2026 年的期刊文章 · 第 29 页

共 1756 篇,按发表时间排,新的在前,每页 40 篇。这一页是 2026-03-21 到 2026-03-16。

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 39 40 41 42 43 44 页

Biometrika Latent Variables & Mixtures Networks & Graphs · 发表于 2026-03-21

Dynamic clustering for heterophilic stochastic block models with time-varying node memberships

K Z Lin, J Lei

摘要

Summary: We consider a time-ordered sequence of networks stemming from stochastic block models in which nodes gradually change their membership over time, and no network at any single time-point contains sufficient signal strength to recover its community structure. To estimate the time-varying community structure, we develop the kernel-debiased sum of squares method that performs spectral clustering after a debiased sum-of-squared aggregation of adjacency matrices. Our theory demonstrates, via a novel bias-variance decomposition, that our new method achieves consistent community detection in each network, even for heterophilic networks, without requiring smoothness in the time-varying dynamics of between-community connectivities. We also prove the identifiability of aligning community structures across time based on how rapidly nodes change communities, and we develop a data-adaptive bandwidth-tuning procedure for our new method. We demonstrate the utility and advantages of our proposed method through simulations and a novel analysis of time-varying dynamics in gene coordination in the human developing brain.

Biometrics Longitudinal Data Clinical Trials Nonparametric & Semiparametric · 发表于 2026-03-20

Nonparametric ANCOVA for longitudinal outcomes in a randomized clinical trial

Rex Shen, Xiaotong Jiang, Changyu Shen, Lu Tian

摘要

The analysis of covariance (ANCOVA) is a commonly used method for correcting bias and improving accuracy in estimating the average treatment effect in randomized clinical trials. In this paper, we focus on using ANCOVA for longitudinal outcomes, where mixed effects regression is the standard approach. The effectiveness of ANCOVA depends on the regression model specification, including how the baseline covariates were used. Unlike traditional methods, we do not assume that the mixed effects model is correctly specified, making our approach nonparametric in nature. We investigate the optimal ANCOVA approach for longitudinal outcomes and show that appropriate covariate adjustment can greatly improve the precision of treatment effect estimates. Unfortunately, determining the optimal ANCOVA adjustment is challenging because it relies on the relationship between longitudinal outcomes and baseline covariates, which is typically unknown. We propose to use cross fitting procedure to estimate the conditional expectation of longitudinal outcomes given baseline covariates to guide the specification of ANCOVA. We provide theoretical derivations and empirical evidence from numerical studies to demonstrate the superiority of our proposed nonparametric ANCOVA method over traditional ANCOVA approaches. Our approach is robust, flexible, and can be easily implemented in practice to improve the accuracy and reliability of treatment effect estimates in clinical trials.

JASA Other · 发表于 2026-03-20

Limiting laws and consistent estimation criteria for fixed and diverging number of spiked eigenvalues

Jianwei Hu, Jingfei Zhang, Jianhua Guo, Ji Zhu

摘要

In this paper, we study limiting laws and consistent estimation criteria for the extreme eigenvalues in a spiked covariance model of dimension p with the number of spikes k. Allowing both p and k to diverge, we derive limiting distributions of the spiked sample eigenvalues using random matrix theory techniques. Notably, our results are established under a general spiked covariance model, where the bulk eigenvalues are allowed to differ, and the spiked eigenvalues need not be uniformly upper bounded or tending to infinity, as have been assumed in the existing literature. Based on the above derived results, we formulate general estimation criteria that can consistently estimate k, while k can be fixed or grow at an order of k=o(n1/3). Our results are established under both Gaussian distributions and general distributions with finite fourth moments, with different growth rate conditions on k. The effectiveness of the proposed estimation criteria is illustrated through simulation studies and applications to three real-world data sets. *(abstract via OpenAlex)*

JRSSB High-Dimensional Networks & Graphs Latent Variables & Mixtures · 发表于 2026-03-20

Authors’ reply to discussion of ‘Statistical exploration of the Manifold Hypothesis’ by Whiteley et al

Nick Whiteley, Annie Gray, Patrick Rubin-Delanchy

摘要

We are very grateful for the numerous, insightful, thought-provoking, and helpful contributions to this discussion. Many of the individual contributed discussions address several topics. Our response is structured to reflect themes which have emerged. Trosset highlights the question of whether existing statistical models can be formulated as special cases of the latent metric space (LMS) model. If so, perhaps the LMS model can shed new light on manifolds those models generate. Trosset refers to earlier work (Rubin-Delanchy, 2020), for example, in which starting from the well-established latent position network model one can show that the emerging ‘manifold’ is a set of low Hausdorff dimension, but nothing is said beyond that about homeomorphism or isometry to the latent space. Perhaps, the most compelling response we can give to Trosset is that the LMS model generalizes the Gaussian process latent variable model (GPLVM) of Lawrence (2003, 2005) in which the latent space is Rd⁠, the kernel must be chosen by the user, and a radial basis kernel is often selected in practice. For such a kernel, observe that the isometry property of Proposition 3 holds. Thus, the GPLVM implicitly generates a manifold which we can describe in the LMS framework; and which turns out to be a simple and standard conception of a nonlinear manifold since its intrinsic geometry is that of a convex region of Euclidean space. With reference to the seminal work by Tenenbaum et al. (2000), Trosset himself has noted the prominence of such manifolds in the manifold learning literature (Trosset & Buyukbas, 2020), where they are often referred to as ‘Swiss rolls’. Reinert and Zhu & Guo draw connections to the latent position network model (Hoff et al., 2002). As we mentioned at the discussion meeting, the LMS model was born in network theory—estimation theory developed for the random dot product graph, and the discovery of manifold structure (of various descriptions) in graph embeddings under the latent position network model. Reinert asks if there are now opportunities to say new things about the latent position network model. Returning to this world, we wonder if there is a very good reason to assume latent space is Euclidean (Hoff et al., 2002). Metric spaces are extremely simple and expressive, and seem perhaps the more natural structure to capture notions of clustering, hierarchy, and other topology. One should note a somewhat imposing barrier to progress on the network modelling front: kernels in latent position network models need not and often are not positive-definite, e.g. in heterophilic settings (Rubin-Delanchy et al., 2022). This seems to impact analysis in quite a fundamental way, possibly requiring pseudo-Euclidean geometry. We are very grateful for the connections made in Maullin’s discussion between our model and Adler and Taylor (2007)’s work on random fields and geometry, of which we were not previously aware. We are yet to appreciate the full extent of these connections, but unexpectedly to us, Maullin’s discussion highlights how the feature map ϕ and derivatives thereof are related to deviation probabilities associated with statistics of interest in hypothesis testing for random fields. It seems that leveraging this connection to estimate such probabilities from data in our model setting would require some kind of extension of our principal component analysis (PCA) consistency result to first and second derivatives of ϕ. Using the analogy of criminal justice separating into the search for evidence (e.g. by a detective) vs. the evaluation of evidence (e.g. by a jury), Tukey (1977) wanted statisticians to understand that exploratory and confirmatory data analysis are very different endeavours. When we think of a statistical model, we are perhaps culturally inclined to consider it in a confirmatory context. In our work, we have tried to put emphasis on the perspective of exploration, since we think this reflects the common practice and usefulness of dimension reduction in searching for structure in data, motivated by the manifold hypothesis. Several discussants comment on moving beyond the structure of our model and the assumptions we consider. Imaizumi highlights interest in heterogeneity and irregularity such as nonuniform intrinsic dimension. On a related theme, Aamari highlights empirical evidence of a union-of-manifolds in real-world data. In fact, there is nothing about the general LMS model that requires there to be a fixed latent dimension. We note that replacing the latent metric space in our model with a disjoint union of metric spaces could be one way to accommodate heterogeneity in geometry (e.g. with points in distinct constituent spaces defined to be at infinite distance from one another), naturally giving rise to a ‘mixture’ model with potentially class-dependent geometry as suggested by Aamari. Simon et al. point out situations in which the distinguishability condition A2 (which amounts to saying ϕ is injective) is not realistic. From a mathematical point of view, we bear in mind that, if ϕ is not injective, M is nevertheless homeomorphic to the quotient space associated with the equivalence relation z∼z′ defined by ϕ(z)=ϕ(z′)⁠. In that sense, M always conveys whatever topological structure of Z can be identified through ϕ. We also note that our consistency theory for PCA does not require assumption A2. Simon et al. also comment on the strength of the assumption that the kernel is of radial form in Proposition 3. In our view, this assumption is interesting because of (i) the strength of conclusions of Proposition 3 regarding isometry which can be deduced from it and (ii) the fact that radial basis kernels, e.g. f(z,z′)=exp(−‖z−z′‖2/τ) for τ>0⁠, which satisfy this assumption, are extremely well-known in Statistics and Machine Learning. Tian et al. also question some of our assumptions. In this context, we would like to clarify that we regard the rank r of the kernel not as something we are free to choose, but rather as what is presented to us by nature in generating data. In broad terms, these kinds of assumptions are perhaps better understood as hypotheses—a possibility (e.g. stationarity) which would have interesting implications (e.g. isometry) if true. We think it is not inherently a bad thing if a hypothesis is strong, as long we are fair and open-minded in our investigations. We welcome Tengyao Wang’s question about what geometric conclusions might be drawn under weaker conditions, we think there is much to be done in this direction. In Section 3, we sketch some mathematical details which we think help expose more of the connection between f and the geometry of M⁠. Kundu and Schmidt-Hieber highlight interest in alternatives to the additive noise model we consider, including the possibility of noise in the latent domain. We note that our general formulation with Z being a metric space allows for many possible interpretations of the latent variables, e.g. that each latent Zi is a collection of ‘signal’ and ‘noise’ variables. This would not materially impact our PCA theory concerning the recovery of ϕ(Z1),…,ϕ(Zn) but would impact how ϕ(Z1),…,ϕ(Zn) should be interpreted in terms of signal vs. noise. Yinqiu He has conducted some experiments which seem to reveal some intriguing possibilities. If, as in their example, the latent metric space admits a natural decomposition into separate variables, it might be incredibly useful to ask how to disentangle these latent variables from the manifold. Sanna Passino and Heard suggest joint dimension reduction of a sequence of data matrices Y1,…,YT⁠, where the underlying latent variables vary over time. We think this could be a very rich framework in which to study time series data. They ask about the role of independence across latent variables Z1,…,Zn in our setup: this independence is used in the proof of Lemma S.8, to establish that the matrix n−1Φ⊤Φ≡n−1∑i=1nϕ(Zi)ϕ(Zi)⊤ is concentrated about its expectation E[n−1Φ⊤Φ]=Λf⁠, which is the diagonal matrix of eigenvalues of the kernel f (the latter equality holds by L2(μ)-orthonormality of the eigenfunctions ukf and the assumption Zi∼μ⁠). In turn, this concentration is used in the proof of Proposition S.3 to relate the eigenvalues of n−1Φ⊤Φ to those of f. We think it is reasonable to expect qualitatively similar concentration behaviour of n−1Φ⊤Φ under some form of weak dependence rather than independence of Z1,…,Zn⁠. Taking this thought further, it could be that n−1Φ⊤Φ≈Λf in situations where Z1,…,Zn are not random draws from μ, but rather deterministic and such that integration (of suitable classes of functions) with respect to the measure n−1∑iδZi approximates integration with respect to μ. For example, when Z is a subset of Euclidean space, μ is the uniform distribution on that set and Z1,…,Zn is a low-discrepancy sequence ‘uniformly spaced’ on Z⁠. Kundu and Schmidt-Hieber also highlight the possibility of considering streaming data and online algorithms, we think there is much to explore there. Gelman mentions the ‘bet on sparsity principle’ (Hastie et al., 2009) and asks about the cognitive dissonance of taking this modelling stance in practice, when we know the truth is not sparse. We think that the sparsity principle discussed could be interpreted quite narrowly to refer to the efficacy of an L1 penalty on linear regression coefficients. Perhaps a broader interpretation of the sparsity principle could invoke less cognitive dissonance. Our paper puts forward two senses in which data might be considered ‘sparse’: they may be concentrated near a low-dimensional linear subspace (i.e. the rank r of the kernel could be low), and they may be concentrated near a low-dimensional manifold within that subspace (i.e. d could be low, in the case Z⊂Rd⁠). If the first dimension (r) is low, this suggests a great opportunity to apply PCA (Theorem 1). If the second dimension (d) is low, there is great hope for, e.g. prediction based on Yi as a covariate, if we use a method which can exploit this low dimensionality. Of course, in reality neither form of sparsity is likely to hold exactly, and what might be treated in our framework as ‘noise’ could contain layers of further structure which we seek could uncover through further model-building iterations. Bordino and Klopp highlight the potential shortcomings of applying spherical projection to the PCA embedding under anisotropic nuisance variation and propose an additional elliptic transformation step which has the potential to correct for it. Examining their arguments for the efficacy of this transformation, we understand the issue they highlight is that, in general, E[ϕ(Z)ϕ(Z)⊤]≠cIr⁠. However, we note that because the eigenfunctions defining ϕ are orthonormal in L2(μ) and Z∼μ⁠, under no further assumptions than we have made in the paper we always have E[ϕ(Z)ϕ(Z)⊤]=Λf⁠, where Λf is the diagonal matrix of eigenvalues of f. This leads us to question if some alternative to the standard PCA-based embedding which re-weights dimensions using empirical eigenvalues may be useful in achieving the robustness Bordino and Klopp ask about. We recognize the significance of the question they raise about extending Theorem 1 to address the ‘whitened’ PCA embedding. The intricacies of the proof techniques prevent us from giving a succinct and categorical response to this question here, but we think this is a very interesting direction for future research. Several correspondents (e.g. Kadhem, Wen, and Dowe, Tian et al.) asked about the stability of sample-splitting for dimension selection, with one random split potentially recommending a different choice of dimension to another. It seems sensible to explore different splits, within computational feasibility, and this opens some interesting questions about how to interpret the variability across splits, and how to combine potentially different recommendations. As it happens, we were a little surprised at how low this variability actually seems to be in the contributed discussions. For example, in Tian et al. the recommendation is always r^=5,6⁠, or 7; moreover, the profiles are all qualitatively similar. Several discussants raise questions about connections between our model and conventional notions in the mathematical theory of differentiable manifolds. In particular, Aamari highlights how, in the case where X:Z→Rp is a deterministic mapping, low dimensionality is captured by the conventional notion of intrinsic dimension of the image of X: the rank of the Jacobian Jz=∇zX(z)∇zX(z)⊤⁠, where X(⋅)=[X1(⋅),…,Xp(⋅)]⁠, and contrasts this with low-dimensionality in our setup being captured in low rank of the kernel f. However, this is not the only way low dimensionality may appear in our setup. Indeed, suppose that Z is some suitable subset of Rd and that f and ϕ are suitably differentiable and well-behaved so that the following formal manipulation is justified: where z=[z1,…,zd]⁠, z=[z1′,…,zd′] and the first equality holds by Mercer’s theorem (we note that differentiability of kernels and feature maps is studied rigorously by, e.g. Steinwart and Christmann (2008, Lemma 4.34)). Comparing (1) with Jz⁠, we see the intrinsic dimension of M={ϕ(z);z∈Z} at a point ϕ(z) is the rank of the d×d matrix Hz with elements: In general, the rank of Hz is not the same as the rank of f, indeed the latter tells us the lowest dimension of any linear subspace of ℓ2 containing M⁠, which may be infinite, whereas the rank of Hz is clearly at most d. In the case f(z,z′)=g(‖z−z′‖2) addressed in Proposition 3 in the paper, by direct calculation Hz=−2g′(0)Id⁠. The identities in (1) and Hz may a to other notions from the theory of differentiable manifolds. highlights work on various notions of from data and a question about whether such notions can be into our setup. this question more in terms of of our how does the kernel f impact notions of of formal for one such notion of Z is a in f is defined and on an set containing and ϕ is the of of M⁠, where the across is over all in M⁠. As by, e.g. Aamari et al. et al. is an on the of M et al., Proposition ϕ is on its image Proposition 1 in the and has a we can a in M as a with a in such that and where is the of a is Euclidean Hz is as in and the equality in Mercer’s From the of can be with if is where is the matrix of second derivatives of at Using in the same way as in we can see that, in general, on various derivatives of f possibly to to the case f(z,z′)=g(‖z−z′‖2) with in Proposition we have and we is a in Z with is further, if Z is being and some In the of these formal some of the of of M is at in We are very grateful to Aamari for connections to the work of Adler et al. about of embeddings of the form where Gaussian In this the kernel f, and ϕ and M⁠, not on and the of tells us as of of using the framework of Adler et al. might establish some between their and the geometry of M⁠. about what geometry is by alternative kernels is a potentially as by In practice, one often that embeddings are and some by e.g. and data in different applying PCA and other geometric the for also be The manifold potentially could be homeomorphic or to but M and are the question of which to in practice. to the of for embedding could potentially some et al., et al., & and highlight some geometric of manifolds of and an interpretation of the manifold hypothesis in terms of low rank of the associated This perspective on the manifold hypothesis seems somewhat different to that in our We not know if there are any direct connections between our setup and geometry, but we the possibility The statistical theory of embeddings and the has a of for We welcome of how this has developed and to at many in this we are to see the work of and which may to the assumption which we on in our theory for and which is in the discussions of Aamari and In turn, more of PCA consistency may of testing about latent structure using as by and discussions. Tengyao highlights the potential to address statistical questions such as of kernel geometry and We think that in such questions it may be to consider some special cases of our general model such as the GPLVM of Lawrence (2003, on which mathematical from statistical theory could be to We are very grateful indeed for of our and the potential for which We are about whether assumptions of dimension of noise might be a vs. the of noise with but we are yet to behaviour of random and matrices is a in our about PCA and the literature on consistency of embedding which and highlight how a random with and Gaussian can be to many low-dimensional with In this sense, it could be that a of noise behaviour which seems to within of the manifold as in the from in the of the In this we think and discussion highlights that there is for mathematical and further of what the manifold hypothesis is or could Indeed, in the in our paper, and many we have in the a which for the efficacy of statistical learning is not only that data are in some but that the or geometry of that low-dimensional structure conveys some useful about the underlying Our setup of a latent metric space through the our homeomorphic is one way to such We are open-minded about the possibility of other which may have their or The theory of statistics conclusions from assumptions that data are or In our work, we a weaker assumption, for of It seems reasonable to expect that our for PCA could be e.g. discussion highlights concentration behaviour of between noise with In et al. and proof of the has to between possibly random with This allows similar concentration to that by to be for between data under our model, with the additional that the independence assumption is to on dimension of the matrix of the ‘signal’ component of the model This may also be a step of weak dependence in the discussion of and ask about connections between our work and the One point of in the special case of the as discussed in the In this setting the of the matrix in Proposition 1 are random variables, so that the structure in Proposition 1 seems to the of random for of dimension the of additive noise in our Of the perspective on a random projection is between our work and the practice of random In the latter the random projection is an whereas in our work it implicitly from our assumptions and is considered of and comment on the connection between Proposition 1 and the We think that may be with the latter in the setting of a say where the basis from the kernel of that We think that some in our setting only in that we are a common basis from the kernel f. our theory for PCA of the points there is potential to beyond this to explore statistical theory for geometric of or of Z1,…,Zn to whatever on from ϕ. Indeed, the of PCA by some kind of nonlinear dimension reduction such as embedding or uniform manifold and projection for of data or is very in practice, and in principle could be formulated as of Z1,…,Zn or Z⁠. simple point could be to consider the case in which Z is an in and to assume isometry as in Proposition 3. related to highlights the work of et al. which a analysis of the recovery of latent manifold structure in the form of associated with of a could be thought of as being Z1,…,Zn e.g. In relation to this asks about under which latent Euclidean distance and distance in M each other to suitable This seems related to ϕ being as a between Z and M with and indeed when and that property holds when f is and is for all et al., Theorem 1). Kundu and Schmidt-Hieber point out that if it were possible to from the data which of the Z1,…,Zn were to this could be used to apply kernel techniques as in We note that et al. work this in a and might the of the as by Kundu and The of additive noise in our model an to which are from the of the Indeed, in work et al., and under and other we uniform on the data dot any assumption of low for whereas for with The of in may us from which Zi and are in Euclidean if we to vary across the latent domain. We are by the potential to our work to the statistical theory of learning discussion highlights et al. work on the implications for network of the and of a on which the data distribution is to be We may this as yet perspective on how the manifold hypothesis can be formulated and interpreted in a context. In similar questions could be under the assumption that some our model is the network and theory by Imaizumi could be in a similar Several draw potential connections to the emerging of and in the interpretation of the geometry of in In this context, the of a network refers to the from an it has through several Several have geometry in these which are perhaps to what might be by our model and natural This for example, of et al., in which we would have to in data under assumptions on the LMS random fields et al., and isometry in the of such as or time et al., which we may expect to see in data under assumptions. As et al. it seems possible and if so quite that could also as general statistical in data. It is not very to us what to of these have to on the geometry of being a result of whereas our could suggest this geometry is in the data. *(abstract via OpenAlex)*

SiM Clinical Trials · 发表于 2026-03-20

Analysis of Stepped‐Wedge Cluster Randomized Trials: A Tutorial Using Marginal Models

Elizabeth L. Turner, John S. Preisser, Ying Zhang, Xueqi Wang, Mark Toles, Samuel Cykert, Fan Li, Paul J. Rathouz

摘要

Stepped‐wedge cluster randomized trials (SW‐CRTs) are one‐way crossover trials that randomize clusters (i.e., groups) of individuals to the time point (period) at which an intervention is introduced into the cluster. In these designs, the intervention under evaluation is introduced into all of the clusters by the end of the study in a series of “steps.” Analysis of SW‐CRTs using marginal models provides a population‐averaged interpretation of the estimated intervention effect and flexible specification of the within‐cluster, marginal pairwise association structure; the latter has practical application in reporting intraclass (i.e., pairwise) correlations and calculating power for CRTs. Despite these features, use of marginal modeling of SW‐CRTs has been mostly limited to applications with working independence and simple exchangeable correlation structures that are suboptimal for multi‐period CRTs when correlation among responses decays over time. However, there have been many methodological developments in marginal modeling of SW‐CRTs over the past fifteen years, particularly on (i) multi‐parameter, within‐cluster correlation structures; (ii) paired generalized estimating equations (GEE) for simultaneous estimation of mean and correlation parameters with standard errors; and, when the number of clusters is small, (iii) corrections to reduce the bias of variance estimators, and that of correlation estimates using matrix‐adjusted estimating equations (MAEE). The goal of the current tutorial is to survey these newer developments and to provide case studies to enable applied researchers to implement GEE/MAEE for marginal model analysis of SW‐CRTs, with application to both cohorts and designs with repeated cross‐sectional samples. The methods are also applicable to multi‐period, parallel‐arm and cluster‐crossover CRTs.

JRSSB Causal Inference Nonparametric & Semiparametric · 发表于 2026-03-19

Proximal causal inference for conditional separable effects

Chan Park, Mats J Stensrud, Eric J Tchetgen Tchetgen

摘要

Scientists regularly pose questions about treatment effects on outcomes conditional on a posttreatment event. However, causal inference in such settings requires care, even in perfectly executed randomized experiments. Recently, the conditional separable effect (CSE) was proposed as an interventionist estimand that corresponds to scientifically meaningful questions in these settings. However, existing results for the CSE require no unmeasured confounding between the outcome and posttreatment event, an assumption frequently violated in practice. In this work, we address this concern by developing new identification and estimation results for the CSE that allow for unmeasured confounding. We establish nonparametric identification of the CSE in observational and experimental settings with time-varying confounders, provided that certain proxy variables for hidden common causes of the posttreatment event and outcome are available. For inference, we characterize an influence function for the CSE under a semiparametric model where nuisance functions are a priori unrestricted. Using modern machine learning methods, we construct nonparametric nuisance function estimators and establish convergence rates that improve upon existing results. Moreover, we develop a consistent, asymptotically linear, and locally semiparametric efficient estimator of the CSE. We illustrate our framework with simulation studies and a real-world cancer therapy trial.

SiM Other · 发表于 2026-03-19

Pairwise Comparisons of k Binomial Responses

Dennis D. Boos, James Schmidt

摘要

Independent binomially distributed data arise in many contexts, such as clinical trials, quality control monitoring, and stratified sampling. Moreover, the scope is much larger because multinomially distributed data from a 2 by contingency table can be viewed as conditionally independent binomial random variables. A standard approach is to use Fisher's conditional test to test equality of the underlying success probabilities. However, researchers often want to know where the important pairwise differences are. Thus, the closed method of pairwise comparisons is here combined with unconditional exact tests for 2 by 2 tables and Fisher's conditional test for larger tables to get values exhibiting strong control of the Family‐Wise Error Rate and excellent power properties. In clinical trials, studies are often multicenter, but the results here pertain only to single‐site studies.

SiM Genetics & Omics Networks & Graphs · 发表于 2026-03-19

Multivariate Regression With Dependence Structures: Evaluating Associations Between Plasma Metabolomics and Alcohol Intake in Older Adults

Yifan Yang, Chixiang Chen, Hwiyoung Lee, Ming Wang, Yuan Zhang, Shuo Chen

摘要

High‐dimensional omics data often exhibit complex yet organized dependencies, characterized by intelligent network properties like high modularity, small‐worldness characteristics, and scale‐free topology. However, integrating these structured interdependencies between omics variables into multivariate regression models presents challenges. The primary difficulty lies in accurately specifying and estimating dependency parameters that capture these network patterns within regression frameworks. Common covariance estimation methods may not preserve these network properties and can also be computationally intensive. To address these challenges, we propose a novel multivariate regression model that incorporates an interconnected community structure, reflecting the organized relationships among omics outcome variables. Our approach includes efficient estimation algorithms, featuring closed‐form regression estimators and likelihood‐based dependence estimators. We also establish the asymptotic properties of estimators to ensure theoretical robustness and hypothesis testing. Extensive simulations demonstrate the enhanced accuracy and sensitivity of our method, as evidenced through benchmarking against existing regression models. We applied our approach to a dataset to assess the associations between 249 metabolomic biomarkers, measured using nuclear magnetic resonance spectroscopy, and alcohol intake among 3984 participants. Results indicate that light alcohol consumption is positively associated with high‐density lipoprotein cholesterol (HDL, i.e., the good cholesterol), high‐density lipoprotein particles, and Apolipoproteins A1, indicators linked to cardiovascular health.

SiM Survival Analysis Longitudinal Data Bayesian Methods · 发表于 2026-03-19

Joint Modeling of Quality of Life and Survival Using a Bayesian Approach in a Retrospective Time Scale

Yizhou Fei, Elizabeth Juarez‐Colunga, Areej El‐Jawahri, Jean S. Kutner, Kathryn Colborn

摘要

Improving patients' quality of life (QoL) is one of the primary goals of palliative care clinical trials. However, a significant challenge in this area is the “truncation by death problem,” where QoL data cannot be observed after a patient dies, potentially introducing bias into statistical analyses. Understanding the impact of truncation by death when estimating the association between QoL and exposure or treatment is essential, especially when a relatively large proportion of subjects die during a study. To address this issue, we propose a Bayesian joint modeling framework that considers dependencies at both the individual and cluster levels while examining longitudinal QoL trajectories and survival outcomes simultaneously. This approach builds on existing joint modeling methods by incorporating cluster‐level random effects. We model QoL on a retrospective scale relative to the time of death, while linking survival via both the subject and cluster‐level random effects. The longitudinal sub‐model also allows for flexible, non‐linear QoL trajectories, which are modeled using penalized regression splines. For the survival sub‐model, we use a proportional hazards frailty model with a Weibull baseline hazard. The model is estimated using a Bayesian framework, implemented via Markov Chain Monte Carlo (MCMC) sampling. To evaluate the performance of our method, we conducted a comprehensive simulation study including scenarios with different numbers of clusters. We also show results from applying this novel methodology to data from the Reducing End of Life Symptoms with Touch (REST) study.

SiM Clinical Trials Bayesian Methods · 发表于 2026-03-19

Covariate Adjustment in Basket Trials Borrowing Information Across Subgroups

Jiyang Ren, David S. Robertson, Haiyan Zheng

摘要

Basket trials are an efficient approach to simultaneously evaluate a single therapy across multiple diseases where patients share a common molecular target. Bayesian hierarchical models (BHMs) are widely used to estimate the treatment effects while accounting for heterogeneity between patient subgroups within a basket trial. However, the use of analysis of covariance (ANCOVA) with treatment‐by‐covariate interaction terms, in this context of patient heterogeneity and small samples, has been largely unexplored, despite the widespread use of ANCOVA for improving estimation precision in traditional settings from a frequentist perspective. In this paper, we propose two covariate‐adjusted BHMs that incorporate ANCOVA into the data model to enhance the estimation precision in basket trials, wherein borrowing of information is permitted across subgroups to a certain extent. Specifically, both ANCOVA without treatment‐by‐covariate interaction terms and ANCOVA with interaction terms are explored in the analysis of basket trials. We perform a simulation study to demonstrate the advantages of covariate‐adjusted BHMs compared to unadjusted BHMs, as well as frequentist ANCOVA models. The BHMs are then retrospectively applied to the analysis of the MAJIC study, a randomized controlled basket trial involving two subtypes of blood cancer.

Biometrics Causal Inference Survival Analysis · 发表于 2026-03-19

A robust covariate-balancing method for estimating individualized treatment with censored data

Rujia Zheng, Wensheng Zhu, Xiaofan Guo

摘要

One of the most essential aspects of precision medicine is the identification of optimal individualized treatment regimen, which recommends treatment decisions to maximize a patient’s expected survival time based on their individual characteristics with censored data. Typically, the expected survival time is required to be estimated first, which is usually based on the posited weighting models (propensity score model and censoring model) or the posited outcome model. However, if any of the above models is misspecified, the estimated treatment regimen is not reliable. In this paper, we consider the contrast value function defined for survival analysis, and propose two robust covariate-balancing estimators of the contrast value function by balancing the covariates of patients through censoring probability and survival function of censoring time in the weights, respectively. Theoretical results prove that the proposed estimators are doubly robust and asymptotically normal under standard regularity conditions. A large number of simulations show the superiority of our methods over the existing methods. We further apply the proposed methods to the China Rural Hypertension Control Project to identify the individualized medication regimens for hypertensive patients by tailoring their individual characteristics to maximize the expected survival time. The results indicate that the estimated optimal treatment regimen improves patients’ survival probability over the 36-month follow-up period.

SiM Clinical Trials · 发表于 2026-03-18

What Makes an Estimand Useful? Guidance on the Choice of Intercurrent Event Strategies

Brennan C. Kahan, Fan Li, Michael O. Harhay, Suzie Cro

摘要

While the use of estimands in randomized trials is increasing, there is little guidance on which intercurrent event strategies should be used. The article by Fleming et al. seeks to address this gap. They argue that strategies such as hypothetical, principal stratum, and while‐alive generally cannot be used to reliably inform decision making, and that treatment policy (and composite for mortality) strategies should be used instead. In this Commentary we argue that there are a variety of settings where strategies such as hypothetical, principal stratum, and while‐alive can reliably inform decision‐making and are preferable to a treatment policy strategy. We provide an alternative approach for selecting intercurrent event strategies, which systematically considers the trade‐off between relevance (whether it addresses a useful question) and reliability (the ability to be estimated such that stakeholders can have confidence in the results) of each strategy in order to identify those that can be used to robustly inform decision‐making. Our overall conclusion is that there is no single intercurrent event strategy that is appropriate in all settings; all strategies can be beneficial when used in appropriate settings, but harmful when used in inappropriate settings.

SiM High-Dimensional · 发表于 2026-03-18

Sensitivity Analysis for Publication Bias in Diagnostic Meta‐Analysis of Sparsity Using the Copas t ‐Statistic Selection Function

Taojun Hu, Yi Zhou, Xiao‐Hua Zhou, Satoshi Hattori

摘要

Publication bias (PB) poses a significant threat to meta‐analysis of diagnostic studies, as studies yielding significant results are more likely to be published in scientific journals, leading to the synthesized diagnostic capacity possibly being overestimated. Sensitivity analysis provides a flexible method to address PB by assuming different proportions of unpublished studies. Most existing methods addressing PB in meta‐analysis of diagnostic studies are based on the bivariate normal model using normal approximations. However, they are unsuitable for meta‐analysis with sparse data, which is common in diagnostic studies with high sensitivities or specificities. Alternatively, the bivariate binomial model relies on the exact within‐study model and has better finite sample properties. To address PB in the bivariate binomial model, we model the selective publication process of diagnostic studies by extending the Copas t ‐statistic model and propose the likelihood conditional on published and estimation strategies. Our proposal provides an interpretable way to address PB on the summary receiver operating characteristic curve, an essential tool for synthesizing diagnostic accuracy. We show the practicability of the proposed method on several real‐world meta‐analyses of diagnostic studies and evaluate the performance by simulation studies.

SiM Survival Analysis Clinical Trials Causal Inference · 发表于 2026-03-18

Sample Size Recalculation in Adaptive Group Sequential Study Designs for Comparing Restricted Mean Survival Times

Carolin Herrmann, Paul Blanche

摘要

Non‐proportional hazards cases are frequently expected in clinical trials with time‐to‐event endpoints (e.g., cardiology, oncology). The relevance of hazard ratios to quantify the treatment effect is questionable and potentially misleading in this context. Hence, alternative methods comparing restricted mean survival times are increasingly promoted. Specific challenges arise when planning clinical trials for comparing restricted mean survival times, as several nuisance parameter estimates are needed for calculating the sample size. Precise estimates might be difficult to obtain at the planning stage and might lead to underpowered trials. One way of dealing with this insecurity is to apply adaptive group sequential study designs with the option to adapt the sample size during an ongoing trial. Within this work, we consider such sample size adaptations, with a specific focus on the context of delayed treatment effects. We compare the performance of an adaptive design with the restricted mean survival time as the primary endpoint with other commonly chosen endpoints in this scenario by means of an extensive simulation study. With our proposed method, adaptive designs with the restricted mean survival time as the primary endpoint are now thoroughly explained. The combination test that we describe can also be useful for other adaptations than sample sizes.

Biometrics Diagnostics & Prediction Bayesian Methods Clinical Trials · 发表于 2026-03-18

SIMBA–a Bayesian decision framework for the identification of optimal biomarker subgroups for cancer basket clinical trials

Shijie Yuan, Jiaxin Liu, Zhihua Gong, Xia Qin, Crystal Qin, Yuan Ji, Peter Müller

摘要

Motivated by a multi-indication basket trial aiming to assess the efficacy of a novel biomarker-targeted therapy in gastric or gastroesophageal junction (G/GEJ), pancreatic, and other related cancers, we consider a statistical design and decision-making framework for such trials. Typically, the investigational therapy in the trial targets a biomarker that is present in multiple cancer indications, and patients with higher biomarker expression tend to exhibit higher response rates, assuming the targeting biomarkers are over-expressed in tumor cells. To enable information sharing across indications, the proposed SIMBA method introduces a Bayesian hierarchical model that defines positive and negative biomarker subgroups and identifies optimal go/no-go decisions. The operating characteristics of SIMBA are assessed via simulations and compared against existing methods in the literature. Overall, SIMBA is constructed to improve the identification of patient sub-populations who may benefit from biomarker-targeted therapeutics.

SiM Causal Inference Econometrics & Economics · 发表于 2026-03-18

Meta‐Analysis of Cost‐Effectiveness

Heejung Bang, Hongwei Zhao

摘要

Systematic review and meta‐analysis are widely accepted approaches for evaluating treatment effectiveness. Meta‐analysis generally addresses statistical aspects of systematic reviews, such as the pooling of treatment effect sizes, assessment of heterogeneity, and statistical inference. To complement treatment effectiveness, cost‐effectiveness is often conducted to encompass both clinical and economic perspectives. However, there are few statistical methods proposed for meta‐analyses of cost‐effectiveness, and none is used widely. In fact, meta‐analysis is currently not encouraged for cost‐effectiveness due to methodological and statistical complexities. In this paper, we propose simple meta‐analytic methods for cost‐effectiveness, which may serve as a starting point for future work. We illustrate the methods using two examples from systematic reviews on wound interventions and mental illness.

SMMR Causal Inference Longitudinal Data · 发表于 2026-03-18

Mediation analysis in longitudinal intervention studies with an ordinal treatment-dependent confounder

Mikko Valtanen, Tommi Härkänen, Matti Uusitupa, Jaakko Tuomilehto, Jaana Lindström, Kari Auranen

摘要

In interventional health studies, causal mediation analysis can be employed to investigate mechanisms through which the intervention affects the targeted health outcome. Identifying direct and indirect effects from empirical data become complicated, however, when a confounder of the mediator-outcome association is itself affected by the treatment. Here, we investigate identification of mediational effects under such post-treatment confounding in a setting with a longitudinal mediator, time-to-event outcome and an ordinal treatment-dependent confounder. If the treatment affects the treatment-dependent confounder only in one direction (monotonicity), we show that the mediational effects are identified up to stratum-specific sensitivity parameters and derive their empirical non-parametric expressions. The feasibility of the monotonicity assumption can be assessed using empirical data, based on restrictions on the marginal distributions of counterfactuals of the treatment-dependent confounder. In an empirical analysis, we use data from the Finnish Diabetes Prevention Study to assess the extent to which the effect of a lifestyle intervention on avoiding type 2 diabetes is mediated through weight reduction in a high-risk population, with other health-related changes acting as treatment-dependent confounders. We avoid pitfalls related to post-treatment conditioning by treating the mediator as a functional entity and defining the time-to-event outcome as a restricted disease-free time.

JRSSB Other · 发表于 2026-03-17

Statistical inference for cell type deconvolution

Dongyue Xie, Lin Gui, Jingshu Wang

摘要

Integrating heterogeneous datasets across different measurement platforms poses fundamental challenges for statistical inference. An important example is cell type deconvolution, where cell type proportions in bulk RNA-seq data are estimated using reference single-cell data from different sources, leading to platform-specific scaling effects, measurement noise, and biological heterogeneity. Existing methods often treat estimated proportions as observed in downstream analyses, potentially compromising validity when comparing multiple individuals. We introduce measurement error adjusted deconvolution, a statistical framework for estimation and inference in deconvolution with externally approximated design matrices. We establish necessary and sufficient conditions for identifiability under arbitrary gene-specific cross-platform scaling differences and develop valid inferential procedures for both individual-level proportions and comparisons across individuals, accounting for gene–gene correlation and shared estimation uncertainty. Simulations and real-data analyses demonstrate competitive estimation accuracy and reliable statistical inference.

Biometrics Survival Analysis High-Dimensional · 发表于 2026-03-16

Variable selection in functional linear Cox model

Yuanzhen Yue, Stella Self, Yichao Wu, Jiajia Zhang, Rahul Ghosal

摘要

Modern biomedical studies frequently collect complex, high-dimensional physiological signals using wearables and sensors along with time-to-event outcomes, making efficient variable selection methods crucial for interpretation and improving the accuracy of survival models. We propose a novel variable selection method for a functional linear Cox model with multiple functional and scalar covariates measured at baseline. We utilize a spline-based semiparametric estimation approach for the functional coefficients and a group minimax concave type penalty, which effectively integrates smoothness and sparsity into the estimation of functional coefficients. An efficient group descent algorithm is used for optimization, and an automated procedure is provided to select optimal values of the smoothing and sparsity parameters. Through simulation studies, we demonstrate the method’s ability to perform accurate variable selection and estimation. The method is applied to 2003–2006 cohort of the National Health and Nutrition Examination Survey data, identifying the key temporally varying distributional patterns of physical activity and demographic predictors related to all-cause mortality. Our analysis sheds light on the intricate association between daily distributional patterns of physical activity and all-cause mortality among older US adults.

JASA Survival Analysis Machine Learning Networks & Graphs · 发表于 2026-03-16

Variable Significance Testing for the Deep Cox Model

Qixian Zhong, Jonas Mueller, Jane-Ling Wang

摘要

Deep learning has become enormously popular in the analysis of complex data, including event time measurements with censoring. To date, deep survival methods have mainly focused on prediction. Such methods are scarcely used in matters of statistical inference such as hypothesis testing. Due to their black-box nature, deep-learned outcomes lack interpretability which limits their use for decision-making in biomedical applications. This article provides estimation and inference methods for the nonparametric Cox model—a flexible family of models with a nonparametric link function to avoid model misspecification. Here we assume the nonparametric link function is modeled via a deep neural network. To perform statistical inference, we use sample splitting and cross-fitting procedures to get neural network estimators and construct test statistic. These procedures enable us to propose a new significance test to examine the association of certain covariates with event times. We establish convergence rates of the neural network estimators, and show that deep learning can overcome the curse of dimensionality in nonparametric regression by learning to exploit low-dimensional structures underlying the data. In addition, we show that our test statistic converges to a normal distribution under the null hypothesis and establish its consistency, in terms of the Type II error, under the alternative hypothesis. Numerical simulations and a real data application demonstrate the usefulness of the proposed test. Supplementary materials for this article are available online. *(abstract via OpenAlex)*

JASA Genetics & Omics Survival Analysis · 发表于 2026-03-16

SurvSTAAR: A Powerful Statistical Framework for Rare Variant Analysis of Time-to-Event Traits in Large-Scale Whole-Genome Sequencing Studies

Yidan Cui, Shiyang Ma, Yuxin Yuan, Nengjie Zhu, Haifeng Chen, Ting Wei, Zilin Li, Xihao Li, Zhangsheng Yu

摘要

The increasing availability of large-scale, population-based whole-genome sequencing (WGS) data enables comprehensive analyses of rare genetic variants, which are crucial for unraveling the genetic mechanisms underlying complex traits and diseases. Time-to-event traits offer the advantage of capturing both diagnosis status and timing, facilitating the identification of genetic variants associated with age of onset, disease progression, and lifespan. However, existing methods primarily focus on quantitative and binary traits, which do not leverage censored time information and have limitations in detecting rare variants associated with disease progression. Here we propose SurvSTAAR, a powerful and comprehensive statistical framework for time-to-event traits in large-scale WGS studies, offering a computationally scalable analytical pipeline for analyzing rare coding and noncoding variants. SurvSTAAR accounts for sample relatedness, population structure, heavily censored traits, and further empowers rare variant association analysis by incorporating functional annotations. We applied SurvSTAAR to analyze the time-to-event trait of Alzheimer’s disease (AD) in 458,773 related samples from the UK Biobank WGS data. We identified putatively novel associations with AD in both coding and noncoding regions and further explored their potential role in disease progression by assessing their effects on protein function, including amino acid changes, structural modifications, and domain disruptions. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work. *(abstract via OpenAlex)*

JASA High-Dimensional Nonparametric & Semiparametric · 发表于 2026-03-16

Sparse Gaussianized Canonical Correlation Analysis with Applications to Portfolio Analysis

He Di, Hui Zou

摘要

Canonical correlation analysis (CCA) is an important statistical technique that explores the linear relationships between two sets of variables. In this article, we propose a new generalization of CCA named sparse Gaussianized CCA (SGCCA) for high-dimensional data analysis. SGCCA has a number of favorable properties. First, it is conceptually easy to comprehend and efficient to implement. Second, it not only yields sparse and nested canonical vectors, but is also invariant against monotone transformations of any of the variables and hence robust to heavy-tailed data which is a well-known issue that severely dampens the classical CCA. Furthermore, SGCCA is shown to enjoy both the estimation consistency and variable selection consistency under a semiparametric copula model with mild regularity conditions. Extensive simulations demonstrate the superior performance of SGCCA over existing CCA methods with or without the underlying datagenerating model being a semiparametric copula model. An analysis of correlation structures between consumer cyclical and non-cyclical stocks is illustrated as an empirical application. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work. *(abstract via PubMed)*

JASA Bayesian Methods High-Dimensional · 发表于 2026-03-16

Provably Efficient Posterior Sampling for Sparse Linear Regression via Measure Decomposition

Andrea Montanari, Yuchen Wu

摘要

We consider the problem of sampling from the posterior distribution of a d-dimensional coefficient vector θ, given linear observations y=Xθ+ε. For sparse Bayesian models, such posteriors are in general multimodal, and therefore challenging to sample from. This observation has prompted the exploration of various heuristics that aim at approximating the posterior distribution. In this article, we study a different approach based on decomposing the posterior distribution into a log-concave mixture of simple product measures. This decomposition allows us to reduce sampling from a multimodal distribution of interest to sampling from a log-concave one, which is tractable and has been investigated in detail. We prove that, under mild conditions on the prior, for random designs, such measure decomposition is generally feasible when the number of samples per parameter n/d exceeds a constant threshold. We thus obtain a provably efficient (polynomial time) sampling algorithm in a regime where this was previously not known. Numerical simulations confirm that the algorithm is practical, and reveal that it has attractive statistical properties compared to state-of-the-art methods. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work. *(abstract via OpenAlex)*

JASA High-Dimensional · 发表于 2026-03-16

Portfolio Analysis in High Dimensions with Tracking Error and Weight Constraints

Mehmet Caner, Qingliang Fan

摘要

This paper explores the statistical properties of forming constrained optimal portfolios within a high-dimensional set of assets. We examine portfolios with tracking error constraints, those with simultaneous tracking error and weight restrictions, and portfolios constrained solely by weight. Tracking error measures portfolio performance against a benchmark (typically an index), while weight constraints determine asset allocation based on regulatory requirements or fund prospectuses. Our approach employs a novel statistical learning technique that integrates factor models with nodewise regression, named the C onstrained R esidual Nodewise O ptimal W eight Regression (CROWN) method. We demonstrate its estimation consistency in large dimensions, even when assets outnumber the portfolio’s time span. Convergence rate results for constrained portfolio weights, risk, and Sharpe Ratio are provided, and simulation and empirical evidence highlight the method’s outstanding performance. *(abstract via OpenAlex)*

SiM Survival Analysis Nonparametric & Semiparametric Networks & Graphs · 发表于 2026-03-16

Permutation Tests Based on the Copula‐Graphic Estimator and Their Use for Survival Tree Construction

Pauline Baur, Markus Pauly, Takeshi Emura

摘要

Survival trees are popular alternatives to Cox or Aalen regression models that offer both modeling flexibility and graphical interpretability. This paper introduces a new algorithm for survival trees that relaxes the assumption of independent censoring. To this end, we use the copula‐graphic estimator to estimate survival functions. This allows us to flexibly specify shape and strength of the dependence of survival and censoring times within survival trees. For splitting, we present a permutation test for the null hypothesis of equal survival. Our test statistic consists of the integrated absolute distance of the groups' copula‐graphic estimators. A first simulation study shows a good type I error and power behavior of the new test. We thereby assess simulation settings of various group sizes, censoring percentages, and grades of dependence generated by Clayton and Frank copulas. Using this test as a splitting criterion, a second simulation study studies the performance of the resulting trees and compares it with that of the usual logrank‐based tree. Lastly, the tree algorithm is applied to real‐world clinical trial data.

SiM Survival Analysis Missing Data & Measurement Error · 发表于 2026-03-16

On Window Mean Survival Time With Interval‐Censored Data

Takuto Iijima, Tomotaka Momozaki, Shuji Ando

摘要

In recent years, cancer clinical trials have increasingly encountered non proportional hazards (NPH) scenarios, particularly with the emergence of immunotherapy. In randomized controlled trials comparing immunotherapy with conventional chemotherapy or placebo, late difference and early crossing survival curves scenarios are commonly observed. In such cases, window mean survival time (WMST), the area under the survival curve within a pre‐specified interval , has gained increasing attention due to its superior power compared to restricted mean survival time (RMST), the area under the survival curve up to a pre‐specified time point. Considering the increasing use of progression‐free survival as a co‐primary endpoint alongside overall survival, there is a critical need to establish a WMST estimation method for interval‐censored data; however, sufficient research has yet to be conducted. To bridge this gap, this study proposes a WMST inference method utilizing one‐point imputations and Turnbull's method. Extensive numerical simulations demonstrate that the WMST estimation method using mid‐point imputation for interval‐censored data exhibits comparable performance to that using Turnbull's method. Since the former facilitates standard error calculation, we adopt it as the standard method. Numerical simulations on two‐sample tests confirm that the proposed WMST testing method have higher power than RMST in late difference and early crossing survival curves scenarios, while having compatible power to the log‐rank test under the PH. Furthermore, even when pre‐specified deviated from the clinically desirable time point, WMST consistently maintains higher power than RMST in late difference and early crossing survival curves scenarios.

JASA Causal Inference Genetics & Omics Nonparametric & Semiparametric · 发表于 2026-03-16,收录于 2026-09-10

Nonparametric Causal Inference for Optogenetics: Sequential Excursion Effects for Dynamic Regimes

Gabriel Loewinger, Alexander W. Levis, Francisco Pereira

摘要

(sourced from arXiv preprint 2405.18597v3 — same paper; tandfonline was 403-blocked): "Optogenetics is a powerful neuroscience technique for studying how neural circuit manipulation affects behavior. Standard analysis conventions discard information and severely limit the scope of the causal questions that can be probed. To address this gap, we 1) draw connections to the causal inference literature on sequentially randomized experiments, 2) propose a non-parametric framework for analyzing "open-loop" (static regime) optogenetics behavioral experiments, 3) derive extensions of history-restricted marginal structural models for dynamic treatment regimes with positivity violations for "closed-loop" designs, and 4) propose a taxonomy of identifiable causal effects that encompass a far richer collection of scientific questions compared to standard methods. From another view, our work extends "excursion effect" methods—popularized recently in the mobile health literature—to enable estimation of causal contrasts for treatment sequences in the presence of positivity violations. We describe sufficient conditions for identifiability of the proposed causal estimands, and provide asymptotic statistical guarantees for a proposed inverse probability-weighted estimator, a multiply-robust estimator (for two intervention timepoints), a framework for hypothesis testing, and a computationally scalable implementation. Finally, we apply our framework to data from a recent neuroscience study and show how it provides insight into causal effects of optogenetics on behavior that are obscured by standard analyses.

JASA Other · 发表于 2026-03-16

Model to Meaning: How to Interpret Statistical Models with R and Python

Brenda Betancourt

摘要

Vincent Arel-Bundock. Boca Raton, FL: Chapman & Hall/CRC Press, 2025, ix + 252 pp., $84.99(P), ISBN: 978-1-032-90872-4.Interpreting statistical models and communicating their implications clearly r... *(abstract via OpenAlex)*

JASA Other · 发表于 2026-03-16

Low-Rank Online Dynamic Assortment with Dual Contextual Information

Seong Jin Lee, Will Wei Sun, Yufeng Liu

摘要

As e-commerce expands, delivering real-time personalized recommendations from vast catalogs poses a critical challenge for retail platforms. Maximizing revenue requires careful consideration of both individual customer characteristics and available item features to continuously optimize assortments over time. In this article, we consider the dynamic assortment problem with dual contexts—user and item features. In high-dimensional scenarios, the quadratic growth of dimensions complicates computation and estimation. To tackle this challenge, we introduce a new low-rank dynamic assortment model to transform this problem into a manageable scale. Then we propose an efficient algorithm that estimates the intrinsic subspaces and uses the upper confidence bound approach to address the exploration-exploitation tradeoff in online decision making. Theoretically, we establish a regret bound of O˜((d1+d2)rT), where d1,d2 represent the dimensions of the user and item features, respectively, r is the rank of the parameter matrix, and T denotes the time horizon. This bound represents a substantial improvement over prior literature, achieved by leveraging the low-rank structure. Extensive simulations and an application to the Expedia hotel recommendation dataset further demonstrate the advantages of our proposed method. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work. *(abstract via OpenAlex)*

JASA Latent Variables & Mixtures Networks & Graphs Optimization & Learning Theory · 发表于 2026-03-16

Hyperbolic Network Latent Space Model with Learnable Curvature

Jinming Li, Gongjun Xu, Ji Zhu

摘要

Network data is ubiquitous in various scientific disciplines, including sociology, economics, and neuroscience. Latent space models are often employed in network data analysis, but the geometric effect of latent space curvature remains a significant, unresolved issue. In this work, we propose a hyperbolic network latent space model with a learnable curvature parameter. We theoretically justify that learning the optimal curvature is essential to minimizing the embedding error across all hyperbolic embedding methods beyond network latent space models. A maximum-likelihood estimation strategy, employing manifold gradient optimization, is developed, and we establish the consistency and convergence rates for the maximum-likelihood estimators, both of which are technically challenging due to the non-linearity and non-convexity of the hyperbolic distance metric. We further demonstrate the geometric effect of latent space curvature and the superior performance of the proposed model through extensive simulation studies and an application using a Facebook friendship network. *(abstract via OpenAlex)*

Biometrics Diagnostics & Prediction Survey & Sampling · 发表于 2026-03-16

Generalized entropy calibration for analyzing voluntary survey data

Yonghyun Kwon, Jae Kwang Kim, Yumou Qiu

摘要

Statistical analysis of voluntary survey data is an important area of research in survey sampling. We consider a unified approach to voluntary survey data analysis under the assumption that the sampling mechanism is ignorable. Generalized entropy calibration is introduced as a unified tool for calibration weighting to control selection bias. We first establish the relationship between the generalized calibration weighting and its dual expression for regression estimation. The dual relationship is critical in identifying the implied regression model and developing model selection for calibration weighting. Also, if a linear regression model for an important study variable is available, then a two-step calibration method can be used to smooth the final weights and achieve statistical efficiency. We establish the double robustness and local efficiency of the proposed estimator. Results from a limited simulation study are also presented.

SiM Diagnostics & Prediction · 发表于 2026-03-16

Evaluating Diagnostic Accuracy of Binary Medical Tests in Multi‐Reader Multi‐Case Study

Seungjae Lee, Sowon Jang, Woojoo Lee

摘要

Multi‐reader multi‐case (MRMC) studies are typically conducted to compare the diagnostic performance of medical modalities, which are evaluated by multiple readers interpreting a common set of cases. One of the primary goals of MRMC analysis for binary diagnostic tests is to compare sensitivities and specificities across different imaging modalities. However, the complex correlation structure that is inherent in MRMC data poses significant challenges for analysis. In practice, a generalized estimating equation, a generalized linear mixed model, and McNemar's test are often used in MRMC analysis. In this paper, we explain the theoretical properties of conditional logistic regression applied to MRMC studies and explore its relationship with Cochran's and McNemar's tests. We illustrate the characteristics of the proposed method through extensive simulation studies and real data analysis.

SiM Longitudinal Data Latent Variables & Mixtures High-Dimensional · 发表于 2026-03-16

Dynamic Factor Analysis for Sparse and Irregular Longitudinal Data: An Application to Metabolite Measurements in a COVID‐19 Study

Jiachen Cai, Robert J. B. Goudie, Brian D. M. Tom

摘要

Factor analysis (FA) can be used to identify key biomarkers in biological processes by assuming that latent biological pathways (statistically, “latent factors”) drive the activity of measurable biomarkers (“observed variables”). However, biological pathways often interact, meaning that the classical FA assumption of independence between factors is questionable. Motivated by sparsely and irregularly collected longitudinal measurements of metabolites in a COVID‐19 study, we propose a dynamic factor analysis model that accounts for cross‐correlations between pathways via a multi‐output Gaussian processes (MOGP) prior on the factor trajectories. To mitigate against overfitting caused by sparsity of longitudinal measurements, we introduce a roughness penalty upon MOGP hyperparameters and allow for non‐zero mean functions. We also propose a scalable stochastic expectation maximization (StEM) algorithm that, in simulations, is both 20 times faster and provides more accurate and stable MOGP hyperparameter estimates than a previously‐proposed Monte Carlo Expectation Maximization algorithm. In the motivating COVID‐19 study, our methodology identifies a kynurenine pathway that affects the clinical severity of patients with COVID‐19 disease and uncovers the role of the biomarker taurine. Our R package DFA4SIL implements the proposed method.

JASA Latent Variables & Mixtures · 发表于 2026-03-16

Deep Clustering Evaluation: How to Validate Internal Clustering Validation Measures

Zeya Wang, Chenglong Ye

摘要

Deep clustering partitions complex high-dimensional data using deep neural networks for clustering. It involves projecting data into lower-dimensional embeddings before partitioning, which embarks unique evaluation challenges. Traditional clustering validation measures, designed for low-dimensional spaces, are problematic for deep clustering for two reasons: (a) the curse of dimensionality when applied to the high-dimensional input data, and (b) unreliable comparison of clustering results when applied to embedded data from different embedding spaces, owing to variations in training procedures and model parameter settings. This article addresses these unresolved and often overlooked challenges in evaluating clustering within deep learning. We propose a systematic evaluation framework for internal clustering validation measures that: (a) theoretically establishes why traditional measures are ineffective when applied to input data or across disparate embedding spaces paired with partitioning outcomes; (b) identifies embedding spaces that endorse reliable evaluations by detecting groups with high agreement in ranking partitioning outcomes; and (c) develops a stable and robust scoring scheme by weighting index values computed across these identified embedding spaces. Experiments show that this new framework aligns better with external measures, effectively reducing the misguidance from the improper use of internal validation measures in deep clustering evaluation. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work. *(abstract via OpenAlex)*

JASA Other · 发表于 2026-03-16

Data-Driven Knowledge Transfer in Batch Q * Learning

Elynn Chen, Xi Chen, Wenbo Jing

摘要

In data-driven decision-making across marketing, healthcare, and education, leveraging large datasets from existing ventures is crucial for navigating high-dimensional feature spaces and addressing data scarcity in new ventures. We investigate knowledge transfer in dynamic decision-making by focusing on batch stationary environments and formally defining task discrepancies through the framework of Markov decision processes (MDPs). We propose the Transfer Fitted Q-Iteration algorithm with general function approximation, which enables direct estimation of the optimal action-state function Q* using both target and source data. Under sieve approximation, we establish the relationship between statistical performance and the MDP task discrepancy, highlighting the influence of source and target sample sizes and task discrepancy on the effectiveness of knowledge transfer. Our theoretical and empirical results demonstrate that the final learning error of the function is significantly reduced compared to the single-task learning rate. *(abstract via OpenAlex)*

Biometrics Genetics & Omics High-Dimensional Nonparametric & Semiparametric · 发表于 2026-03-16

DNN-based semiparametric AFT model for integrating genomic and pathological imaging data in cancer prognosis

Jingmao Li, Qingzhao Zhang, Shuangge Ma

摘要

Modeling prognosis has critical implications in cancer research and clinical practice. Many studies have been conducted, built on genomic (and omics in general) and pathological imaging data. In recent research, a handful of studies have also integrated the two types of data, taking advantage of their complementary signals. In this study, we further advance cancer prognosis modeling by developing a semiparametric accelerated failure time model. For modeling the high-dimensional genomic variables—where strong interpretability is desired, we assume parametric effects. For modeling the high-dimensional pathological imaging features—where more flexible/effective modeling can be preferred over interpretability, we assume non-parametric effects. Different from many existing studies, such non-parametric effects are estimated using deep neural networks. To differentiate important genomic and pathological imaging variables from noises, we impose penalization on both the parametric and non-parametric effects. In particular, for the non-parametric effects, we apply group penalization to the first-layer weights. The asymptotic selection and estimation consistency and normality properties are carefully established, which can provide a uniquely strong ground. Computation is examined. Simulation demonstrates competitive performance of the proposed approach. In the analysis of The Cancer Genome Atlas data on lung cancer, the proposed approach leads to satisfactory and sensible findings different from the alternatives.

SiM Clinical Trials · 发表于 2026-03-16,收录于 2026-09-10

Clinical Trial Simulation: Planning With the OCTAVE Framework, Implementation and Validation Principles

K. M. Lee, B. Choodari-Oskooei, M. J. Grayling, P. Jacko, P. K. Kimani, A. Mukherjee, P. Pallmann, T. Parke, D. S. Robertson, Z. Wang, C. Yap, T. Jaki

摘要

(verbatim, in full — via Newcastle University ePrints repository, cross-confirmed via PMC/PubMed): "© 2026 The Author(s). Statistics in Medicine published by John Wiley & Sons Ltd. The adoption of complex innovative clinical trial designs has steadily increased in recent years. These are trial designs that have one or more unconventional features—often resulting in multiple stages—with the goal of improving on conventional single-stage, fixed-setting designs in terms of efficiency, for example, by reducing the required sample size or the time to establish findings about an intervention. The motivation for these designs may not be difficult to follow, but their set-up and implementation is usually more challenging. Statistical properties of these designs can also be difficult to compute. Clinical trial simulation (CTS), which uses software to generate artificial data for learning, can be conducted to identify the (optimal) setting of a clinical trial, evaluate the design's statistical properties under some hypothetical scenarios for sensitivity analysis, and compare different design set-ups and data analysis strategies, all of which contribute to a better understanding of the value of unconventional features before implementing the design in an actual clinical trial. Existing literature on simulation primarily focuses on the evaluation of statistical analysis methods, with less attention on the detailed specification and planning of CTS. This tutorial presents a new framework, called OCTAVE, for outlining the details of CTS, provides practical recommendations for their implementation, and addresses key computational considerations. The target audience is trial statisticians who are involved in designing and analyzing clinical trials. This tutorial covers a range of complex innovative designs, without the expectation that readers are familiar with the mentioned examples.

JASA Causal Inference · 发表于 2026-03-16

Blessing from Human-AI Interaction: Super Policy Learning in Confounded Environments

Jiayi Wang, Chengchun Shi, Zhengling Qi

摘要

As AI becomes more prevalent throughout society, effective methods of integrating humans and AI systems that leverage their respective strengths and mitigate risk have become an important priority. In this article, we introduce the paradigm of super policy learning that takes advantage of Human-AI interaction for data driven sequential decision making. This approach uses the observed action, either from AI or humans, as input for achieving a stronger oracle in policy learning for the decision maker (humans or AI). In the decision process with unmeasured confounding, the actions taken by past agents can offer valuable insights into undisclosed information. By including this information for the policy search in a novel and legitimate manner, the proposed super policy learning will yield a super-policy that is guaranteed to outperform both the standard optimal policy and the behavior one (e.g., past agents’ actions). We call this stronger oracle a blessing from human-AI interaction. Furthermore, to address the issue of unmeasured confounding in finding super-policies using the batch data, a number of nonparametric and causal identifications are established under the framework of proximal causal inference. Building upon on these novel identification results, we develop several super-policy learning algorithms and systematically study their theoretical properties such as finite-sample regret guarantee. Finally, we illustrate the effectiveness of our proposal through extensive simulations and real-world applications. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work. *(abstract via OpenAlex)*

JASA Other · 发表于 2026-03-16

Ball Impurity: Measuring Heterogeneity in General Metric Spaces

Menglu Che, Ting Li, Wenliang Pan, Xueqin Wang, Heping Zhang

摘要

Data in various domains, such as neuroimaging and network data analysis, often come in complex forms without possessing a Hilbert structure. The complexity necessitates innovative approaches for effective analysis. We propose a novel measure of heterogeneity, ball impurity, which is designed to work with complex non-Euclidean objects. Our approach extends the notion of impurity to general metric spaces, providing a versatile tool for feature selection and tree models. The ball impurity measure exhibits desirable properties, such as the triangular inequality, and is computationally tractable, enhancing its practicality and usefulness. Extensive experiments on synthetic data and real data from the UK Biobank validate the efficacy of our approach in capturing data heterogeneity. Remarkably, our results compare favorably with state-of-the-art methods in metric spaces, highlighting the potential of ball impurity as a valuable tool for addressing complex data analysis tasks. *(abstract via OpenAlex)*

JASA Causal Inference · 发表于 2026-03-16

Balancing Weights for Causal Inference in Observational Factorial Studies

Ruoqi Yu, Peng Ding

摘要

Many scientific questions in biomedical, environmental, and psychological research involve understanding the effects of multiple factors on outcomes. While factorial experiments are ideal for this purpose, randomized controlled treatment assignment is generally infeasible in many empirical studies. Therefore, investigators must rely on observational data, where drawing reliable causal inferences for multiple factors remains challenging. As the number of treatment combinations grows exponentially with the number of factors, some treatment combinations can be rare or missing by chance in observed data, further complicating factorial effects estimation. To address these challenges, we propose a novel weighting method tailored to observational studies with multiple factors. Our approach uses weighted observational data to emulate a randomized factorial experiment, enabling simultaneous estimation of the effects of multiple factors and their interactions. Our investigations reveal a crucial nuance: achieving balance among covariates, as in single-factor scenarios, is necessary but insufficient for unbiasedly estimating factorial effects; balancing the factors is also essential in multi-factor settings. Moreover, we extend our weighting method to handle missing treatment combinations in observed data. Finally, we study the asymptotic behavior of the new weighting estimators and propose a consistent variance estimator, providing reliable inferences on factorial effects in observational studies. *(abstract via OpenAlex)*

第 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 页