinal predictors when predicting event probabilities. The method, implemented in the R package DynForest, internally transforms the time-dependent predictors at each node of each tree into time-fixed features (using mixed models) that can then be used as splitting candidates. The final individual event probability is computed as the average of leaf-specific Aalen–Johansen estimators over the trees. Using simulations, we compared the performances of DynForest to accurately predict an event with (i) a joint modeling alternative when considering two longitudinal predictors only, and with (ii) a regression calibration method that ignores the informative truncation by the event when dealing with a large number of longitudinal predictors. Through an application in dementia research, we also illustrated how DynForest can be used to develop a dynamic prediction tool for dementia from multimodal repeated markers, and quantify the importance of each marker.

JRSSB Causal Inference · 发表于 2023-10-27

Dynamic synthetic control method for evaluating treatment effects in auto-regressive processes

Xiangyu Zheng, Song Xi Chen

摘要

Motivated by evaluating the effects of air pollution alerts on air quality, we propose the dynamic synthetic control method for micro-level data with time-varying confounders and spatial dependence under an auto-regressive model setting. We employ the empirical likelihood to define the synthetic control weights, which ensures a unique solution and permits theoretical analysis. The dynamic matching increases the feasibility of matching and enables us to assess the unconfoundedness assumption using pre-treatment data. For statistical inference, we develop a normalised placebo test to address the asymmetry issue. The method is illustrated and evaluated on numerical simulations and a case study on air pollution alerts.

Biometrika Networks & Graphs · 发表于 2023-10-26

Deep Kronecker network

Long Feng, Guang Yang

摘要

Summary: We develop a novel framework for the analysis of medical imaging data, including magnetic resonance imaging, functional magnetic resonance imaging, computed tomography and more. Medical imaging data differ from general images in two main aspects: (i) the sample size is often considerably smaller and (ii) the interpretation of the model is usually more crucial than predicting the outcome. As a result, standard methods such as convolutional neural networks cannot be directly applied to medical imaging analysis. Therefore, we propose the deep Kronecker network, which can adapt to the low sample size constraint and offer the desired model interpretation. Our approach is versatile, as it works for both matrix- and tensor-represented image data and can be applied to discrete and continuous outcomes. The deep Kronecker network is built upon a Kronecker product structure, which implicitly enforces a piecewise smooth property on coefficients. Moreover, our approach resembles a fully convolutional network as the Kronecker structure can be expressed in a convolutional form. Interestingly, our approach also has strong connections to the tensor regression framework proposed by Zhou et al. (2013), which imposes a canonical low-rank structure on tensor coefficients. We conduct both classification and regression analyses using real magnetic resonance imaging data from the Alzheimer’s Disease Neuroimaging Initiative to demonstrate the effectiveness of our approach.

SMMR Other · 发表于 2023-10-25

Multivariate control charts for monitoring a bivariate correlated count process with application to meningococcal disease

Hanhan Li, Cong Li

摘要

In recent years, with the increasing number and complexity of infectious diseases, the idea of using control charts to monitor public health and disease has been proposed. In this paper, we study multivariate control charts for monitoring a bivariate integer-valued autocorrelation process with bivariate Poisson distribution and select the optimal control scheme by comparing the performance of control charts. Furthermore, the meningococcal patient event in two states in Australia serves as an example to illustrate the application of these methods. The results show that the D exponentially weighted moving average control scheme can detect the changes in the mean value faster, which is a significant advantage.

SiM Survival Analysis Nonparametric & Semiparametric High-Dimensional · 发表于 2023-10-24

Kernel Cox partially linear regression: Building predictive models for cancer patients' survival

Yaohua Rong, Sihai Dave Zhao, Xia Zheng, Yi Li

摘要

Wide heterogeneity exists in cancer patients' survival, ranging from a few months to several decades. To accurately predict clinical outcomes, it is vital to build an accurate predictive model that relates the patients' molecular profiles with the patients' survival. With complex relationships between survival and high‐dimensional molecular predictors, it is challenging to conduct nonparametric modeling and irrelevant predictors removing simultaneously. In this article, we build a kernel Cox proportional hazards semi‐parametric model and propose a novel regularized garrotized kernel machine (RegGKM) method to fit the model. We use the kernel machine method to describe the complex relationship between survival and predictors, while automatically removing irrelevant parametric and nonparametric predictors through a LASSO penalty. An efficient high‐dimensional algorithm is developed for the proposed method. Comparison with other competing methods in simulation shows that the proposed method always has better predictive accuracy. We apply this method to analyze a multiple myeloma dataset and predict the patients' death burden based on their gene expressions. Our results can help classify patients into groups with different death risks, facilitating treatment for better clinical outcomes.

JASA Other · 发表于 2023-10-24

Fast Approximation of the Shapley Values Based on Order-of-Addition Experimental Designs

Liuqing Yang, Yongdao Zhou, Haoda Fu, Min-Qian Liu, Wei Zheng

摘要

Shapley value is originally a concept in econometrics to fairly distribute both gains and costs to players in a coalition game. In the recent decades, its application has been extended to other areas such as marketing, engineering and machine learning. For example, it produces reasonable solutions for problems in sensitivity analysis, local model explanation toward the interpretable machine learning, node importance in social network, attribution models, etc. However, it could be very expensive to compute the Shapley value. Specifically, in a d-player coalition game, calculating a Shapley value requires the evaluation of d! or 2d marginal contribution values, depending on whether we are taking the permutation or combination formulation of the Shapley value. Hence, it becomes infeasible to calculate the Shapley value when d is reasonably large. A common remedy is to take a random sample of the permutations to surrogate for the complete list of permutations. We find an advanced sampling scheme can be designed to yield much more accurate estimation of the Shapley value than the simple random sampling (SRS). Our sampling scheme is based on combinatorial structures in the field of design of experiments (DOE), particularly the order-of-addition experimental designs for the study of how the orderings of components would affect the output. We show that the obtained estimates are unbiased, and can sometimes deterministically recover the original Shapley value. Both theoretical and simulations results show that our DOE-based sampling scheme outperforms SRS in terms of estimation accuracy. Surprisingly, it is also slightly faster than SRS. Lastly, real data analysis is conducted for the C. elegans nervous system and the 9/11 terrorist network. Supplementary materials for this article are available online. *(abstract via OpenAlex)*

JRSSB Bayesian Methods · 发表于 2023-10-24

Bayesian predictive decision synthesis

Emily Tallman, Mike West

摘要

Decision-guided perspectives on model uncertainty expand traditional statistical thinking about managing, comparing, and combining inferences from sets of models. Bayesian predictive decision synthesis (BPDS) advances conceptual and theoretical foundations, and defines new methodology that explicitly integrates decision-analytic outcomes into the evaluation, comparison, and potential combination of candidate models. BPDS extends recent theoretical and practical advances based on both Bayesian predictive synthesis and empirical goal-focused model uncertainty analysis. This is enabled by the development of a novel subjective Bayesian perspective on model weighting in predictive decision settings. Illustrations come from applied contexts including optimal design for regression prediction and sequential time series forecasting for financial portfolio decisions.

SiM Clinical Trials · 发表于 2023-10-23

Minimization in randomized clinical trials

Elisabeth Coart, Perrine Bamps, Emmanuel Quinaux, Geneviève Sturbois, Everardo D. Saad, Tomasz Burzykowski, Marc Buyse

摘要

In randomized trials, comparability of the treatment groups is ensured through allocation of treatments using a mechanism that involves some random element, thus controlling for confounding of the treatment effect. Completely random allocation ensures comparability between the treatment groups for all known and unknown prognostic factors. For a specific trial, however, imbalances in prognostic factors among the treatment groups may occur. Although accidental bias can be avoided in the presence of such imbalances by stratifying the analysis, most trialists, regulatory agencies, and other stakeholders prefer a balanced distribution of prognostic factors across the treatment groups. Some procedures attempt to achieve balance in baseline covariates, by stratifying the allocation for these covariates, or by dynamically adapting the allocation using covariate information during the trial (covariate‐adaptive procedures). In this Tutorial, the performance of minimization, a popular covariate‐adaptive procedure, is compared with two other commonly used procedures, completely random allocation and stratified blocked designs. Using individual patient data of 2 clinical trials (in advanced ovarian cancer and age‐related macular degeneration), the procedures are compared in terms of operating characteristics (using asymptotic and randomization tests), predictability of treatment allocation, and achieved balance. Fifty actual trials of various sizes that applied minimization for treatment allocation are used to investigate the achieved balance. Implementation issues of minimization are described. Minimization procedures are useful in all trials but especially when (1) many major prognostic factors are known, (2) many centers of different sizes accrue patients, or (3) the trial sample size is moderate.

SiM Survival Analysis Nonparametric & Semiparametric · 发表于 2023-10-22

Semiparametric estimation and testing for panel count data with informative interval‐censored failure event

Li Liu, Wen Su, Xingqiu Zhao

摘要

Panel count data and interval‐censored data are two types of incomplete data that often occur in event history studies. Almost all existing statistical methods are developed for their separate analysis. In this paper, we investigate a more general situation where a recurrent event process and an interval‐censored failure event occur together. To intuitively and clearly explain the relationship between the recurrent current process and failure event, we propose a failure time‐dependent mean model through a completely unspecified link function. To overcome the challenges arising from the blending of nonparametric components and parametric regression coefficients, we develop a two‐stage conditional expected likelihood‐based estimation procedure. We establish the consistency, the convergence rate and the asymptotic normality of the proposed two‐stage estimator. Furthermore, we construct a class of two‐sample tests for comparison of mean functions from different groups. The proposed methods are evaluated by extensive simulation studies and are illustrated with the skin cancer data that motivated this study.

JASA Nonparametric & Semiparametric · 发表于 2023-10-20

Semiparametrically Efficient Method for Enveloped Central Space

Linquan Ma, Jixin Wang, Han Chen, Lan Liu

摘要

The estimation of the central space is at the core of the sufficient dimension reduction (SDR) literature. However, it is well known that the finite-sample estimation suffers from collinearity among predictors. Cook et al. (2013) proposed the predictor envelope method under linear models that can alleviate the problem by targeting a bigger space - which not only envelopes the central information, but also partitions the predictors by finding an uncorrelated set of material and immaterial predictors. One limitation of the predictor envelope is that it has strong distributional and modeling assumptions and therefore, it cannot be readily used in semiparametric settings where SDR usually nests. In this paper, we generalize the envelope model by defining the enveloped central space and propose a semiparametric method to estimate it. We derive the entire class of regular and asymptotically linear (RAL) estimators as well as the locally and globally semiparametrically efficient estimators for the enveloped central space. Based on the connection between predictor envelope and partial least square (PLS), our methods can also be used to calculate the PLS space beyond linearity. In the simulations, our methods are shown to be both robust and accurate for estimating the enveloped central space under different settings. Moreover, the downstream analysis using state-of-the-art methods such as machine learning (ML) methods has the potential to achieve much better predictions. We further illustrate our methods in a heart failure study. *(abstract via PubMed)*

JASA Nonparametric & Semiparametric Missing Data & Measurement Error · 发表于 2023-10-20

Nonparametric Finite Mixture: Applications in Overcoming Misclassification Bias

Zi Ye, Solomon W. Harrar

摘要

Investigating the differential effect of treatments in groups defined by patient characteristics is of paramount importance in personalized medicine research. In some studies, participants are first classified as having or not of the characteristic of interest by diagnostic tools, but such classifiers may not be perfectly accurate. The impact of diagnostic misclassification in statistical inference has been recently investigated in parametric model contexts and shown to introduce severe bias in estimating treatment effects and give grossly inaccurate inferences. The article aims to address these problems in a fully nonparametric setting. Methods for consistently estimating and testing meaningful yet nonparametric treatment effects are developed. Along the way, we also construct estimators for misclassification error rates and investigate their asymptotic properties. The proposed methods are applicable for outcomes measured in ordinal, discrete, or continuous scales. They do not require any assumptions, such as the existence of moments. Simulation results show significant advantages of the proposed methods in bias reduction, coverage probability, and power. The applications of the proposed methods are illustrated with gene expression profiling of bronchial airway brushing in asthmatic and healthy control subjects. Supplementary materials for this article are available online. *(abstract via OpenAlex)*

SMMR Causal Inference Optimization & Learning Theory · 发表于 2023-10-20

Modified interactive Q-learning for attenuating the impact of model misspecification with treatment effect heterogeneity

Yuan Zhang, David M Vock, Megan E Patrick, Thomas A Murray

摘要

A sequential multiple assignment randomized trial, which incorporates multiple stages of randomization, is a popular approach for collecting data to inform personalized and adaptive treatments. There is an extensive literature on statistical methods to analyze data collected in sequential multiple assignment randomized trials and estimate the optimal dynamic treatment regime. Q-learning with linear regression is widely used for this purpose due to its ease of implementation. However, model misspecification is a common problem with this approach, and little attention has been given to the impact of model misspecification when treatment effects are heterogeneous across subjects. This article describes the integrative impact of two possible types of model misspecification related to treatment effect heterogeneity: omitted early-stage treatment effects in late-stage main effect model, and violated linearity assumption between pseudo-outcomes and predictors despite non-linearity arising from the optimization operation. The proposed method, aiming to deal with both types of misspecification concomitantly, builds interactive models into modified parametric Q-learning with Murphy’s regret function. Simulations show that the proposed method is robust to both sources of model misspecification. The proposed method is applied to a two-stage sequential multiple assignment randomized trial with embedded tailoring aimed at reducing binge drinking in first-year college students.

JASA Networks & Graphs Clinical Trials Testing & Uncertainty · 发表于 2023-10-20

Graphical Model Inference with Erosely Measured Data

Lili Zheng, Genevera I. Allen

摘要

In this paper, we investigate the Gaussian graphical model inference problem in a novel setting that we call erose measurements, referring to irregularly measured or observed data. For graphs, this results in different node pairs having vastly different sample sizes which frequently arises in data integration, genomics, neuroscience, and sensor networks. Existing works characterize the graph selection performance using the minimum pairwise sample size, which provides little insights for erosely measured data, and no existing inference method is applicable. We aim to fill in this gap by proposing the first inference method that characterizes the different uncertainty levels over the graph caused by the erose measurements, named GI-JOE (Graph Inference when Joint Observations are Erose). Specifically, we develop an edge-wise inference method and an affiliated FDR control procedure, where the variance of each edge depends on the sample sizes associated with corresponding neighbors. We prove statistical validity under erose measurements, thanks to careful localized edge-wise analysis and disentangling the dependencies across the graph. Finally, through simulation studies and a real neuroscience data example, we demonstrate the advantages of our inference methods for graph selection from erosely measured data. *(abstract via PubMed)*

JASA Bayesian Methods · 发表于 2023-10-20

Generalized Bayesian Inference for Discrete Intractable Likelihood

Takuo Matsubara, Jeremias Knoblauch, François-Xavier Briol, Chris. J. Oates

摘要

Discrete state spaces represent a major computational challenge to statistical inference, since the computation of normalization constants requires summation over large or possibly infinite sets, which can be impractical. This article addresses this computational challenge through the development of a novel generalized Bayesian inference procedure suitable for discrete intractable likelihood. Inspired by recent methodological advances for continuous data, the main idea is to update beliefs about model parameters using a discrete Fisher divergence, in lieu of the problematic intractable likelihood. The result is a generalized posterior that can be sampled from using standard computational tools, such as Markov chain Monte Carlo, circumventing the intractable normalizing constant. The statistical properties of the generalized posterior are analyzed, with sufficient conditions for posterior consistency and asymptotic normality established. In addition, a novel and general approach to calibration of generalized posteriors is proposed. Applications are presented on lattice models for discrete spatial data and on multivariate models for count data, where in each case the methodology facilitates generalized Bayesian inference at low computational cost. Supplementary materials for this article are available online. *(abstract via OpenAlex)*

JRSSB Other · 发表于 2023-10-20

Another look at bandwidth-free inference: a sample splitting approach

Yi Zhang, Xiaofeng Shao

摘要

The bandwidth-free tests for a multi-dimensional parameter have attracted considerable attention in econometrics and statistics literature. These tests can be conveniently implemented due to their tuning-parameter free nature and possess more accurate size as compared to the traditional heteroskedasticity and autocorrelation consistent-based approaches. However, when sample size is small/medium, these bandwidth-free tests exhibit large size distortion when both the dimension of the parameter and the magnitude of temporal dependence are moderate, making them unreliable to use in practice. In this paper, we propose a sample splitting-based approach to reduce the dimension of the parameter to one for the subsequent bandwidth-free inference. Our SS–SN (sample splitting plus self-normalisation) idea is broadly applicable to many testing problems for time series, including mean testing, testing for zero autocorrelation, and testing for a change point in multivariate mean, among others. Specifically, we propose two types of SS–SN test statistics and derive their limiting distributions under both the null and alternatives and show their effectiveness in alleviating size distortion via simulations. In addition, we obtain the limiting distributions for both SS–SN test statistics in the multivariate mean testing problem when the dimension is allowed to diverge.

JASA Networks & Graphs · 发表于 2023-10-19

Power and Multicollinearity in Small Networks: A Discussion of “Tale of Two Datasets: Representativeness and Generalisability of Inference for Samples of Networks” by Krivitsky, Coletti, and Hens

George G. Vega Yon

摘要

The recent work by Krivitsky, Coletti & Hens [KCH] provides an important new contribution to the Exponential-Family Random Graph Models [ERGMs], a start-to-finish approach to dealing with multi-network ERGMs. Although multi-network ERGMs have been around for a while (mostly in the form of block-diagonal models and multi-level ERGMs, see Duxbury and Wertsching (2023), Wang et al. (2013), Slaughter and Koehly (2016)), not much care has been given to the estimation and post-estimation steps. In their paper, Krivitsky, Coletti & Hens give a detailed layout of how to build, estimate, and analyze multi-ERGMs with heterogeneous data sources. In this comment, I will focus on two issues the authors did not discuss, namely, sample size requirements and multicollinearity. *(abstract via PubMed)*

Biometrika Nonparametric & Semiparametric Bayesian Methods · 发表于 2023-10-19

Nonparametric priors with full-range borrowing of information

F Ascolani, B Franzolini, A Lijoi, I Prünster

摘要

Summary: Modelling of the dependence structure across heterogeneous data is crucial for Bayesian inference, since it directly impacts the borrowing of information. Despite extensive advances over the past two decades, most available methods only allow for nonnegative correlations. We derive a new class of dependent nonparametric priors that can induce correlations of any sign, thus introducing a new and more flexible idea of borrowing of information. This is achieved thanks to a novel concept, which we term hyper-tie, and represents a direct and simple measure of dependence. We investigate prior and posterior distributional properties of the model and develop algorithms to perform posterior inference. Illustrative examples on simulated and real data show that the proposed method outperforms alternatives in terms of prediction and clustering.

Biometrika Other · 发表于 2023-10-19

Likelihood-based inference under nonconvex boundary constraints

J Y Wang, Z S Ye, Y Chen

摘要

Likelihood-based inference under nonconvex constraints on model parameters has become increasingly common in biomedical research. In this paper, we establish large-sample properties of the maximum likelihood estimator when the true parameter value lies at the boundary of a nonconvex parameter space. We further derive the asymptotic distribution of the likelihood ratio test statistic under nonconvex constraints on model parameters. A general Monte Carlo procedure for generating the limiting distribution is provided. The theoretical results are demonstrated by five examples in Anderson’s stereotype logistic regression model, genetic association studies, gene-environment interaction tests, cost-constrained linear regression and fairness-constrained linear regression.

SMMR Survival Analysis Functional Data Survey & Sampling · 发表于 2023-10-19

Functional proportional hazards mixture cure model with applications in cancer mortality in NHANES and post ICU recovery

Rahul Ghosal, Marcos Matabuena, Jiajia Zhang

摘要

We develop a functional proportional hazards mixture cure model with scalar and functional covariates measured at the baseline. The mixture cure model, useful in studying populations with a cure fraction of a particular event of interest is extended to functional data. We employ the expectation–maximization algorithm and develop a semiparametric penalized spline-based approach to estimate the dynamic functional coefficients of the incidence and the latency part. The proposed method is computationally efficient and simultaneously incorporates smoothness in the estimated functional coefficients via roughness penalty. Simulation studies illustrate a satisfactory performance of the proposed method in accurately estimating the model parameters and the baseline survival function. Finally, the clinical potential of the model is demonstrated in two real data examples that incorporate rich high-dimensional biomedical signals as functional covariates measured at the baseline and constitute novel domains to apply cure survival models in contemporary medical situations. In particular, we analyze (i) minute-by-minute physical activity data from the National Health And Nutrition Examination Survey 2003–2006 to study the association between diurnal patterns of physical activity at baseline and all cancer mortality through 2019 while adjusting for other biological factors; (ii) the impact of daily functional measures of disease severity collected in the intensive care unit on post intensive care unit recovery and mortality event. Our findings provide novel epidemiological insights into the association between daily patterns of physical activity and cancer mortality. Software implementation and illustration of the proposed estimation method are provided in R.

SiM Bayesian Methods Clinical Trials · 发表于 2023-10-19

BEATS: Bayesian hybrid design with flexible sample size adaptation for time‐to‐event endpoints

Dehua Bi, Meizi Liu, Jianchang Lin, Rachael Liu

摘要

As the roles of historical trials and real‐world evidence in drug development have substantially increased, several approaches have been proposed to leverage external data and improve the design of clinical trials. While most of these approaches focus on methodology development for borrowing information during the analysis stage, there is a risk of inadequate or absent enrollment of concurrent control due to misspecification of heterogeneity from external data, which can result in unreliable estimates of treatment effect. In this study, we introduce a Bayesian hybrid design with flexible sample size adaptation (BEATS) that allows for adaptive borrowing of external data based on the level of heterogeneity to augment the control arm during both the design and interim analysis stages. Moreover, BEATS extends the Bayesian semiparametric meta‐analytic predictive prior (BaSe‐MAP) to incorporate time‐to‐event endpoints, enabling optimal borrowing performance. Initially, BEATS calibrates the expected sample size and initial randomization ratio based on heterogeneity among the external data. During the interim analysis, flexible sample size adaptation is performed to address conflicts between the concurrent and historical control, while also conducting futility analysis. At the final analysis, estimation is provided by incorporating the calibrated amount of external data. Therefore, our proposed design allows for an approximation of an ideal randomized controlled trial with an equal randomization ratio while controlling the size of the concurrent control to benefit patients and accelerate drug development. BEATS also offers optimal power and robust estimation through flexible sample size adaptation when conflicts arise between the concurrent control and external data.

JRSSB Clinical Trials Bayesian Methods Testing & Uncertainty · 发表于 2023-10-19

Authors' reply to the Discussion of ‘Estimating means of bounded random variables by betting’

Ian Waudby-Smith, Aaditya Ramdas

摘要

We are grateful to each of the discussants for their rich and insightful discussions of our paper. We were thrilled to see that our work has piqued their interest, and it was a delight reading through discussions written by colleagues and friends, old and new. More broadly, we are delighted to see the widespread interest in betting, concentration inequalities, and anytime-valid inference from experts in a wide variety of fields including probability, mathematical statistics, machine learning, and online learning. The eclectic range of discussants aptly represents the diverse set of fields in which modern betting tools have their roots. In the paragraphs to follow, we will reply to discussants in the order that we received their discussions, saving our more detailed replies to the votes of thanks (of Peter Grünwald and Gergely Neu) for the end. We are in particular honoured that Anastasios Angelopoulos and colleagues have started using the name ‘WSR bounds’ to refer to the betting-based CIs we developed. We also thank Angelopoulos for the striking illustration of how our bounds can improve on existing methods when used within the Risk Controlling Prediction Set (RCPS) framework applied to tumor segmentation (Bates et al., 2021). More broadly, we view the RCPS and Prediction Powered Inference frameworks of Bates et al. (2021) and Angelopoulos et al. (2023) as much-needed tools in the assumption-lean statistical inference toolbox, and we are delighted that our bounds have helped sharpen parts of these tools. Art Owen makes some thoughtful observations relating the hindsight-optimal bets (optimal values of λ) to empirical likelihood (EL) weights, building on our observations in Appendix E. In particular, Owen notes that alternative weights are also sometimes used in empirical likelihood, for example λ:=exp{−λT(xi−m)} is motivated by entropy criteria. We note that such a formula happens to correspond exactly with the first Taylor expansion in the derivation of our ‘approximate GRAPA’ strategy, leading to perhaps another connection between betting and empirical likelihood that we were not aware of earlier. Owen also mentions interesting power and optimality guarantees that EL inherits from these weights. While we suspect that this may translate to a certain optimal power guarantee for the GRAPA betting strategy, it is not yet obvious to us what precise statements can be made. David Siegmund raises an interesting open question about sequentially estimating the density of a distribution using histogram binning. This task is only possible if some amount of smoothness is imposed (e.g. if the density were assumed to be Lipschitz continuous with a known constant). There may be hope to perform data-dependent binning (without sample-splitting) using the quantile-conditional histogram binning techniques (Gupta & Ramdas, 2021) that arose in the calibration literature, but it is indeed an open problem to sequentialise such a construction, for example progressively adding points to bins and then adaptively splitting bins with too many points. Siegmund also mentions applications to cumulative distribution functions (CDFs). Both Howard and Ramdas (2022) and Manole and Ramdas (2023) provide time- and quantile-uniform confidence bands for the CDF, without any restrictions on the underlying distribution. Notably, Howard et al. (2021, Theorem 5) attain widths that depend on the sample size and on the quantile being estimated, as Siegmund alludes to. These bounds may also be useful for Owen’s open problem connected to the Berk–Jones confidence bands (Berk & Jones, 1979). While we fully agree with Siegmund that methods often used in sequential clinical trials (e.g. group-sequential methods and repeated confidence intervals Jennison & Turnbull, 1984, 1989) can be tighter than anytime-valid CSs when applied to problems with fixed horizons, we view these two paradigms as complementary and targeting somewhat different sequential settings. Firstly, as Howard also brings up in his discussion, group-sequential methods are typically justified via asymptotic normality while much of the anytime-valid literature is focused on non-asymptotic guarantees. This is sometimes a matter of philosophical preference, but there are cases where non-asymptotics are quite important to the problem at hand. For example, in a post-election risk-limiting audit (RLA) (Stark, 2008), paper ballots are examined sequentially (often without replacement) from a stack to ensure that an erroneously announced winning candidate will be corrected with probability at least 1−α—here, α is the ‘risk limit’. To save time and resources, it is desirable to stop an RLA as soon as possible, and this can sometimes occur at unexpectedly small sample sizes, making it difficult to place much trust in asymptotics. We have already seen martingale-based methods and CSs have important applications for RLAs (see Spertus & Stark, 2022; Stark, 2020; Waudby-Smith, Stark, et al., 2021 and the papers of Philip Stark and colleagues more generally). However, even when resorting to asymptotics, anytime-valid procedures such as ‘asymptotic confidence sequences’ (Waudby-Smith, Arbour, et al., 2021) may be desirable since they do away with needing to pre-specify maximum sample sizes and/or interim analysis times. While such pre-specifications may fit naturally in settings with strong governmental regulations, we still find it worthwhile to develop new methods for other applications that allow for more flexibility. We thank Hien Nguyen for the comparison to universal inference (UI) in the parametric setting of mean estimation for Beta random variables. Similar to Nguyen, we were somewhat surprised to see that betting CSs managed to outperform UI in some but not all cases, despite the latter directly exploiting the parametric structure of the problem. One explanation for this phenomenon may be that UI constructs a two-dimensional confidence set for the (α,β) parameters of a Beta distribution, from which an implied one-dimensional set for the mean can be deduced, while our martingale-based confidence sets target the mean directly. Nevertheless, it is interesting to see how these methods compare in practice, and we thank Nguyen for running these simulations. Li, Li, and Dai make an empirical comparison between our confidence sequences to a sequence of L geometrically spaced bootstrap CIs, perhaps as a sanity check to compare our method’s performance against a simple competitor. Specifically, they compute one bootstrap (1−α/L)-CI CNB for each sample size N∈{2l:l=1,…,L}⁠, and set CnB←CNB for all N⩽n<2N⁠. They find that betting-based confidence sets are sometimes tighter and sometimes looser than these sequential bootstrap CIs, but the latter become very loose for large sample sizes. It is important to also keep in mind that even for a fixed sample size n, bootstrap CIs are only valid asymptotically (i.e. as n→∞⁠) while we only focused on non-asymptotic confidence sets that enjoy guarantees of the form P(μ∈C˙n)⩾1−α or P(∀t,μ∈C¯t)⩾1−α for CIs C˙n and CSs (C¯t)t=1∞⁠, respectively. While we focused on non-asymptotics in this paper, there do exist CSs that trade non-asymptotic validity for tightness and versatility—for example, Waudby-Smith, Arbour, et al. (2021) derive the so-called ‘asymptotic confidence sequences’ that only make finite moment assumptions similar to the central limit theorem, and unlike the geometrically spaced bootstrap CIs, these remain tight in large samples (and indeed shrink to a zero width in almost all cases). We thank Martin Larsson and Johannes Ruf for their crisp summary of the core of our methods using intuitive financial terms. Larsson and Ruf also show how the diversified Kelly process (KtdKelly)t=1∞—which is itself an average of several capital processes obtained via individual betting strategies—has an alternative representation as a single process with a capital-reweighted betting strategy. The mentioned connections to the portfolio theory of Cover (1991) are illuminating, and it would be interesting to explore similar representations of other betting strategies and combinations thereof. We are grateful to Philip Stark for his kind words as well as his excellent historical overview of the area. Stark mentions that it may be of interest to compare our bounds to Phan et al. (2021) and indeed we have a simulation comparing to it in the supplement (Waudby-Smith & Ramdas, 2023, Figure 16) where we find that our methods produces shorter intervals than theirs across sample sizes and different simulations (but their bound is quite competitive for low-variance continuous distributions). Further, their method requires an i.i.d. assumption, is much more computationally expensive than ours, and it does not have a time-uniform nor without-replacement analogue. Nevertheless, they use an interesting and rather different proof technique that deserves further exploration. We are also pleased to see how (Spertus & Stark, 2022) have used properties of martingales and e-values to derive union-intersection tests with stratified sampling. Rong Jiang and Keming Yu discuss the choices of λ in Hoeffding and our Predictable Plug-in empirical Bernstein (PrPl-EB) CIs, namely noting that the former involve a single data-independent choice while the latter involve several data-dependent tuning-parameters. While it is true that Hoeffding’s CI has a single data-independent tuning parameter, this is only because it cannot adapt to features of the distribution other than the bounds on the support—indeed, one could have derived a predictable plug-in Hoeffding CI with several tuning parameters, but the Hoeffding supermartingale would not benefit from this. Our empirical Bernstein CIs on the other hand, are built from a different process that can adapt to the unknown variance σ2 using a value of λ that depends on σ2⁠. Since σ2 is unknown to us, we estimate it in a careful way—i.e. through predictable plugins—using the fact that the empirical Bernstein process nevertheless forms a non-negative supermartingale, yielding non-asymptotic validity for any sample size n (including for any n<100 as queried by Jiang and Yu). The choice of c∈(0,1) has some effect on finite sample tightness (but not on validity), but one can inspect λtPrPl−EB and see that for sufficiently large t, the value of c will be of no importance. This is additionally illustrated by the fact that in the i.i.d. case, our PrPl-EB CIs attain an asymptotic width of 2σ2log(2/α)/n a.s. regardless of how c is chosen. Thus c=1/2 is simply a default practical suggestion, and it is unlikely that changing c will be able to improve much on this default. While Jiang and Yu correctly point out that H. Wang and Ramdas (2023) derive CSs for the mean under weaker assumptions than us (such as a known (1+δ)th moment), it is important to keep in mind that their bounds are not adaptive to the unknown variance (and provably cannot be without further assumptions), while that was a focal point of our work. Moreover, their use of the PrPl technique in the unbounded case was inspired by our work. We thank Ruodu Wang for his kind words and for the interesting discussion of admissibility. We fully agree with Wang that this e-testing approach is an incredibly versatile (model-free) tool that can be used to derive tests and confidence sets for other functionals, and as discussed in a previous paragraph, we are quite enthusiastic about his clean and elegant application to risk measures among other quantities (Q. Wang et al., 2022). This is also directly relevant to Jiang and Yu’s interest in studying other functionals such as the mode, expectile, and extremile, as is the work on elicitable functionals by Casgrain et al. (2022), and the earlier mentioned works of Howard and Ramdas (2022) and Manole and Ramdas (2023) for quantiles and CDFs. Wang brings up a sharp point about some betting strategies using features of the distribution—such as the mean, variance, etc.—to arrive at (λt)t=1∞⁠, and that this may be tailored towards settings with some degree of stationarity. We have two responses. (1) Firstly, some betting strategies such as an average of wealths from constant strategies (e.g. hgKelly) are agnostic to any distributional features and yet seem to do well in practice. (2) Second and finally, inherent in the problem statement itself is a ‘stationarity’ condition, i.e. the fact that E(Xt∣X1,…,Xt−1)=μ for some fixed and unchanging μ. There are other works that derive CSs for functionals like the running average mean so far μ~t:=1t∑i=1tμi⁠, where μi=E[Xi∣X1,…,Xi−1]⁠; see Howard et al. (2021), Waudby-Smith, Arbour, et al. (2021), and Waudby-Smith et al. (2022). We thank Steve Howard—whose own prior work on time-uniform confidence sequences forms an important foundation on which we build—for his kind words. We echo his desire to see more collaborations between the fields of statistics, information theory, online learning, and finance; our other works have already benefited from such collaborations. Howard raises three important questions regarding vector-valued functional estimation, comparisons to CLT-based CIs, and non-stationarity. First, notice that it is in principle straightforward to extend our martingales to vector-valued observations and parameters by writing Kt(μ):=∏i=1t(1+λi⊤(Xi−μ)) where λi,Xi,μ∈Rd⁠. While this may enable computationally efficient testing procedures, computing the set {m∈[0,1]d:Kt(m)<1/α} appears expensive, and we do not know of neat shortcuts. Some progress towards efficient variance-adaptive confidence sets in higher dimensions has recently been made in Whitehouse et al. (2023), where the authors derive multivariate generalisations of the empirical-Bernstein technique of Howard et al. (2021), amongst other self-normalised time-uniform bounds. As for comparisons to CLT-based CIs, we are in agreement that non-asymptotics are sometimes a hard sell, but whether or not this ought to be the case is a separate philosophical matter altogether. Nevertheless, one can derive non-asymptotic CIs with CLT-like asymptotic performance via Berry–Esseen bounds (assuming an upper bound on the third moment, which is the case in our [0,1]-bounded setting) and a recent preprint by Austern and Mackey (2022) seems to have taken this further. We share Howard’s enthusiasm for these types of results. Finally, Howard asks whether there are counterexamples showing that betting-based CSs break in settings where Waudby-Smith and Ramdas us that our capital processes being martingales is to the observations being with mean and this in the of For an empirical Waudby-Smith et al. Figure this where an empirical by Howard et al., 2021 but for to the running mean while a betting-based to. are the betting-based CSs are tighter in practice. We share the of Philip and Phan that confidence intervals will form important building for machine applications including and et al. a confidence to proof has the for (see also & We are also in agreement regarding the empirical performance of these and we that a proof of would be an important We would like to thank not only for his discussion of our paper, but also for many of the in probability and on which our work historical the method of connections to in statistics, and probability, universal online learning, and his Martin brings up a of our confidence intervals (and even using tools from probability such as sets and they are for problems like to be but it is an and explore a new (and to application of our to a distribution. We were with this and we reading about their We hope that their to in theory or in We to the votes of thanks by Peter Grünwald and Gergely us with Grünwald on the i.i.d. setting are We by for his words at the of the We share his enthusiasm for applications of betting and confidence sequences to theory, and indeed recent work of et al. (2023) and work by et al. (2023) such The latter on the setting using concentration and the former a rather and unbounded settings. We enjoy neat to and are in agreement that procedures are not only by but also whether may have the been but not In of the of confidence sequences are to their validity under Grünwald also makes some thoughtful on the of inference in for a setting as as estimating of random a approach would by an prior distribution that appears difficult to because it all possible continuous and with any within (and their In non-asymptotic guarantees are by our simple and betting martingales other confidence This phenomenon was in a by that statistical problems do not have to be as one is to but is for To estimate a an to a prior on the of all and then find the The not about the of the distribution and can on much regarding optimality of and are in a recent and Ramdas (2023) show bounds and upper that several betting-based confidence intervals and sequences of the paper are in the of an in which the is Finally, to the of the of We are grateful to Gergely for his kind words about our paper and we share his interest in to the the betting framework may We share his enthusiasm for the work of and only to the and not the other three in our we are quite by repeated use of the to a paper that more than a We that an earlier paper by these authors & not make connections to testing nor to confidence nor it use the processes directly as we it is of to the sequential testing and estimation literature by and and the probability literature by and In their 2021 preprint was their first work which time-uniform CSs that the form in our paper. is not to their in any and we are about their new We simply to to the discussion of three in our paper, and while we agree with some of in we will some in that we could have used betting strategies with more guarantees on their writing what betting one use the confidence an point of that we the use of betting but we to and are with using any that works well for our statistical and indeed point out that any does in a valid confidence Nevertheless, we agree that it is not obvious which betting one use in However, despite our own use of such there is no to which the of a supermartingale guarantees widths of the derived confidence For an one only to the supermartingale CI width is a.s. by a constant betting which itself does not the it depends on the α and not on the true mean For this many of our strategies were derived with tightness of width in and not the For example, our Predictable Plug-in Bernstein confidence intervals attain an asymptotic width of 2σ2log(2/α)/n that of Bernstein confidence intervals exactly including (Waudby-Smith & Ramdas, 2023, and this was the only fully empirical confidence that we were aware of with such a about recent work of and Ramdas (2023) have also that our betting confidence intervals with the are at least as It is not known whether the guarantee when using strategies in the derivation of confidence To confidence intervals and sequences use different betting and in the width for may not be the the points to and (2021) to betting strategies that have but only one of their three has an asymptotic width guarantee (and it is the of the they for sample However, the work of and Ramdas (2023) that a wide of strategies enjoy certain a connection between the of the and the of the in the as to in our reply to confidence intervals depend on the order in which the is to the points to and (2021) as of confidence sequences but as their and do depend on the order of the since they the running of the confidence sequence at each their to not running their bounds would then indeed be and their guarantees. We that not all of our bounds depend on the order of the one confidence sequence that can be in several is that from the in Waudby-Smith and Ramdas which is and does not depend on the order of the as the running is not that our for the on of the somewhat We and may rather for a betting that is from the if that is an important However, while procedures to when the is we are much more with this. we do not view the of the as a if it also with For example, in a different modern inference is possible to sample so that the and intervals do not the 2022). The can be about the likelihood of et al. which can be made in the as by different and even of such as for In many of these cases, some of concentration of in to ensure that the practical across of the we have this to be true in our as so that the mathematical on the order of the appears to be Ramdas and Manole (2023) show how or methods can more we do not see as to be at all and in some cases, it could even be if it procedures in the latter and if the has practical It is that we the of Grünwald and indeed all the and we hope our to the As has been by and recent the of betting confidence in far the case in this paper, but we hope that the we set in studying this problem provide mathematical and in of for other *(abstract via OpenAlex)*

SiM Latent Variables & Mixtures Optimization & Learning Theory · 发表于 2023-10-18

Unsupervised learning for medical data: A review of probabilistic factorization methods

Dorien Neijzen, Gerton Lunter

摘要

We review popular unsupervised learning methods for the analysis of high‐dimensional data encountered in, for example, genomics, medical imaging, cohort studies, and biobanks. We show that four commonly used methods, principal component analysis, K‐means clustering, nonnegative matrix factorization, and latent Dirichlet allocation, can be written as probabilistic models underpinned by a low‐rank matrix factorization. In addition to highlighting their similarities, this formulation clarifies the various assumptions and restrictions of each approach, which eases identifying the appropriate method for specific applications for applied medical researchers. We also touch upon the most important aspects of inference and model selection for the application of these methods to health data.

JASA Networks & Graphs · 发表于 2023-10-18

A Tale of Two Datasets: Representativeness and Generalisability of Inference for Samples of Networks

Pavel N. Krivitsky, Pietro Coletti, Niel Hens

摘要

The last two decades have seen considerable progress in foundational aspects of statistical network analysis, but the path from theory to application is not straightforward.Two large, heterogeneous samples of small networks of within-household contacts in Belgium were collected using two different but complementary sampling designs: one smaller but with all contacts in each household observed, the other larger and more representative but recording contacts of only one person per household.We wish to combine their strengths to learn the social forces that shape household contact formation and facilitate simulation for prediction of disease spread, while generalising to the population of households in the region.To accomplish this, we describe a flexible framework for specifying multi-network models in the exponential family class and identify the requirements for inference and prediction under this framework to be consistent, identifiable, and generalisable, even when data are incomplete; explore how these requirements may be violated in practice; and develop a suite of quantitative and graphical diagnostics for detecting violations and suggesting improvements to candidate models.We report on the effects of network size, geography, and household roles on household contact patterns (activity, heterogeneity in activity, and triadic closure). *(abstract via OpenAlex)*

JASA Machine Learning Optimization & Learning Theory · 发表于 2023-10-17

Value Enhancement of Reinforcement Learning via Efficient and Robust Trust Region Optimization

Chengchun Shi, Zhengling Qi, Jianing Wang, Fan Zhou

摘要

Reinforcement learning (RL) is a powerful machine learning technique that enables an intelligent agent to learn an optimal policy that maximizes the cumulative rewards in sequential decision making. Most of methods in the existing literature are developed in online settings where the data are easy to collect or simulate. Motivated by high stake domains such as mobile health studies with limited and pre-collected data, in this article, we study offline reinforcement learning methods. To efficiently use these datasets for policy optimization, we propose a novel value enhancement method to improve the performance of a given initial policy computed by existing state-of-the-art RL algorithms. Specifically, when the initial policy is not consistent, our method will output a policy whose value is no worse and often better than that of the initial policy. When the initial policy is consistent, under some mild conditions, our method will yield a policy whose value converges to the optimal one at a faster rate than the initial policy, achieving the desired “value enhancement” property. The proposed method is generally applicable to any parameterized policy that belongs to certain pre-specified function class (e.g., deep neural networks). Extensive numerical studies are conducted to demonstrate the superior performance of our method. Supplementary materials for this article are available online. *(abstract via OpenAlex)*

JASA Other · 发表于 2023-10-17

Unified Unconditional Regression for Multivariate Quantiles, M-Quantiles, and Expectiles

Luca Merlo, Lea Petrella, Nicola Salvati, Nikos Tzavidis

摘要

In this article, we develop a unified regression approach to model unconditional quantiles, M-quantiles and expectiles of multivariate dependent variables exploiting the multidimensional Huber’s function. To assess the impact of changes in the covariates across the entire unconditional distribution of the responses, we extend the work of Firpo, Fortin, and Lemieux by running a mean regression of the recentered influence function on the explanatory variables. We discuss the estimation procedure and establish the asymptotic properties of the derived estimators. A data-driven procedure is also presented to select the tuning constant of the Huber’s function. The validity of the proposed methodology is explored with simulation studies and through an application using the Survey of Household Income and Wealth 2016 conducted by the Bank of Italy. Supplementary materials for this article are available online. *(abstract via OpenAlex)*

SiM Optimization & Learning Theory Genetics & Omics Testing & Uncertainty · 发表于 2023-10-17

Rank selection for non‐negative matrix factorization

Yun Cai, Hong Gu, Toby Kenney

摘要

Non‐Negative Matrix Factorization (NMF) is a widely used dimension reduction method that factorizes a non‐negative data matrix into two lower dimensional non‐negative matrices: one is the basis or feature matrix which consists of the variables and the other is the coefficients matrix which is the projections of data points to the new basis. The features can be interpreted as sub‐structures of the data. The number of sub‐structures in the feature matrix is also called the rank. This parameter controls the model complexity and is the only tuning parameter for the NMF model. An appropriate rank will extract the key latent features while minimizing the noise from the original data. However due to the large amount of optimization error always present in the NMF computation, the rank selection has been a difficult problem. We develop a novel rank selection method based on hypothesis testing, using a deconvolved bootstrap distribution to assess the significance level accurately. Through simulations, we compare our method with a rank selection method based on hypothesis testing using bootstrap distribution without deconvolution and a method based on cross‐validation; we demonstrate that our method is not only accurate at estimating the true ranks for NMF, especially when the features are hard to distinguish, but also efficient at computation. When applied to real microbiome data (eg, OTU data and functional metagenomic data), our method also shows the ability to extract interpretable subcommunities in the data.

SiM Machine Learning Diagnostics & Prediction · 发表于 2023-10-17

Calibrating machine learning approaches for probability estimation: A comprehensive comparison

Francisco M. Ojeda, Max L. Jansen, Alexandre Thiéry, Stefan Blankenberg, Christian Weimar, Matthias Schmid, Andreas Ziegler

摘要

Statistical prediction models have gained popularity in applied research. One challenge is the transfer of the prediction model to a different population which may be structurally different from the model for which it has been developed. An adaptation to the new population can be achieved by calibrating the model to the characteristics of the target population, for which numerous calibration techniques exist. In view of this diversity, we performed a systematic evaluation of various popular calibration approaches used by the statistical and the machine learning communities for estimating two‐class probabilities. In this work, we first provide a review of the literature and, second, present the results of a comprehensive simulation study. The calibration approaches are compared with respect to their empirical properties and relationships, their ability to generalize precise probability estimates to external populations and their availability in terms of easy‐to‐use software implementations. Third, we provide code from real data analysis allowing its application by researchers. Logistic calibration and beta calibration, which estimate an intercept plus one and two slope parameters, respectively, consistently showed the best results in the simulation studies. Calibration on logit transformed probability estimates generally outperformed calibration methods on nontransformed estimates. In case of structural differences between training and validation data, re‐estimation of the entire prediction model should be outweighted against sample size of the validation data. We recommend regression‐based calibration approaches using transformed probability estimates, where at least one slope is estimated in addition to an intercept for updating probability estimates in validation studies.

SiM Survival Analysis Nonparametric & Semiparametric High-Dimensional · 发表于 2023-10-16

On semiparametric accelerated failure time models with time‐varying covariates: A maximum penalised likelihood estimation

Ding Ma, Jun Ma, Petra L. Graham

摘要

The accelerated failure time (AFT) model offers an important and useful alternative to the conventional Cox proportional hazards model, particularly when the proportional hazards assumption for a Cox model is violated. Since an AFT model is basically a log‐linear model, meaningful interpretations of covariate effects on failure times can be made directly. However, estimation of a semiparametric AFT model imposes computational challenges even when it only has time‐fixed covariates, and the situation becomes much more complicated when time‐varying covariates are included. In this paper, we propose a penalised likelihood approach to estimate the semiparametric AFT model with right‐censored failure time, where both time‐fixed and time‐varying covariates are permitted. We adopt the Gaussian basis functions to construct a smooth approximation to the nonparametric baseline hazard. This model fitting method requires a constrained optimisation approach. A comprehensive simulation study is conducted to demonstrate the performance of the proposed method. An application of our method to a motor neuron disease data set is provided.

JRSSB Other 讨论 / 评论 · 发表于 2023-10-16

Art Owen’s contribution to the Discussion of ‘Estimating means of bounded random variables by betting’ by Waudby-Smith and Ramdas

Art B Owen

摘要

I congratulate the authors on some very interesting work making connections between likelihood, betting and martingales. What caught my eye was the connection to the empirical likelihood (EL) (Owen, 2001) and the dual likelihood (Mykland, 1995). The hindsight optimal λ solves 0=∑i=1t(xi−m)/(1+λT(xi−m))=0 corresponding to the observation weights wi∝1/(1+λT(xi−m)) used in EL. Two common alternatives use weights wi∝(1−λT(xi−m)) and wi∝exp(−λT(xi−m)) arising from L2 and entropy criteria, respectively (for different vectors λ). The entropy weights connect to exponential tilting and logistic regression; see, for instance, Hainmueller (2012). The EL weights perform a kind of reciprocal tilting that gives them some special power properties. Kitamura (2003) shows that EL tests cannot be dominated by other regular tests for moment restrictions, in a large deviations sense. His result is a nonparametric counterpart to the likelihood test optimality result of Hoeffding (1965) for multinomial distributions. Lazar and Mykland (1998) find that for true parametric models, EL matches their power to second order and at third order either the empirical or the parametric tests could have greater power. In some overspecified moment models, EL has such high power for detecting lack of fit that, under lack of fit, there cannot exist any pseudo-true value of the parameter for which the maximum EL estimate is root-n consistent (Schennach, 2007). We can now add the present authors’ hindsight optimality to this list of power properties. A similar power optimality is achieved by the confidence bands of Berk and Jones (1979). They use the most significant binomial likelihood ratio from all n order statistics to form confidence bands with greater power than any weighted Kolmogorov–Smirnov test. It would be very interesting to see if the authors’ methods could produce an always valid version of the Berk–Jones bands. *(abstract via OpenAlex)*

SiM Testing & Uncertainty · 发表于 2023-10-15

Hypothesis testing procedure for binary and multi‐class F1‐scores in the paired design

Kanae Takahashi, Kouji Yamamoto, Aya Kuchiba, Ayumi Shintani, Tatsuki Koyama

摘要

In modern medicine, medical tests are used for various purposes including diagnosis, disease screening, prognosis, and risk prediction. To quantify the performance of the binary medical test, we often use sensitivity, specificity, and negative and positive predictive values as measures. Additionally, the ‐score, which is defined as the harmonic mean of precision (positive predictive value) and recall (sensitivity), has come to be used in the medical field due to its favorable characteristics. The ‐score has been extended for multi‐class classification, and two types of ‐scores have been proposed for multi‐class classification: a micro‐averaged ‐score and a macro‐averaged ‐score. The micro‐averaged ‐score pools per‐sample classifications across classes and then calculates the overall ‐score, whereas the macro‐averaged ‐score computes an arithmetic mean of the ‐scores for each class. Additionally, Sokolova and Lapalme gave an alternative definition of the macro‐averaged ‐score as the harmonic mean of the arithmetic means of the precision and recall over classes. Although some statistical methods of inference for binary and multi‐class ‐scores have been proposed, the methodology development of hypothesis testing procedure for them has not been fully progressing yet. Therefore, we aim to develop hypothesis testing procedure for comparing two ‐scores in paired study design based on the large sample multivariate central limit theorem.

SiM Survival Analysis Bayesian Methods Nonparametric & Semiparametric · 发表于 2023-10-15

Clayton copula for survival data with dependent censoring: An application to a tuberculosis treatment adherence data

Silvana Schneider, Rodrigo Citton P. dos Reis, Maicon M. F. Gottselig, Patrícia Fisch, Daniela Riva Knauth, Álvaro Vigo

摘要

Ignoring the presence of dependent censoring in data analysis can lead to biased estimates, for example, not considering the effect of abandonment of the tuberculosis treatment may influence inferences about the cure probability. In order to assess the relationship between cure and abandonment outcomes, we propose a copula Bayesian approach. Therefore, the main objective of this work is to introduce a Bayesian survival regression model, capable of taking into account the dependent censoring in the adjustment. So, this proposed approach is based on Clayton's copula, to provide the relation between survival and dependent censoring times. In addition, the Weibull and the piecewise exponential marginal distributions are considered in order to fit the times. A simulation study is carried out to perform comparisons between different scenarios of dependence, different specifications of prior distributions, and comparisons with the maximum likelihood inference. Finally, we apply the proposed approach to a tuberculosis treatment adherence dataset of an HIV cohort from Alvorada‐RS, Brazil. Results show that cure and abandonment outcomes are negatively correlated, that is, as long as the chance of abandoning the treatment increases, the chance of tuberculosis cure decreases.

SiM Latent Variables & Mixtures Survival Analysis · 发表于 2023-10-15

A lean additive frailty model: With an application to clustering of melanoma in Norwegian families

Mari Brathovde, Tron A. Moger, Odd O. Aalen, Tom Grotmol, Marit B. Veierød, Morten Valberg

摘要

Additive frailty models are used to model correlated survival data. However, the complexity of the models increases with cluster size to the extent that practical usage becomes increasingly challenging. We present a modification of the additive genetic gamma frailty (AGGF) model, the lean AGGF (L‐AGGF) model, which alleviates some of these challenges by using a leaner additive decomposition of the frailty. The performances of the models were compared and evaluated in a simulation study. The L‐AGGF model was used to analyze population‐wide data on clustering of melanoma in 2 391 125 two‐generational Norwegian families, 1960‐2015. Using this model, we could analyze the complete data set, while the original model limited the analysis to a restricted data set (with cluster sizes ). We found a substantial clustering of melanoma in Norwegian families and large heterogeneity in melanoma risk across the population, where 52% of the frailty was attributed to the 10% of the population at highest unobserved risk. Due to the improved scalability, the L‐AGGF model enables a wider range of analyses of population‐wide data compared to the AGGF model. Moreover, the methods outlined here make it possible to perform these analyses in a computationally efficient manner.

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