2024 年的期刊文章 · 第 13 页

共 1515 篇,按发表时间排,新的在前,每页 40 篇。这一页是 2024-07-01 到 2024-06-16。

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

JRSSB Bayesian Methods Testing & Uncertainty 讨论 / 评论 · 发表于 2024-07-01

Alexander Ly’s contribution to the Discussion of ‘Safe testing’ by Grünwald, de Heide, and Koolen

Alexander Ly

摘要

The article by Grünwald, de Heide, and Koolen not only presents a comprehensive approach to anytime-valid testing but also offers profound insights and inspiration for statistical practice. However, there is a potential inconvenience in cases where achieving gro(w) optimality necessitates the use of point priors. In these instances, the specified point of the point prior never vanishes from the associated anytime-valid confidence interval, and it prohibits powerful inferences for parameter values smaller than the chosen point. While a prior with full support can address these issues, selecting such a prior often lacks clear guidance from an optimality criterion. Is it worth exploring an adaptation of the gro(w) criterion to ensure that the optimal solution involves a prior with full support? If so, what would be the most effective approach to achieve this adaptation? *(abstract via OpenAlex)*

SiM Networks & Graphs Bayesian Methods High-Dimensional · 发表于 2024-06-30

Time‐varying dynamic Bayesian network learning for an fMRI study of emotion processing

Lizhe Sun, Aiying Zhang, Faming Liang

摘要

This article presents a novel method for learning time‐varying dynamic Bayesian networks. The proposed method breaks down the dynamic Bayesian network learning problem into a sequence of regression inference problems and tackles each problem using the Markov neighborhood regression technique. Notably, the method demonstrates scalability concerning data dimensionality, accommodates time‐varying network structure, and naturally handles multi‐subject data. The proposed method exhibits consistency and offers superior performance compared to existing methods in terms of estimation accuracy and computational efficiency, as supported by extensive numerical experiments. To showcase its effectiveness, we apply the proposed method to an fMRI study investigating the effective connectivity among various regions of interest (ROIs) during an emotion‐processing task. Our findings reveal the pivotal role of the subcortical‐cerebellum in emotion processing.

SiM Causal Inference Econometrics & Economics · 发表于 2024-06-30

Online causal inference with application to near real‐time post‐market vaccine safety surveillance

Lan Luo, Malcolm Risk, Xu Shi

摘要

Streaming data routinely generated by social networks, mobile or web applications, e‐commerce, and electronic health records present new opportunities to monitor the impact of an intervention on an outcome via causal inference methods. However, most existing causal inference methods have been focused on and applied to static data, that is, a fixed data set in which observations are pooled and stored before performing statistical analysis. There is thus a pressing need to turn static causal inference into online causal learning to support near real‐time monitoring of treatment effects. In this paper, we present a framework for online estimation and inference of treatment effects that can incorporate new information as it becomes available without revisiting prior observations. We show that, under mild regularity conditions, the proposed online estimator is asymptotically equivalent to the offline oracle estimator obtained by pooling all data. Our proposal is motivated by the need for near real‐time vaccine effectiveness and safety monitoring, and our proposed method is applied to a case study on COVID‐19 vaccine safety surveillance.

JRSSB Bayesian Methods 讨论 / 评论 · 发表于 2024-06-28

Stefano Rizzelli’s contribution to the Discussion of ‘Safe testing’ by Grünwald, de Heide, and Koolen

Stefano Rizzelli

摘要

I congratulate the authors on this extremely interesting work, having the indisputable merit to introduce formal optimality criteria for testing (GRO, GROW, REGROW) using e-variables, under increasingly complex conditions. An illuminating description of how they relate to each other is offered at the beginning of Section 4. Notably, if the alternative Θ1={θ*} is simple and the true data generating distribution is Pθ*⁠, the GRO(θ*) can be linked (under regularity assumptions) to a notion of oracle prior distributionW0*(⋅)≡W0*(θ*;⋅) on the null set Θ0⁠, yielding via PW0*=∫Pθ(⋅)W0*(dθ) the reverse information projection (Csiszár & Matus, 2003; Grünwald & Mehta, 2020) of Pθ* on the class of Bayesian marginal distributions PW⁠, which stem from all priors W on the null set (see also Koolen & Grünwald, 2022, Theorem 12). With W0*(θ*;⋅) at hand, an oracle would construct the optimal e-variable as E*=pθ*(Y)/pW0*(Y)⁠. In practice, θ* is unknown and so is W0*(θ*;⋅)⁠. One may thus only target e-variables whose expected growth is close to the optimum no matter what θ* actually is. The GROW and REGROW criteria are introduced to this aim. A somewhat natural question is whether the oracle prior can be sensibly inferred from the data, producing where W^0*≡W0*(θ^;⋅) and θ^ is a statistic that conveys information on θ*⁠, in such a way that E^* is an approximately optimal e-variable. Similarly, whenever the statistical model includes a main parameter δ and a nuisance γ, and if the GROW e-variable to test δ is known to be E*=g(Y;γ) in the idealistic case where γ is given, one wonders whether in practice a “good” e-variable E^*=g(Y;γ^) can be obtained by simply plugging a suitable estimator γ^ of γ in g(Y;⋅)⁠. These solutions, which could be regarded as shortcuts to the use of the more rigorous GROW and REGROW criteria, would be reminiscent of empirical Bayes practices in Bayesian inference (see, e.g. Petrone et al., 2014, Sections 2 and 4.2). Intuition can be gathered from the prototypical case of an i.i.d. Gaussian sample Y=(Y1,…,Yn) from N(μ,σ2) , where Θ0={(0,σ):σ>0} is tested versus Θ1={(μ,σ):μ≠0,σ>0}⁠. Exploiting Theorem 4.3 in Koolen and Grünwald (2022), to construct (1) with naive choice θ^=(n−1∑i=1nYi,n−1∑i=1nYi2−(n−1∑i=1nYi)2) immediately reveals that E^* is not an e-variable. In its simplicity, this example suggests that the construction of effective e-test statistics requires careful evaluation, emphasizing the importance of judiciously devised criteria. *(abstract via OpenAlex)*

JASA Spatial & Time Series Nonparametric & Semiparametric Meta-Analysis & Data Integration · 发表于 2024-06-28

Distributed Heterogeneity Learning for Generalized Partially Linear Models with Spatially Varying Coefficients

Shan Yu, Guannan Wang, Li Wang

摘要

Spatial heterogeneity is of great importance in social, economic, and environmental science studies. The spatially varying coefficient model is a popular and effective spatial regression technique to address spatial heterogeneity. However, accounting for heterogeneity comes at the cost of reducing model parsimony. To balance flexibility and parsimony, this article develops a class of generalized partially linear spatially varying coefficient models which allow the inclusion of both constant and spatially varying effects of covariates. Another significant challenge in many applications comes from the enormous size of the spatial datasets collected from modern technologies. To tackle this challenge, we design a novel distributed heterogeneity learning (DHL) method based on bivariate spline smoothing over a triangulation of the domain. The proposed DHL algorithm has a simple, scalable, and communication-efficient implementation scheme that can almost achieve linear speedup. In addition, this article provides rigorous theoretical support for the DHL framework. We prove that the DHL constant coefficient estimators are asymptotic normal and the DHL spline estimators reach the same convergence rate as the global spline estimators obtained using the entire dataset. The proposed DHL method is evaluated through extensive simulation studies and analyses of U.S. loan application data. *(abstract via PubMed)*

JRSSB Testing & Uncertainty 讨论 / 评论 · 发表于 2024-06-27

Vladimir Vovk’s contribution to the Discussion of ‘Safe testing’ by Grünwald, de Heide, and Koolen

Vladimir Vovk

摘要

The main result of this paper (Theorem 1) is elegant and satisfying, but in this comment I will concentrate on one of its limitations, namely the condition infW0D(Q‖PW0)<∞⁠. (Analogous conditions are present in all generalizations given in the paper.) Even in the toy binary case with an infinite horizon, this condition is typically violated. Consider, e.g., the case of Markov alternative hypotheses Q considered in Ramdas et al. (2022): a nonIID Markov probability measure (or a Jeffreys-type mixture of such measures) and any exchangeable probability measure are mutually singular. A possible counter-argument is that infinities are irrelevant in real life, and we can always truncate a potentially infinite sequence of observations, but the problem is that our results will depend on the details of truncation. If we do not know the number of observations in advance and just would like to have an online measure of evidence found against the null hypothesis (such as a test martingale), we cannot use Theorem 1. It is interesting that narrow parametric models with a given simple alternative (as in Ramdas et al., 2022, or Vovk et al., 2022, Sect. 9.2) admit a simple solution: just replace Grünwald et al.’s average in the denominator of E* (see Theorem 1) by the supremum. But this idea does not work for nontrivial nonparametric models (Vovk et al., 2022, Remark 9.8). Our book (Vovk et al., 2022, Part III) presents a framework, which we call conformal testing, for testing nonparametric null hypotheses (first of all the hypothesis of exchangeability) in general situations typical of machine learning. Restricting ourselves to the class of conformal test martingales with a given conformity measure, we construct an optimal testing procedure, the Bayes–Kelly algorithm, whose optimality property is akin to Grünwald et al.’s one; it is stated explicitly in the extended version of this comment (Vovk, 2024, Theorem 3.2). We spell out the details of the procedure and prove general optimality properties (not restricted to conformal testing) in the binary case (Vovk et al., 2022, Sect. 9.2), but the procedure itself is general, and the efficiency of conformal testing is demonstrated in empirical studies reported in (Vovk et al., 2022, Chap. 8). Of course, it is not surprising that Theorem 1 has limitations (and the authors discuss some), but it is a great first step that opens up interesting directions of further research. One such direction is to design optimal (in Grünwald et al.’s sense) conformity measures in conformal testing. The following contributions were received in writing after the meeting: *(abstract via OpenAlex)*

JRSSB Testing & Uncertainty 讨论 / 评论 · 发表于 2024-06-27

Proposer of the vote of thanks to Grünwald, de Heide, and Koolen and contribution to the Discussion of ‘Safe testing’

Ruodu Wang

摘要

The contribution of Peter Grünwald, Rianne de Heide, and Wouter Koolen (GDK henceforth) is an important piece in the theory of e-values and e-processes. Methods based on e-values are developed to overcome several drawbacks (but not all, of course) in traditional methods based on p-values and those involving statistical malpractice. Due to GDK and a few others around the same time (around 2019) the research theme of e-values has been growing rapidly and healthily over the past few years in multiple exciting directions. An overview was provided by Ramdas et al. (2023), but there has been much more progresses since then. In addition to the various statistical advantages discussed by GDK, there is a notable advantage from a mathematical point of view: The theory of martingales is essential for this methodology, allowing statisticians to appreciate and contribute to this classic theory in probability. Conversely, the theory of e-values has successfully attracted probabilists and financial mathematicians to join the journey of statistics. Methods based on e-values have Type I error guarantee under finite sample and optional stopping, as well illustrated in GDK and the recent literature. These methods are sometimes criticized for their power. Traditionally, power under a specific alternative probability Q is measured by Q(rejecting the null)⁠. For a fixed-sample testing problem, with this definition of power and a prespecified Type I error control, methods based on e-values typically have less power compared to those based on properly chosen p-values (a result of the Neymann–Pearson lemma). Nevertheless, power in the context of e-test may be defined differently: For a specific alternative Q, the growth-rate optimality in GDK is a criterion to maximize ΠQ(E):=EQ[logE] over e-variables E; see Theorem 1 of GDK. Section 2 of Shafer (2021) also has an extensive discussion on this quantity, with an intimate connection to Kelly’s criterion in investment theory. The desirability of optimizing ΠQ(E) and its composite-alternative version is crucial for the main results of GDK. The quantity ΠQ(E) is called the e-power of E under Q by Vovk and Wang (2024), where it is used to define relative efficiency of e-tests. Many papers on e-tests have adapted this criterion as a notion of power. Despite the intuitive reasons to use ΠQ(E) to represent power, there are several issues to discuss. A main argument in GDK and related papers for using ΠQ(E) is that, in a sequential setting, e-variables E1,E2,… are meant to be multiplicative. For simplicity, suppose that ΠQ(Et) is the same for each t∈N⁠. Under some conditions and in an asymptotic sense, ΠQ(E1) is the average rate of growth as t→∞ for the e-processes M=(Mt)t∈N given by However, in practice, e-tests are used mainly in finite sample and often not multiplicative; for instance, Waudby-Smith and Ramdas (2024) use e-processes built from a betting strategy or a weighted arithmetic average. In such cases, the interpretation of the e-power of individual e-variables is not clear. Even under multiplication and independence, we can find simple, perhaps pathological, examples to scrutinize the intuition of the e-power measured by ΠQ⁠. Below we use P for a simple null. Suppose that E1,E2,… are independent e-variables for P. If E1 satisfies ΠQ(E1)=−∞ like in (a) above and Et has finite e-power for all t≥2⁠, then the process M built in (1) satisfies ΠQ(Mt)=−∞ for all t∈N⁠, no matter how fast it grows after t=1 (in particular, one can easily build M with Mt→∞ in probability under Q). In each example above, we find e-variables or e-processes that are intuitively powerful but have negative infinite e-power. Taking a reciprocal in (b) and (c), we also find e-processes with positive infinite e-power but declines to 0 with probability 1 under Q. On the positive side, Proposition 1 of Vovk and Wang (2024) showed that ΠQ enjoys a natural and desirable property: Combining several independent e-tests of positive power (measured by ΠQ⁠) by partial products, averages, or U-statistics always results in positive power. Of course, it is not my intention to say that ΠQ is not a suitable notion of e-power. The purpose of the discussion here is to explain that e-power deserves more discussions to fully understand its advantages and limitations. I anticipate that this issue will be solved by researchers in the field in a few years, with both theoretical and empirical advances, and the functional ΠQ will be accepted as the correct notion of e-power, with properly acknowledged caveats. *(abstract via OpenAlex)*

JRSSB Testing & Uncertainty 讨论 / 评论 · 发表于 2024-06-27

Neil Dey, Ryan Martin, and Jonathan P. Williams’ contribution to the Discussion of ‘Safe Testing’ by Grünwald, de Heide, and Koolen

Neil Dey, Ryan Martin, Jonathan P Williams

摘要

Congratulations to Grünwald, de Heide, and Koolen (GHK) for their contribution to the exciting and rapidly growing literature on safe/anytime-valid inference. GHK’s proposal focused on problems in which a statistical model is available that determines the quantity of interest—the model parameter—and provides a likelihood function that drives the e-value construction. When data are observed and analysed sequentially, however, the data analyst typically cannot say with any certainty what statistical model might be appropriate. Our comments below focus on what can be done along the lines of safe/anytime-valid inference without a correctly specified statistical model. Two recent developments deserve mention. First, Park, Balakrishnan, and Wasserman, in their recent arXiv paper (https://arxiv.org/abs/2307.04034) develop a version of the universal inference framework (Wasserman et al., 2020) that accommodates model misspecification but is not designed for the online setting. Second, the nonparametric approach in Waudby-Smith and Ramdas (2024) offers anytime-valid inference, but focuses exclusively on the mean of the data-generating process. In our recent arXiv paper (https://arxiv.org/abs/2402.00202), we take a different approach, starting from the basic question: What is the inferential target? From classical M-estimation to modern machine learning, the inferential target is often the solution to a suitable optimization problem. These solutions can be cast as risk minimizers: Given a loss function ℓθ(z)⁠, the inferential target is θ⋆=argminθR(θ)⁠, with R(θ)=∫ℓθ(z)P(dz) the risk. We generalize the universal inference framework by replacing the statistical model’s negative log-likelihood with a multiple of the empirical risk, Rn(θ)=n−1∑i=1nℓθ(Zi)⁠, and prove that it offers safe, anytime-valid inference on risk minimizers. Circling back to GHK’s focus on growth rate optimality, we show that our generalized universal e-process for testing H0:θ⋆∈Θ0 vs. H1:θ⋆∈Θ1⁠, has (asymptotic) log-growth rate The above claim is analogous to Theorem 2 in the recent arXiv report by Dixit and Martin (https://arxiv.org/abs/2309.13441). Our conjecture is that (1) gives the ‘optimal’ asymptotic first-order growth rate achievable by an e-process relative to the given learning problem (P,ℓ)⁠. If our conjecture is true, then given the similarities between GHK’s growth rate and (1), one could approach this more general problem similarly to GHK, by defining an e-process based on a growth rate optimality criterion, and we expect its logarithm would be proportional to an empirical risk difference. Further, the corresponding ‘growth rate optimal’ e-process would have to be similar to what we proposed in our aforementioned paper. *(abstract via OpenAlex)*

JRSSB Other 讨论 / 评论 · 发表于 2024-06-27

Martin Larsson, Aaditya Ramdas, and Johannes Ruf’s contribution to the Discussion of ‘Safe testing’ by Grünwald, de Heide, and Koolen

Martin Larsson, Aaditya Ramdas, Johannes Ruf

摘要

We congratulate the authors on a stimulating article that has inspired many follow-up works. We discuss the central concept of growth-rate optimality (GRO). For a point null and alternative, the GRO e-variable is their likelihood ratio (Shafer, 2021). The generalization to composite nulls takes centre stage in Grünwald et al. (2024). Their use of the reverse information projection (RIPr) builds on Li (1999), and has been recently extended in Lardy et al. (2023). Inspired by their centrality, we report on our efforts to significantly generalize these results. Our preprint (Larsson et al., 2024) considers testing a composite null P against a point alternative Q⁠. We establish a powerful result: under no conditions whatsoever (e.g. reference/dominating measure), there always exists a unique ‘optimal’ e-variable that we call the numeraire X*⁠: for every other e-variable X, we have EQ[X/X*]≤1⁠. In particular, X* also satisfies EQ[log(X/X*)]≤0 (log-optimality), as well as EQ[1/X*]≤1⁠. X* also identifies a particular measure P* by the definition dP*/dQ=1/X*⁠. In general, P* is a subprobability measure in the ‘bipolar’ of P⁠. In particular, X*=dQ/dP* is a generalized likelihood ratio of Q against P⁠, and is the only e-variable with such a representation. Further, P* coincides with the RIPr when additional assumptions are made that are required for the latter to exist [such as Q≪P for some P∈P⁠, σ-convexity of P⁠, and finite maximum description gain in Lardy et al. (2023)]. Thus, we contend that going forward, absent any assumptions on Q,P⁠, the above should be the default definition of the RIPr. We also present a more general optimality theory that studies general concave utilities (beyond the logarithm), exemplified by a reverse Rényi projection, which also exists under no conditions. Surprisingly, our results imply that the universal inference e-variable (Wasserman et al., 2020) is (essentially) always inadmissible. Practically, the numeraire is now on a somewhat equal footing with universal inference. While the latter seems easy to apply, it is has only been applicable in a few nonparametric situations thus far (like testing log-concavity). In contrast, we explicitly calculate the numeraire and RIPr in many nonparametric settings where there is no reference measure, and universal inference does not apply; we explore several such cases in the paper (testing symmetry, sub-Gaussian means, bounded means). We are currently working to extend these ideas to composite alternatives using standard techniques like the method of mixtures, and to sequential settings using the mixture and plug-in methods; see Ramdas et al. (2023) for details. *(abstract via OpenAlex)*

JRSSB Testing & Uncertainty 讨论 / 评论 · 发表于 2024-06-26

Thorsten Dickhaus’s contribution to the Discussion of ‘Safe testing’ by Grünwald, de Heide, and Koolen

Thorsten Dickhaus

摘要

We discuss the paper “Safe Testing” from the perspective of multiple testing research. *(abstract via OpenAlex)*

JASA Networks & Graphs · 发表于 2024-06-26

Population-Level Balance in Signed Networks

Weijing Tang, Ji Zhu

摘要

Statistical network models are useful for understanding the underlying formation mechanism and characteristics of complex networks. However, statistical models for signed networks have been largely unexplored. In signed networks, there exist both positive (e.g., like, trust) and negative (e.g., dislike, distrust) edges, which are commonly seen in real-world scenarios. The positive and negative edges in signed networks lead to unique structural patterns, which pose challenges for statistical modeling. In this article, we introduce a statistically principled latent space approach for modeling signed networks and accommodating the well-known balance theory, that is, “the enemy of my enemy is my friend” and “the friend of my friend is my friend.” The proposed approach treats both edges and their signs as random variables, and characterizes the balance theory with a novel and natural notion of population-level balance. This approach guides us towards building a class of balanced inner-product models, and toward developing scalable algorithms via projected gradient descent to estimate the latent variables. We also establish non-asymptotic error rates for the estimates, which are further verified through simulation studies. In addition, we apply the proposed approach to an international relation network, which provides an informative and interpretable model-based visualization of countries during World War II. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work. *(abstract via OpenAlex)*

JASA Missing Data & Measurement Error Causal Inference · 发表于 2024-06-26

Mediation Analysis with the Mediator and Outcome Missing Not at Random

Shuozhi Zuo, Debashis Ghosh, Peng Ding, Fan Yang

摘要

Mediation analysis is widely used for investigating direct and indirect causal pathways through which an effect arises. However, many mediation analysis studies are challenged by missingness in the mediator and outcome. In general, when the mediator and outcome are missing not at random, the direct and indirect effects are not identifiable without further assumptions. We study the identifiability of the direct and indirect effects under some interpretable mechanisms that allow for missing not at random in the mediator and outcome. We evaluate the performance of statistical inference under those mechanisms through simulation studies and illustrate the proposed methods via the National Job Corps Study. *(abstract via PubMed)*

JASA Genetics & Omics · 发表于 2024-06-26

Joint Tensor Modeling of Single Cell 3D Genome and Epigenetic Data with Muscle

Kwangmoon Park, Sündüz Keleş

摘要

Emerging single cell technologies that simultaneously capture long-range interactions of genomic loci together with their DNA methylation levels are advancing our understanding of three-dimensional genome structure and its interplay with the epigenome at the single cell level. While methods to analyze data from single cell high throughput chromatin conformation capture (scHi-C) experiments are maturing, methods that can jointly analyze multiple single cell modalities with scHi-C data are lacking. Here, we introduce Muscle, a semi-nonnegative joint decomposition of Multiple single cell tensors, to jointly analyze 3D conformation and DNA methylation data at the single cell level. Muscle takes advantage of the inherent tensor structure of the scHi-C data, and integrates this modality with DNA methylation. We developed an alternating least squares algorithm for estimating Muscle parameters and established its optimality properties. Parameters estimated by Muscle directly align with the key components of the downstream analysis of scHi-C data in a cell type specific manner. Evaluations with data-driven experiments and simulations demonstrate the advantages of the joint modeling framework of Muscle over single modality modeling and a baseline multi modality modeling for cell type delineation and elucidating associations between modalities. Muscle is publicly available at https://github.com/keleslab/muscle. *(abstract via PubMed)*

Biometrics Causal Inference · 发表于 2024-06-26

Differential recall bias in estimating treatment effects in observational studies

Suhwan Bong, Kwonsang Lee, Francesca Dominici

SiM Genetics & Omics Missing Data & Measurement Error · 发表于 2024-06-26

Addressing dispersion in mis‐measured multivariate binomial outcomes: A novel statistical approach for detecting differentially methylated regions in bisulfite sequencing data

Kaiqiong Zhao, Karim Oualkacha, Yixiao Zeng, Cathy Shen, Kathleen Klein, Lajmi Lakhal‐Chaieb, Aurélie Labbe, Tomi Pastinen, Marie Hudson, Inés Colmegna, Sasha Bernatsky, Celia M. T. Greenwood

摘要

Motivated by a DNA methylation application, this article addresses the problem of fitting and inferring a multivariate binomial regression model for outcomes that are contaminated by errors and exhibit extra‐parametric variations, also known as dispersion. While dispersion in univariate binomial regression has been extensively studied, addressing dispersion in the context of multivariate outcomes remains a complex and relatively unexplored task. The complexity arises from a noteworthy data characteristic observed in our motivating dataset: non‐constant yet correlated dispersion across outcomes. To address this challenge and account for possible measurement error, we propose a novel hierarchical quasi‐binomial varying coefficient mixed model, which enables flexible dispersion patterns through a combination of additive and multiplicative dispersion components. To maximize the Laplace‐approximated quasi‐likelihood of our model, we further develop a specialized two‐stage expectation‐maximization (EM) algorithm, where a plug‐in estimate for the multiplicative scale parameter enhances the speed and stability of the EM iterations. Simulations demonstrated that our approach yields accurate inference for smooth covariate effects and exhibits excellent power in detecting non‐zero effects. Additionally, we applied our proposed method to investigate the association between DNA methylation, measured across the genome through targeted custom capture sequencing of whole blood, and levels of anti‐citrullinated protein antibodies (ACPA), a preclinical marker for rheumatoid arthritis (RA) risk. Our analysis revealed 23 significant genes that potentially contribute to ACPA‐related differential methylation, highlighting the relevance of cell signaling and collagen metabolism in RA. We implemented our method in the R Bioconductor package called “SOMNiBUS.

SiM Causal Inference Clinical Trials · 发表于 2024-06-24

Propensity score weighted multi‐source exchangeability models for incorporating external control data in randomized clinical trials

Wei Wei, Yunxuan Zhang, Satrajit Roychoudhury, the Alzheimer's Disease Neuroimaging Initiative

摘要

Among clinical trialists, there has been a growing interest in using external data to improve decision‐making and accelerate drug development in randomized clinical trials (RCTs). Here we propose a novel approach that combines the propensity score weighting (PW) and the multi‐source exchangeability modelling (MEM) approaches to augment the control arm of a RCT in the rare disease setting. First, propensity score weighting is used to construct weighted external controls that have similar observed pre‐treatment characteristics as the current trial population. Next, the MEM approach evaluates the similarity in outcome distributions between the weighted external controls and the concurrent control arm. The amount of external data we borrow is determined by the similarities in pretreatment characteristics and outcome distributions. The proposed approach can be applied to binary, continuous and count data. We evaluate the performance of the proposed PW‐MEM method and several competing approaches based on simulation and re‐sampling studies. Our results show that the PW‐MEM approach improves the precision of treatment effect estimates while reducing the biases associated with borrowing data from external sources.

JASA Machine Learning Networks & Graphs Spatial & Time Series · 发表于 2024-06-24

Neural Networks for Geospatial Data

Wentao Zhan, Abhirup Datta

摘要

Analysis of geospatial data has traditionally been model-based, with a mean model, customarily specified as a linear regression on the covariates, and a Gaussian process covariance model, encoding the spatial dependence. While nonlinear machine learning algorithms like neural networks are increasingly being used for spatial analysis, current approaches depart from the model-based setup and cannot explicitly incorporate spatial covariance. We propose NN-GLS, embedding neural networks directly within the traditional Gaussian process (GP) geostatistical model to accommodate non-linear mean functions while retaining all other advantages of GP, like explicit modeling of the spatial covariance and predicting at new locations via kriging. In NN-GLS, estimation of the neural network parameters for the non-linear mean of the Gaussian Process explicitly accounts for the spatial covariance through use of the generalized least squares (GLS) loss, thus extending the linear case. We show that NN-GLS admits a representation as a special type of graph neural network (GNN). This connection facilitates the use of standard neural network computational techniques for irregular geospatial data, enabling novel and scalable mini-batching, backpropagation, and kriging schemes. We provide methodology to obtain uncertainty bounds for estimation and predictions from NN-GLS. Theoretically, we show that NN-GLS will be consistent for irregularly observed spatially correlated data processes. We also provide a finite sample concentration rate, which quantifies the need to accurately model the spatial covariance in neural networks for dependent data. To our knowledge, these are the first large-sample results for any neural network algorithm for irregular spatial data. We demonstrate the methodology through numerous simulations and an application to air pollution modeling. We develop a software implementation of NN-GLS in the Python package geospaNN. *(abstract via PubMed)*

JASA Causal Inference High-Dimensional Meta-Analysis & Data Integration · 发表于 2024-06-24

Doubly Robust Augmented Model Accuracy Transfer Inference with High Dimensional Features

Doudou Zhou, Molei Liu, Mengyan Li, Tianxi Cai

摘要

Transfer learning is crucial for training models that generalize to unlabeled target populations using labeled source data, especially in real-world studies where label scarcity and covariate shift are common. While most research focuses on model estimation, there is limited literature on transfer inference for model accuracy despite its importance. We introduce a novel Doubly Robust Augmented Model Accuracy Transfer InferenCe (DRAMATIC) method for point and interval estimation of commonly used classification performance measures in an unlabeled target population with labeled source data. DRAMATIC derives and evaluates a potentially misspecified risk model for a binary response, leveraging high-dimensional adjustment features from both source and target data. It builds on an imputation model for the response mean and a density ratio model to characterize distributional shifts. The method constructs doubly robust estimators that are valid when either model is correctly specified and certain sparsity assumptions hold. Simulations show negligible bias in point estimation and satisfactory empirical coverage levels in confidence intervals. The utility of DRAMATIC is illustrated by transferring a genetic risk prediction model and its accuracy evaluation for type II diabetes across two patient cohorts in Mass General Brigham (MGB). Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work. *(abstract via PubMed)*

SiM Testing & Uncertainty Networks & Graphs · 发表于 2024-06-24

Differential network knockoff filter with application to brain connectivity analysis

Jiadong Ji, Zhendong Hou, Yong He, Lei Liu, Fuzhong Xue, Hao Chen, Zhongshang Yuan

摘要

The brain functional connectivity can typically be represented as a brain functional network, where nodes represent regions of interest (ROIs) and edges symbolize their connections. Studying group differences in brain functional connectivity can help identify brain regions and recover the brain functional network linked to neurodegenerative diseases. This process, known as differential network analysis focuses on the differences between estimated precision matrices for two groups. Current methods struggle with individual heterogeneity in measuring the brain connectivity, false discovery rate (FDR) control, and accounting for confounding factors, resulting in biased estimates and diminished power. To address these issues, we present a two‐stage FDR‐controlled feature selection method for differential network analysis using functional magnetic resonance imaging (fMRI) data. First, we create individual brain connectivity measures using a high‐dimensional precision matrix estimation technique. Next, we devise a penalized logistic regression model that employs individual brain connectivity data and integrates a new knockoff filter for FDR control when detecting significant differential edges. Through extensive simulations, we showcase the superiority of our approach compared to other methods. Additionally, we apply our technique to fMRI data to identify differential edges between Alzheimer's disease and control groups. Our results are consistent with prior experimental studies, emphasizing the practical applicability of our method.

SiM Genetics & Omics · 发表于 2024-06-24

A multivariate to multivariate approach for voxel‐wise genome‐wide association analysis

Qiong Wu, Yuan Zhang, Xiaoqi Huang, Tianzhou Ma, L. Elliot Hong, Peter Kochunov, Shuo Chen

摘要

The joint analysis of imaging‐genetics data facilitates the systematic investigation of genetic effects on brain structures and functions with spatial specificity. We focus on voxel‐wise genome‐wide association analysis, which may involve trillions of single nucleotide polymorphism (SNP)‐voxel pairs. We attempt to identify underlying organized association patterns of SNP‐voxel pairs and understand the polygenic and pleiotropic networks on brain imaging traits. We propose a bi‐clique graph structure (ie, a set of SNPs highly correlated with a cluster of voxels) for the systematic association pattern. Next, we develop computational strategies to detect latent SNP‐voxel bi‐cliques and an inference model for statistical testing. We further provide theoretical results to guarantee the accuracy of our computational algorithms and statistical inference. We validate our method by extensive simulation studies, and then apply it to the whole genome genetic and voxel‐level white matter integrity data collected from 1052 participants of the human connectome project. The results demonstrate multiple genetic loci influencing white matter integrity measures on splenium and genu of the corpus callosum.

SiM High-Dimensional Machine Learning Bayesian Methods · 发表于 2024-06-23

Integrative deep learning with prior assisted feature selection

Feifei Wang, Ke Jia, Yang Li

摘要

Integrative analysis has emerged as a prominent tool in biomedical research, offering a solution to the “small and large ” challenge. Leveraging the powerful capabilities of deep learning in extracting complex relationship between genes and diseases, our objective in this study is to incorporate deep learning into the framework of integrative analysis. Recognizing the redundancy within candidate features, we introduce a dedicated feature selection layer in the proposed integrative deep learning method. To further improve the performance of feature selection, the rich previous researches are utilized by an ensemble learning method to identify “prior information”. This leads to the proposed prior assisted integrative deep learning (PANDA) method. We demonstrate the superiority of the PANDA method through a series of simulation studies, showing its clear advantages over competing approaches in both feature selection and outcome prediction. Finally, a skin cutaneous melanoma (SKCM) dataset is extensively analyzed by the PANDA method to show its practical application.

SiM Bayesian Methods Survival Analysis Longitudinal Data · 发表于 2024-06-23

Bayesian survival analysis with INLA

Danilo Alvares, Janet van Niekerk, Elias Teixeira Krainski, Håvard Rue, Denis Rustand

摘要

This tutorial shows how various Bayesian survival models can be fitted using the integrated nested Laplace approximation in a clear, legible, and comprehensible manner using the INLA and INLAjoint R‐packages. Such models include accelerated failure time, proportional hazards, mixture cure, competing risks, multi‐state, frailty, and joint models of longitudinal and survival data, originally presented in the article “Bayesian survival analysis with BUGS.” In addition, we illustrate the implementation of a new joint model for a longitudinal semicontinuous marker, recurrent events, and a terminal event. Our proposal aims to provide the reader with syntax examples for implementing survival models using a fast and accurate approximate Bayesian inferential approach.

Biometrika Spatial & Time Series · 发表于 2024-06-22

Testing serial dependence or cross dependence for time series with underreporting

Keyao Wei, Lengyang Wang, Yingcun Xia

摘要

In practice, it is common for collected data to be underreported, an issue that is particularly prevalent in fields such as the social sciences, ecology and epidemiology. Drawing inferences from such data using conventional statistical methods can lead to incorrect conclusions. In this paper, we study tests for serial or cross dependence in time series data that are subject to underreporting. We introduce new test statistics, develop corresponding group-of-blocks bootstrap techniques and establish their consistency. The methods are shown via simulation studies to be efficient and are used to identify key factors responsible for the spread of dengue fever and the occurrence of cardiovascular disease.

JASA Other · 发表于 2024-06-21

Rational Kriging

V. Roshan Joseph

摘要

This article proposes a new kriging that has a rational form.It is shown that the generalized least squares estimator of the mean from rational kriging is much more well behaved than that of ordinary kriging.Parameter estimation and uncertainty quantification for rational kriging are proposed using a Gaussian process framework.A generalized version of rational kriging is also proposed, which includes ordinary and rational kriging as special cases.Extensive simulations carried out over a wide class of functions show that the generalized rational kriging performs on par or better than both ordinary and rational kriging in terms of prediction and uncertainty quantification.The only extra step needed for generalized rational kriging over ordinary kriging is the computation of Perron eigenvector of an augmented correlation matrix which can be computed in near linear time and therefore, its overall computational complexity is no more than that of ordinary kriging.The potential applications of the new kriging methods in the emulation of computationally expensive models and model calibration problems are illustrated with real and simulated examples.Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work. *(abstract via OpenAlex)*

SiM Other · 发表于 2024-06-20

Robust inference methods for meta‐analysis involving influential outlying studies

Hisashi Noma, Shonosuke Sugasawa, Toshi A. Furukawa

摘要

Meta‐analysis is an essential tool to comprehensively synthesize and quantitatively evaluate results of multiple clinical studies in evidence‐based medicine. In many meta‐analyses, the characteristics of some studies might markedly differ from those of the others, and these outlying studies can generate biases and potentially yield misleading results. In this article, we provide effective robust statistical inference methods using generalized likelihoods based on the density power divergence. The robust inference methods are designed to adjust the influences of outliers through the use of modified estimating equations based on a robust criterion, even when multiple and serious influential outliers are present. We provide the robust estimators, statistical tests, and confidence intervals via the generalized likelihoods for the fixed‐effect and random‐effects models of meta‐analysis. We also assess the contribution rates of individual studies to the robust overall estimators that indicate how the influences of outlying studies are adjusted. Through simulations and applications to two recently published systematic reviews, we demonstrate that the overall conclusions and interpretations of meta‐analyses can be markedly changed if the robust inference methods are applied and that only the conventional inference methods might produce misleading evidence. These methods would be recommended to be used at least as a sensitivity analysis method in the practice of meta‐analysis. We have also developed an R package, robustmeta, that implements the robust inference methods.

SiM Other · 发表于 2024-06-19

Renewable risk assessment of heterogeneous streaming time‐to‐event cohorts

Jie Ding, Jialiang Li, Xiaoguang Wang

摘要

The analysis of streaming time‐to‐event cohorts has garnered significant research attention. Most existing methods require observed cohorts from a study sequence to be independent and identically sampled from a common model. This assumption may be easily violated in practice. Our methodology operates within the framework of online data updating, where risk estimates for each cohort of interest are continuously refreshed using the latest observations and historical summary statistics. At each streaming stage, we introduce parameters to quantify the potential discrepancy between batch‐specific effects from adjacent cohorts. We then employ penalized estimation techniques to identify nonzero discrepancy parameters, allowing us to adaptively adjust risk estimates based on current data and historical trends. We illustrate our proposed method through extensive empirical simulations and a lung cancer data analysis.

SiM Missing Data & Measurement Error Bayesian Methods Diagnostics & Prediction · 发表于 2024-06-19

Nonparametric empirical Bayes biomarker imputation and estimation

Alton Barbehenn, Sihai Dave Zhao

摘要

Biomarkers are often measured in bulk to diagnose patients, monitor patient conditions, and research novel drug pathways. The measurement of these biomarkers often suffers from detection limits that result in missing and untrustworthy measurements. Frequently, missing biomarkers are imputed so that down‐stream analysis can be conducted with modern statistical methods that cannot normally handle data subject to informative censoring. This work develops an empirical Bayes ‐modeling method for imputing and denoising biomarker measurements. We establish superior estimation properties compared to popular methods in simulations and with real data, providing the useful biomarker measurement estimations for down‐stream analysis.

SiM Survival Analysis Machine Learning Networks & Graphs · 发表于 2024-06-18

deepAFT: A nonlinear accelerated failure time model with artificial neural network

Patrick A. Norman, Wanlu Li, Wenyu Jiang, Bingshu E. Chen

摘要

The Cox regression model or accelerated failure time regression models are often used for describing the relationship between survival outcomes and potential explanatory variables. These models assume the studied covariates are connected to the survival time or its distribution or their transformations through a function of a linear regression form. In this article, we propose nonparametric, nonlinear algorithms (deepAFT methods) based on deep artificial neural networks to model survival outcome data in the broad distribution family of accelerated failure time models. The proposed methods predict survival outcomes directly and tackle the problem of censoring via an imputation algorithm as well as re‐weighting and transformation techniques based on the inverse probabilities of censoring. Through extensive simulation studies, we confirm that the proposed deepAFT methods achieve accurate predictions. They outperform the existing regression models in prediction accuracy, while being flexible and robust in modeling covariate effects of various nonlinear forms. Their prediction performance is comparable to other established deep learning methods such as deepSurv and random survival forest methods. Even though the direct output is the expected survival time, the proposed AFT methods also provide predictions for distributional functions such as the cumulative hazard and survival functions without additional learning efforts. For situations where the popular Cox regression model may not be appropriate, the deepAFT methods provide useful and effective alternatives, as shown in simulations, and demonstrated in applications to a lymphoma clinical trial study.

SiM Causal Inference · 发表于 2024-06-18

Sensitivity analysis for principal ignorability violation in estimating complier and noncomplier average causal effects

Trang Quynh Nguyen, Elizabeth A. Stuart, Daniel O. Scharfstein, Elizabeth L. Ogburn

摘要

An important strategy for identifying principal causal effects (popular estimands in settings with noncompliance) is to invoke the principal ignorability (PI) assumption. As PI is untestable, it is important to gauge how sensitive effect estimates are to its violation. We focus on this task for the common one‐sided noncompliance setting where there are two principal strata, compliers and noncompliers. Under PI, compliers and noncompliers share the same outcome‐mean‐given‐covariates function under the control condition. For sensitivity analysis, we allow this function to differ between compliers and noncompliers in several ways, indexed by an odds ratio, a generalized odds ratio, a mean ratio, or a standardized mean difference sensitivity parameter. We tailor sensitivity analysis techniques (with any sensitivity parameter choice) to several types of PI‐based main analysis methods, including outcome regression, influence function (IF) based and weighting methods. We discuss range selection for the sensitivity parameter. We illustrate the sensitivity analyses with several outcome types from the JOBS II study. This application estimates nuisance functions parametrically – for simplicity and accessibility. In addition, we establish rate conditions on nonparametric nuisance estimation for IF‐based estimators to be asymptotically normal – with a view to inform nonparametric inference.

SiM Other · 发表于 2024-06-18

Data fusion for predicting long‐term program impacts

Michael W. Robbins, Sebastian Bauhoff, Lane Burgette

SiM Bayesian Methods Latent Variables & Mixtures · 发表于 2024-06-18

Bayesian mixture modelling with ranked set samples

Amirhossein Alvandi, Sedigheh Omidvar, Armin Hatefi, Mohammad Jafari Jozani, Omer Ozturk, Nader Nematollahi

摘要

We consider the Bayesian estimation of the parameters of a finite mixture model from independent order statistics arising from imperfect ranked set sampling designs. As a cost‐effective method, ranked set sampling enables us to incorporate easily attainable characteristics, as ranking information, into data collection and Bayesian estimation. To handle the special structure of the ranked set samples, we develop a Bayesian estimation approach exploiting the Expectation‐Maximization (EM) algorithm in estimating the ranking parameters and Metropolis within Gibbs Sampling to estimate the parameters of the underlying mixture model. Our findings show that the proposed RSS‐based Bayesian estimation method outperforms the commonly used Bayesian counterpart using simple random sampling. The developed method is finally applied to estimate the bone disorder status of women aged 50 and older.

Biometrics Bayesian Methods Machine Learning Networks & Graphs · 发表于 2024-06-18

A Bayesian convolutional neural network-based generalized linear model

Yeseul Jeon, Won Chang, Seonghyun Jeong, Sanghoon Han, Jaewoo Park

摘要

Convolutional neural networks (CNNs) provide flexible function approximations for a wide variety of applications when the input variables are in the form of images or spatial data. Although CNNs often outperform traditional statistical models in prediction accuracy, statistical inference, such as estimating the effects of covariates and quantifying the prediction uncertainty, is not trivial due to the highly complicated model structure and overparameterization. To address this challenge, we propose a new Bayesian approach by embedding CNNs within the generalized linear models (GLMs) framework. We use extracted nodes from the last hidden layer of CNN with Monte Carlo (MC) dropout as informative covariates in GLM. This improves accuracy in prediction and regression coefficient inference, allowing for the interpretation of coefficients and uncertainty quantification. By fitting ensemble GLMs across multiple realizations from MC dropout, we can account for uncertainties in extracting the features. We apply our methods to biological and epidemiological problems, which have both high-dimensional correlated inputs and vector covariates. Specifically, we consider malaria incidence data, brain tumor image data, and fMRI data. By extracting information from correlated inputs, the proposed method can provide an interpretable Bayesian analysis. The algorithm can be broadly applicable to image regressions or correlated data analysis by enabling accurate Bayesian inference quickly.

SiM Clinical Trials Bayesian Methods Survival Analysis · 发表于 2024-06-17

Improved mortality analysis in early‐phase dose‐ranging clinical trials for emergency medical diseases using Bayesian time‐to‐event models with active comparators

Xiaosong Shi, Jo A. Wick, Renee' L. Martin, Jonathan Beall, Robert Silbergleit, Gaylan L. Rockswold, William G. Barsan, Frederick K. Korley, Sarah Rockswold, Byron J. Gajewski

摘要

Emergency medical diseases (EMDs) are the leading cause of death worldwide. A time‐to‐death analysis is needed to accurately identify the risks and describe the pattern of an EMD because the mortality rate can peak early and then decline. Dose‐ranging Phase II clinical trials are essential for developing new therapies for EMDs. However, most dose‐finding trials do not analyze mortality as a time‐to‐event endpoint. We propose three Bayesian dose‐response time‐to‐event models for a secondary mortality analysis of a clinical trial: a two‐group (active treatment vs control) model, a three‐parameter sigmoid EMAX model, and a hierarchical EMAX model. The study also incorporates one specific active treatment as an active comparator in constructing three new models. We evaluated the performance of these six models and a very popular independent model using simulated data motivated by a randomized Phase II clinical trial focused on identifying the most effective hyperbaric oxygen dose to achieve favorable functional outcomes in patients with severe traumatic brain injury. The results show that the three‐group, EMAX, and EMAX model with an active comparator produce the smallest averaged mean squared errors and smallest mean absolute biases. We provide a new approach for time‐to‐event analysis in early‐phase dose‐ranging clinical trials for EMDs. The EMAX model with an active comparator can provide valuable insights into the mortality analysis of new EMDs or other conditions that have changing risks over time. The restricted mean survival time, a function of the model's hazards, is recommended for displaying treatment effects for EMD research.

Biometrics Causal Inference Nonparametric & Semiparametric · 发表于 2024-06-17

Doubly robust estimation and sensitivity analysis for marginal structural quantile models

Chao Cheng, Liangyuan Hu, Fan Li

摘要

The marginal structure quantile model (MSQM) provides a unique lens to understand the causal effect of a time-varying treatment on the full distribution of potential outcomes. Under the semiparametric framework, we derive the efficiency influence function for the MSQM, from which a new doubly robust estimator is proposed for point estimation and inference. We show that the doubly robust estimator is consistent if either of the models associated with treatment assignment or the potential outcome distributions is correctly specified, and is semiparametric efficient if both models are correct. To implement the doubly robust MSQM estimator, we propose to solve a smoothed estimating equation to facilitate efficient computation of the point and variance estimates. In addition, we develop a confounding function approach to investigate the sensitivity of several MSQM estimators when the sequential ignorability assumption is violated. Extensive simulations are conducted to examine the finite-sample performance characteristics of the proposed methods. We apply the proposed methods to the Yale New Haven Health System Electronic Health Record data to study the effect of antihypertensive medications to patients with severe hypertension and assess the robustness of the findings to unmeasured baseline and time-varying confounding.

JRSSB Causal Inference Bayesian Methods Machine Learning · 发表于 2024-06-17

Authors’ reply to the Discussion of ‘Parameterizing and simulating from causal models’

Robin J Evans, Vanessa Didelez

摘要

We thank all the discussants for their interesting and thoughtful contributions. Our responses below are grouped by the themes raised. We thank Dr Seaman for his fascinating response. As to the approach to simulation he describes, which we informally call ‘the inversion method’, we had indeed discovered it thanks to an intervention by Evans’ DPhil student Xi Lin. Freed from rejection sampling, this makes the method much more scalable in terms of the dimensions of the variables, meaning that longitudinal models become much more feasible without the need to resort to any sort of approximation. His work with Prof. Keogh using the risk function is an example of this. We feel that the additional flexibility that is afforded by a full vine (or other) parameterization is something that could be useful in certain contexts, such as when we have some understanding of the mechanisms that relate earlier time-varying confounders to the outcome. Prof. Silva’s idea to combine a frugal parameterization for a marginal structural model and a structural causal model is very interesting. Though Richardson and Spirtes (2002) do indeed use linear structural equation models, we suspect that the intriguing idea of using potential outcomes in the manner suggested in Prof. Silva’s comment would not have occurred to them! We note that the parameterization he uses is the same as our own and, aesthetically speaking, we prefer our original formulation; however, the point that structural equation models are slightly more flexible than our paper suggests is well made. Prof. Guo’s density ratio formulation of the frugal model is an attractive idea, and suggests some generalizations of the method. We also note that the partially marginal model in his Example 2 has appeared in our presentations on this topic in various forms over the years. One idea was that, when modelling the health effects (Y) of alcohol consumption (X), Z1 might represent the genotype of an individual, and that Z2 might be some environmental factor such as socio-economic status that we wish to control for. The key role of the likelihood ratio in marginal structural models for causal inference for time-to-event settings has also been pointed out by Røysland et al. (2022), where a similar argument is used to show the property of eliminability, which is related to marginalization. Prof. Dawid’s question about dynamic treatment models mirrors our own line of thinking with almost frightening precision! We intend for this to be the subject of our next manuscript. It may appear straightforward, using the inversion method mentioned above, to simulate from deterministic regimes. A difficulty will lie in resolving any possible incompatibility between a given marginal model and the implicit constraints of a dynamic regime—the two may not be arbitrarily combinable. Moreover, it could be a further challenge to obtain a smooth parameterization that contrasts two such regimes. As we note in Section 8 of the paper, similar quantities seem to lead inevitably to a description that contains singularities. Dr Seaman and Prof. Martinussen both ponder whether using maximum likelihood estimation (MLE) with a parametric nuisance function can really be a reasonable comparison for a semiparameteric approach in which these functions are left largely unspecified. We would respond that it still seems reasonable to us to use the MLE as an ‘optimal’ baseline, even if no method that makes reasonable assumptions can ever attain that variance. However, it would be an important insight if it turned out that in some cases a particular semiparametric method performs almost as well as the MLE. We certainly agree that the likelihood is useful for a Bayesian analysis—see our response to Steiner and Steel’s contribution below. Prof. Martinussen also mentions that model-free approaches may be preferable to fitting a fully parametric model. We agree with this, to the extent that the model may be misspecified, and say as much just above Remark 5.2. However, we wish to emphasize that the understanding of the model that is gained by being able to specify the separate pieces is extremely useful. Prof. Battey also discusses model misspecification, pointing to a paper that will shortly be released as a preprint. This paper demonstrates that conditions under which defining a model incorrectly will not affect inference may be weaker than was previously known. Relying solely on correct specification of the outcome model may be rather a strong assumption in practice, however, and this fact has motivated much of the inverse propensity weighting and doubly robust literature in causal inference. Dispensing with the propensity score carries some dangers regarding robustness as is well-known from the recent double/debiased machine learning literature. Prof. Bhansali also suggests that the approach we take may be too general for data analysis. Indeed, we agree that having a full likelihood function is not necessary for inference, but we note that there are some advantages to this approach. As mentioned above, they can serve as ‘optimal’ comparator, e.g. for moment-based methods. Indeed, generative models are increasingly used in machine learning-based inference methods. Steiner and Steel’s contribution illustrates one of our main hopes for further work—that one can much more easily allow Bayesian inference in a causal model. This is both because we provide a likelihood, which is necessary for fully Bayesian methods to work, and also because the variation independence properties make sensible prior specification much easier. We are grateful to them for the examples of this that they provide, and the lack of sensitivity to the prior that they illustrate is reassuring. We thank Profs Richardson and Robins for their discussion. On the question of mediation, we are happy to elaborate on the comments in Section 8. In the case where all random variables are binary, we can use a (marginal) log-linear parameterization to describe the model. These are just ordinary log-linear parameters, but may be defined in a particular margin of the entire distribution; see the recent overview by Rudas and Bergsma (2023) for a fuller explanation. In this case, in order to have a smooth parameterization of the whole model, it is necessary to specify an effect parameter precisely once for each subset of the variables (see Theorem 3 of Bergsma & Rudas, 2002). We use the notation λLM to denote the log-linear parameter of an effect corresponding to the variables XL within the margin XM⁠, where necessarily L⊆M⁠. For example, if L={1,2} and M={1,2}⁠, then λLM would be an (unconditional) odds ratio between X1 and X2⁠; if L={1} then it would parameterize the ‘main effect’ of X1 within the margin X1,X2⁠. We will consider the full model, including the two versions of treatment, A and A′⁠, as separate variables; this is illustrated in Figure 1. For mediation the functional of interest is the distribution of Y∣A,A′⁠, where observationally P(A=A′)=1⁠, but not interventionally. This can be parameterized by the log-linear quantities λYAA′Y,λAYAA′Y,λA′YAA′Y,λAA′YAA′Y (here we abuse notation slightly). However, the graph in Figure 1 imposes the conditional independence constraint that Y⊥⊥A′∣A,M⁠, which requires that both λA′YAA′MY=λAA′YAA′MY=0⁠. Given the conflicting margins within which the parameter and the constraint must be defined, it seems not to be possible to model the A′-Y interaction and impose the necessary conditional independence constraint, whilst maintaining a smooth parameterization of the entire model. The results of Evans (2015) can be used to show that one can use the parameters λYAA′Y,λAYAA′Y to control the ‘direct’ effect in a mediation model, but we cannot have full control over the ‘indirect’ effect in the same way. Graph representing a mediation model. Richardson and Robins’ proposal does indeed yield variation independence, but it is not locally smooth at points in the model where A′⊥⊥M⁠. *(abstract via OpenAlex)*

SiM Latent Variables & Mixtures Longitudinal Data High-Dimensional · 发表于 2024-06-17

A sparse factor model for clustering high‐dimensional longitudinal data

Zihang Lu, Noirrit Kiran Chandra

摘要

Recent advances in engineering technologies have enabled the collection of a large number of longitudinal features. This wealth of information presents unique opportunities for researchers to investigate the complex nature of diseases and uncover underlying disease mechanisms. However, analyzing such kind of data can be difficult due to its high dimensionality, heterogeneity and computational challenges. In this article, we propose a Bayesian nonparametric mixture model for clustering high‐dimensional mixed‐type (eg, continuous, discrete and categorical) longitudinal features. We employ a sparse factor model on the joint distribution of random effects and the key idea is to induce clustering at the latent factor level instead of the original data to escape the curse of dimensionality. The number of clusters is estimated through a Dirichlet process prior. An efficient Gibbs sampler is developed to estimate the posterior distribution of the model parameters. Analysis of real and simulated data is presented and discussed. Our study demonstrates that the proposed model serves as a useful analytical tool for clustering high‐dimensional longitudinal data.

SiM Longitudinal Data Nonparametric & Semiparametric · 发表于 2024-06-16

Generalized single index modeling of longitudinal data with multiple binary responses

Zibo Tian, Peihua Qiu

摘要

In health and clinical research, medical indices (eg, BMI) are commonly used for monitoring and/or predicting health outcomes of interest. While single‐index modeling can be used to construct such indices, methods to use single‐index models for analyzing longitudinal data with multiple correlated binary responses are underdeveloped, although there are abundant applications with such data (eg, prediction of multiple medical conditions based on longitudinally observed disease risk factors). This article aims to fill the gap by proposing a generalized single‐index model that can incorporate multiple single indices and mixed effects for describing observed longitudinal data of multiple binary responses. Compared to the existing methods focusing on constructing marginal models for each response, the proposed method can make use of the correlation information in the observed data about different responses when estimating different single indices for predicting response variables. Estimation of the proposed model is achieved by using a local linear kernel smoothing procedure, together with methods designed specifically for estimating single‐index models and traditional methods for estimating generalized linear mixed models. Numerical studies show that the proposed method is effective in various cases considered. It is also demonstrated using a dataset from the English Longitudinal Study of Aging project.

SiM Survival Analysis Causal Inference · 发表于 2024-06-16

Comparing two hazard curves when there is a treatment time‐lag effect

Xiaoxi Zhang, Somnath Datta, Peihua Qiu

摘要

In cancer and other medical studies, time‐to‐event (eg, death) data are common. One major task to analyze time‐to‐event (or survival) data is usually to compare two medical interventions (eg, a treatment and a control) regarding their effect on patients' hazard to have the event in concern. In such cases, we need to compare two hazard curves of the two related patient groups. In practice, a medical treatment often has a time‐lag effect, that is, the treatment effect can only be observed after a time period since the treatment is applied. In such cases, the two hazard curves would be similar in an initial time period, and the traditional testing procedures, such as the log‐rank test, would be ineffective in detecting the treatment effect because the similarity between the two hazard curves in the initial time period would attenuate the difference between the two hazard curves that is reflected in the related testing statistics. In this paper, we suggest a new method for comparing two hazard curves when there is a potential treatment time‐lag effect based on a weighted log‐rank test with a flexible weighting scheme. The new method is shown to be more effective than some representative existing methods in various cases when a treatment time‐lag effect is present.

SiM Causal Inference Clinical Trials Bayesian Methods · 发表于 2024-06-16

Assurance methods for designing a clinical trial with a delayed treatment effect

James A. Salsbury, Jeremy E. Oakley, Steven A. Julious, Lisa V. Hampson

摘要

An assurance calculation is a Bayesian alternative to a power calculation. One may be performed to aid the planning of a clinical trial, specifically setting the sample size or to support decisions about whether or not to perform a study. Immuno‐oncology is a rapidly evolving area in the development of anticancer drugs. A common phenomenon that arises in trials of such drugs is one of delayed treatment effects, that is, there is a delay in the separation of the survival curves. To calculate assurance for a trial in which a delayed treatment effect is likely to be present, uncertainty about key parameters needs to be considered. If uncertainty is not considered, the number of patients recruited may not be enough to ensure we have adequate statistical power to detect a clinically relevant treatment effect and the risk of an unsuccessful trial is increased. We present a new elicitation technique for when a delayed treatment effect is likely and show how to compute assurance using these elicited prior distributions. We provide an example to illustrate how this can be used in practice and develop open‐source software to implement our methods. Our methodology has the potential to improve the success rate and efficiency of Phase III trials in immuno‐oncology and for other treatments where a delayed treatment effect is expected to occur.

SiM Clinical Trials · 发表于 2024-06-16

A multi‐arm multi‐stage platform design that allows preplanned addition of arms while still controlling the family‐wise error

Peter Greenstreet, Thomas Jaki, Alun Bedding, Chris Harbron, Pavel Mozgunov

摘要

There is growing interest in platform trials that allow for adding of new treatment arms as the trial progresses as well as being able to stop treatments part way through the trial for either lack of benefit/futility or for superiority. In some situations, platform trials need to guarantee that error rates are controlled. This paper presents a multi‐stage design, that allows additional arms to be added in a platform trial in a preplanned fashion, while still controlling the family‐wise error rate, under the assumption of known number and timing of treatments to be added, and no time trends. A method is given to compute the sample size required to achieve a desired level of power and we show how the distribution of the sample size and the expected sample size can be found. We focus on power under the least favorable configuration which is the power of finding the treatment with a clinically relevant effect out of a set of treatments while the rest have an uninteresting treatment effect. A motivating trial is presented which focuses on two settings, with the first being a set number of stages per active treatment arm and the second being a set total number of stages, with treatments that are added later getting fewer stages. Compared to Bonferroni, the savings in the total maximum sample size are modest in a trial with three arms, <1% of the total sample size. However, the savings are more substantial in trials with more arms.

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