2025 年的期刊文章 · 第 28 页

共 1491 篇,按发表时间排,新的在前,每页 40 篇。这一页是 2025-01-24 到 2025。

按期刊、按主题看 →

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

SiM Bayesian Methods Spatial & Time Series · 发表于 2025-01-24

A Bayesian Multivariate Model With Temporal Dependence on Random Partition of Areal Data for Mosquito‐Borne Diseases

Jessica Pavani, Fernando Andrés Quintana

摘要

More than half of the world's population is exposed to mosquito‐borne diseases, leading to millions of cases and hundreds of thousands of deaths every year. Analyzing this type of data is complex and poses several interesting challenges, mainly due to the usually vast geographic area involved, the peculiar temporal behavior, and the potential correlation between infections. Motivation for this work stems from the analysis of tropical disease data, namely, the number of cases of dengue and chikungunya, for the 145 microregions in Southeast Brazil from 2018 to 2022. As a contribution to the literature on multivariate disease data, we develop a flexible Bayesian multivariate spatio‐temporal model where temporal dependence is defined for areal clusters. The model features a prior distribution for the random partition of areal data that incorporates neighboring information. It also incorporates an autoregressive structure and terms related to seasonal patterns into temporal components that are disease‐ and cluster‐specific. Furthermore, it considers a multivariate directed acyclic graph autoregressive structure to accommodate spatial and inter‐disease dependence. We explore the properties of the model through simulation studies and show results that prove our proposal compares well to competing alternatives. Finally, we apply the model to the motivating dataset with a twofold goal: finding clusters of areas with similar temporal trends for some of the diseases and exploring the existence of correlation between two diseases transmitted by the same mosquito.

SiM Other · 发表于 2025-01-23

Testing One Primary and Two Secondary Endpoints in a Two‐Stage Group Sequential Trial With Extensions

Ajit C. Tamhane, Dong Xi, Cyrus R. Mehta, Alexander Romanenko, Jiangtao Gou

摘要

We study the problem of testing multiple secondary endpoints conditional on a primary endpoint being significant in a two‐stage group sequential procedure, focusing on two secondary endpoints. This extends our previous work with one secondary endpoint. The test for the secondary null hypotheses is a closed procedure. Application of the Bonferroni test for testing the intersection of the secondary hypotheses results in the Holm procedure while application of the Simes test results in the Hochberg procedure. The focus of the present paper is on developing normal theory analogs of the abovementioned ‐value based tests that take into account (i) the gatekeeping effect of the test on the primary endpoint and (ii) correlations between the endpoints. The normal theory boundaries are determined by finding the least favorable configuration of the correlations and so their knowledge is not needed to apply these procedures. The ‐value based procedures are easy to apply but they are less powerful than their normal theory analogs because they do not take into account the correlations between the endpoints and the gatekeeping effect referred to above. On the other hand, the normal theory procedures are restricted to two secondary endpoints and two stages mainly because of computational difficulties with more than two secondary endpoints and stages. Comparisons between the two types of procedures are given in terms of secondary powers. The sensitivity of the secondary type I error rate and power to unequal information times is studied. Numerical examples and a real case study illustrate the procedures.

SMMR Nonparametric & Semiparametric Survival Analysis · 发表于 2025-01-23

Semiparametric estimator for the covariate-specific receiver operating characteristic curve

Pablo Martínez-Camblor, Juan Carlos Pardo-Fernández

摘要

The study of the predictive ability of a marker is mainly based on the accuracy measures provided by the so-called confusion matrix. Besides, the area under the receiver operating characteristic curve has become a popular index for summarizing the overall accuracy of a marker. However, the nature of the relationship between the marker and the outcome, and the role that potential confounders play in this relationship could be fundamental in order to extrapolate the observed results. Directed acyclic graphs commonly used in epidemiology and in causality, could provide good feedback for learning the possibilities and limits of this extrapolation applied to the binary classification problem. Both the covariate-specific and the covariate-adjusted receiver operating characteristic curves are valuable tools, which can help to a better understanding of the real classification abilities of a marker. Since they are strongly related with the conditional distributions of the marker on the positive (subjects with the studied characteristic) and negative (subjects without the studied characteristic) populations, the use of proportional hazard regression models arises in a very natural way. We explore the use of flexible proportional hazard Cox regression models for estimating the covariate-specific and the covariate-adjusted receiver operating characteristic curves. We study their large- and finite-sample properties and apply the proposed estimators to a real-world problem. The developed code (in R language) is provided on Supplemental Material.

SiM Causal Inference · 发表于 2025-01-23

Reverse Time‐to‐Death as Time‐Scale in Time‐to‐Event Analysis for Studies of Advanced Illness and Palliative Care

Yin Bun Cheung, Xiangmei Ma, Isha Chaudhry, Nan Liu, Qingyuan Zhuang, Grace Meijuan Yang, Chetna Malhotra, Eric Andrew Finkelstein

摘要

Incidence of adverse outcome events rises as patients with advanced illness approach end‐of‐life. Exposures that tend to occur near end‐of‐life, for example, use of wheelchair, oxygen therapy and palliative care, may therefore be found associated with the incidence of the adverse outcomes. We propose a concept of reverse time‐to‐death (rTTD) and its use for the time‐scale in time‐to‐event analysis based on partial likelihood to mitigate the time‐varying confounding. We used data on community‐based palliative care uptake (exposure) and emergency department visits (outcome) among patients with advanced cancer in Singapore to illustrate. We compare the results against that of the common practice of using time‐on‐study (TOS) as time‐scale. Graphical analysis demonstrated that cancer patients receiving palliative care had higher rate of emergency department visits than non‐recipients mainly because they were closer to end‐of‐life, and that rTTD analysis made comparison between patients at the same time‐to‐death. In analysis of a decedent cohort, emergency department visits in relation to palliative care using TOS time‐scale showed significant increase in hazard ratio estimate when observed time‐varying covariates were omitted from statistical adjustment (% change‐in‐estimate = 16.2%; 95% CI 6.4% to 25.6%). There was no such change in otherwise the same analysis using rTTD (% change‐in‐estimate = 3.1%; 95% CI ‐1.0% to 8.5%), demonstrating the ability of rTTD time‐scale to mitigate confounding that intensifies in relation to time‐to‐death. A similar pattern was found in the full cohort. Simulations demonstrated that the proposed method had smaller relative bias and root mean square error than TOS‐based analysis. In conclusion, use of rTTD as time‐scale in time‐to‐event analysis provides a simple and robust approach to control time‐varying confounding in studies of advanced illness, even if the confounders are unmeasured.

SiM Longitudinal Data Missing Data & Measurement Error · 发表于 2025-01-23

Multiple Imputation for Longitudinal Data: A Tutorial

Rushani Wijesuriya, Margarita Moreno‐Betancur, John B. Carlin, Ian R. White, Matteo Quartagno, Katherine J. Lee

摘要

Longitudinal studies are frequently used in medical research and involve collecting repeated measures on individuals over time. Observations from the same individual are invariably correlated and thus an analytic approach that accounts for this clustering by individual is required. While almost all research suffers from missing data, this can be particularly problematic in longitudinal studies as participation often becomes harder to maintain over time. Multiple imputation (MI) is widely used to handle missing data in such studies. When using MI, it is important that the imputation model is compatible with the proposed analysis model. In a longitudinal analysis, this implies that the clustering considered in the analysis model should be reflected in the imputation process. Several MI approaches have been proposed to impute incomplete longitudinal data, such as treating repeated measurements of the same variable as distinct variables or using generalized linear mixed imputation models. However, the uptake of these methods has been limited, as they require additional data manipulation and use of advanced imputation procedures. In this tutorial, we review the available MI approaches that can be used for handling incomplete longitudinal data, including where individuals are clustered within higher‐level clusters. We illustrate implementation with replicable R and Stata code using a case study from the Childhood to Adolescence Transition Study.

SMMR Causal Inference Survival Analysis · 发表于 2025-01-23

Multicategory matched learning for estimating optimal individualized treatment rules in observational studies with application to a hepatocellular carcinoma study

Xuqiao Li, Qiuyan Zhou, Ying Wu, Ying Yan

摘要

One primary goal of precision medicine is to estimate the individualized treatment rules that optimize patients’ health outcomes based on individual characteristics. Health studies with multiple treatments are commonly seen in practice. However, most existing individualized treatment rule estimation methods were developed for the studies with binary treatments. Many require that the outcomes are fully observed. In this article, we propose a matching-based machine learning method to estimate the optimal individualized treatment rules in observational studies with multiple treatments when the outcomes are fully observed or right-censored. We establish theoretical property for the proposed method. It is compared with the existing competitive methods in simulation studies and a hepatocellular carcinoma study.

SiM Other · 发表于 2025-01-23

Bioequivalence Design With Sampling Distribution Segments

Luke Hagar, Nathaniel T. Stevens

摘要

In bioequivalence design, power analyses dictate how much data must be collected to detect the absence of clinically important effects. Power is computed as a tail probability in the sampling distribution of the pertinent test statistics. When these test statistics cannot be constructed from pivotal quantities, their sampling distributions are approximated via repetitive, time‐intensive computer simulation. We propose a novel simulation‐based method to quickly approximate the power curve for many such bioequivalence tests by efficiently exploring segments (as opposed to the entirety) of the relevant sampling distributions. Despite not estimating the entire sampling distribution, this approach prompts unbiased sample size recommendations. We illustrate this method using two‐group bioequivalence tests with unequal variances and overview its broader applicability in clinical design. All methods proposed in this work can be implemented using the developed dent package in R.

SiM Other · 发表于 2025-01-23

Allocation Predictability of Individual Assignments in Restricted Randomization Designs for Two‐Arm Equal Allocation Trials

Wenle Zhao, Sherry Livingston

摘要

This manuscript derives the allocation predictability measured by the correct guess probability and the probability of being deterministic for individual treatment assignments, as well as the averages of a randomization sequence, based on the treatment imbalance transition matrix and the conditional allocation probability. The methods described are applicable to restricted randomization designs that satisfy the following criteria: (1) two‐arm equal allocation, (2) restriction of maximum tolerated imbalance, and (3) conditional allocation probability fully determined by the observed current treatment imbalance. Analytical results indicate that, for two‐arm equal allocation trials, allocation predictability alternates by the odd/even sequence order of the treatment assignment. Additionally, the sequence average allocation predictability converges to its asymptotic value significantly more slowly than the allocation predictability for individual assignment does. Consequently, comparisons of allocation predictability between different randomization designs based on sequence averages are sensitive to sequence length. Using sequence average allocation predictability may underestimate the risk of selection bias for individual assignment. This discrepancy is particularly pronounced for short sequence lengths, where individual assignment predictability can be substantially higher than the sequence average.

SiM Bayesian Methods Clinical Trials Causal Inference · 发表于 2025-01-22

Matching‐Assisted Power Prior for Incorporating Real‐World Data in Randomized Clinical Trial Analysis

Ruoyuan Qian, Biqing Yang, Xinyi Xu, Bo Lu

摘要

Leveraging external data information to supplement randomized clinical trials has been a popular topic in recent years, especially for medical device and drug discovery. In rare diseases, it is very challenging to recruit patients and run a large‐scale randomized trial. To take advantage of real‐world data from historical trials on the same disease, we can run a small hybrid trial and borrow historical controls to increase the power. But the borrowing needs to be conducted in a statistically principled manner. Bayesian power prior methods and propensity score adjustments have been discussed in the literature. In this paper, we propose a matching‐assisted power prior approach to better mitigate observed bias when incorporating external data. A subset of comparable external subjects is selected by groups through template matching, and different weights are assigned to these groups based on their similarity to the current study population. Power priors are then implemented to incorporate the information into Bayesian inference. Unlike conventional power prior methods, which discount all control patients similarly, matching pre‐selects good controls, hence improved the quality of external data being borrowed. We compare its performance with the existing propensity score‐integrated power prior approach through simulation studies and illustrate the implementation using data from a real acupuncture clinical trial.

JASA Causal Inference Latent Variables & Mixtures · 发表于 2025-01-22

Inferring Causal Effect of a Digital Communication Strategy under a Latent Sequential Ignorability Assumption and Treatment Noncompliance

Yuki Ohnishi, Bikram Karmakar, Wreetabrata Kar

摘要

Organizations are increasingly relying on digital communications, such as targeted e-mails and mobile notifications, to engage with their audiences. Despite the evident advantages like cost-effectiveness and customization, assessing the effectiveness of such communications from observational data poses various statistical challenges. An immediate challenge is to adjust for targeting rules used in these communications. When digital communications involve a sequence of e-mails or notifications, however, further adjustments are required to correct for selection bias arising from previous communications influencing the subsequent ones and to deal with noncompliance issues, for example, not opening the e-mail. This article addresses these challenges in a study of promotional e-mail sequences sent by a U.S. retailer. We use a Bayesian methodology for causal inference from longitudinal data, considering targeting, noncompliance, and sequential confounding with unmeasured variables. The methodology serves three objectives: to evaluate the average treatment effect of any deterministic e-mailing strategy, to compare the effectiveness of these strategies across varying compliance behaviors, and to infer optimal strategies for distinct customer segments. Our analysis finds, among other things, that certain promotional e-mails effectively maintain engagement among individuals who have regularly received such incentives, and individuals who consistently open their e-mails exhibit reduced sensitivity to promotional content. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work. *(abstract via OpenAlex)*

SiM Genetics & Omics Bayesian Methods · 发表于 2025-01-22

Bayesian Modeling of Cancer Outcomes Using Genetic Variables Assisted by Pathological Imaging Data

Yunju Im, Rong Li, Shuangge Ma

摘要

With the increasing maturity of genetic profiling, an essential and routine task in cancer research is to model disease outcomes/phenotypes using genetic variables. Many methods have been successfully developed. However, oftentimes, empirical performance is unsatisfactory because of a “lack of information.” In cancer research and clinical practice, a source of information that is broadly available and highly cost‐effective comes from pathological images, which are routinely collected for definitive diagnosis and staging. In this article, we consider a Bayesian approach for selecting relevant genetic variables and modeling their relationships with a cancer outcome/phenotype. We propose borrowing information from (manually curated, low‐dimensional) pathological imaging features via reinforcing the same selection results for the cancer outcome and imaging features. We further develop a weighting strategy to accommodate the scenario where information borrowing may not be equally effective for all subjects. Computation is carefully examined. Simulations demonstrate competitive performance of the proposed approach. We analyze TCGA (The Cancer Genome Atlas) LUAD (lung adenocarcinoma) data, with overall survival and gene expressions being the outcome and genetic variables, respectively. Findings different from the alternatives and with sound properties are made.

Biometrics Survival Analysis Meta-Analysis & Data Integration Nonparametric & Semiparametric · 发表于 2025-01-21

Improving estimation efficiency for survival data analysis by integrating a coarsened time-to-event outcome from an external study

Daxuan Deng, Lijun Zhang, Hao Feng, Vernon M Chinchilli, Chixiang Chen, Ming Wang

摘要

In the era of big data, increasing availability of data makes combining different data sources to obtain more accurate estimations a popular topic. However, the development of data integration is often hindered by the heterogeneity in data forms across studies. In this paper, we focus on a case in survival analysis where we have primary study data with a continuous time-to-event outcome and complete covariate measurements, while the data from an external study contain an outcome observed at regular intervals, and only a subset of covariates is measured. To incorporate external information while accounting for the different data forms, we posit working models and obtain informative weights by empirical likelihood, which will be used to construct a weighted estimator in the main analysis. We have established the theory demonstrating that the new estimator has higher estimation efficiency compared to the conventional ones, and this advantage is robust to working model misspecification, as confirmed in our simulation studies. To assess its utility, we apply our method to accommodate data from the National Alzheimer’s Coordinating Center to improve the analysis of the Alzheimer’s Disease Neuroimaging Initiative Phase 1 study.

SMMR Clinical Trials · 发表于 2025-01-19

The effect of estimating prevalences on the population-wise error rate

Remi Luschei, Werner Brannath

摘要

The population-wise error rate is a type I error rate for clinical trials with multiple target populations. In such trials, a treatment is tested for its efficacy in each population. The population-wise error rate is defined as the probability that a randomly selected, future patient will be exposed to an inefficient treatment based on the study results. It can be understood and computed as an average of strata-specific family wise error rates and involves the prevalences of these strata. A major issue of this concept is that the prevalences are usually unknown in practice, so that the population-wise error rate cannot be directly controlled. Instead, one could use an estimator based on the given sample, like their maximum-likelihood estimator under a multinomial distribution. In this article, we demonstrate through simulations that this does not substantially inflate the true population-wise error rate. We differentiate between the expected population-wise error rate, which is almost perfectly controlled, and study-specific values of the population-wise error rate which are conditioned on all subgroup sample sizes and vary within a narrow range. Thereby, we consider up to eight different overlapping populations and moderate to large sample sizes. In these settings, we also consider the maximum strata-wise family wise error rate, which is found to be, on average, at least bounded by twice the significance level used for population-wise error rate control.

SMMR Survival Analysis Clinical Trials · 发表于 2025-01-19

Group sequential design using restricted mean survival time as the primary endpoint in clinical trials

Zhaojin Li, Xiang Geng, Yawen Hou, Zheng Chen

摘要

The proportional hazards (PH) assumption is often violated in clinical trials. If the most commonly used Log-rank test is used for trial design in non-proportional hazard (NPH) cases, it will result in power loss or inflation, and the effect measures hazard ratio will become difficult to interpret. To circumvent the issue caused by the NPH for trial design and to make the effect measures easy to interpret and communicate, two simulation-free methods about restricted mean survival time group sequential (GS-RMST) design are introduced in this study: the independent increment GS-RMST (GS-RMSTi) design and the non-independent increment GS-RMST (GS-RMSTn) design. For the above two designs, the corresponding analytic expression of the variance-covariance matrix, the calculations of the stopping boundaries and sample size are given in the study. Simulation studies show that both designs can achieve the corresponding nominal type I error and nominal power. The GS-RMSTn simulation studies show that the Max-Combo test group sequential design is robust in different NPH scenarios and is suitable for discovering whether there is a treatment effect difference. However, it does not have a corresponding easy-to-interpret effect measure indicating effect difference magnitude. GS-RMST performs well in both PH and NPH scenarios, and it can obtain time-scale effect measures that are easy to understand by both physicians and patients. Examples of both GS-RMST designs are also illustrated.

SMMR Meta-Analysis & Data Integration Causal Inference Clinical Trials · 发表于 2025-01-19

Estimating target population treatment effects in meta-analysis with individual participant-level data

Hwanhee Hong, Lu Liu, Elizabeth A Stuart

摘要

Meta-analysis of randomized controlled trials is commonly used to evaluate treatments and inform policy decisions because it provides comprehensive summaries of all available evidence. However, meta-analyses are limited to draw population inference of treatment effects because they usually do not define target populations of interest specifically, and results of the individual randomized controlled trials in those meta-analyses may not generalize to the target populations. To leverage evidence from multiple randomized controlled trials in the generalizability context, we bridge the ideas from meta-analysis and causal inference. We integrate meta-analysis with causal inference approaches estimating target population average treatment effect. We evaluate the performance of the methods via simulation studies and apply the methods to generalize meta-analysis results from randomized controlled trials of treatments on schizophrenia to adults with schizophrenia who present to usual care settings in the United States. Our simulation results show that all methods perform comparably and well across different settings. The data analysis results show that the treatment effect in the target population is meaningful, although the effect size is smaller than the sample average treatment effect. We recommend applying multiple methods and comparing the results to ensure robustness, rather than relying on a single method.

SMMR Survival Analysis Clinical Trials · 发表于 2025-01-19

Efficient estimation of the marginal mean of recurrent events in randomized controlled trials

Luca Genetti, Giuliana Cortese, Henrik Ravn, Thomas Scheike

摘要

Recurrent events data are often encountered in biomedical settings, where individuals may also experience a terminal event such as death. A useful estimand to summarize such data is the marginal mean of the cumulative number of recurrent events up to a specific time horizon, allowing also for the possible presence of a terminal event. Recently, it was found that augmented estimators can estimate this quantity efficiently, providing improved inference. Improvement in efficiency by the use of covariate adjustment is increasing in popularity as the methods get further developed, and is supported by regulatory agencies EMA (2015) and FDA (2023). Motivated by these arguments, this article presents novel efficient estimators for clinical data from randomized controlled trials, accounting for additional information from auxiliary covariates. Moreover, in randomized studies when both right censoring and competing risks are present, we propose a novel doubly augmented estimator of the marginal mean , which has two optimal augmentation components due to censoring and randomization. We provide theoretical and asymptotic details for the novel estimators, also confirmed by simulation studies. Then, we discuss how to improve efficiency, both theoretically by computing the expected amount of variance reduction, and practically by showing the performance of different working regression models that are needed in the augmentation, when they are correctly specified or misspecified. The methods are applied to the LEADER study, a randomized controlled trial that studied cardiovascular safety of treatments in type 2 diabetes patients.

SMMR Other · 发表于 2025-01-19

A contaminated regression model for count health data

Arnoldus F Otto, Johannes T Ferreira, Salvatore Daniele Tomarchio, Andriëtte Bekker, Antonio Punzo

摘要

In medical and health research, investigators are often interested in countable quantities such as hospital length of stay (e.g., in days) or the number of doctor visits. Poisson regression is commonly used to model such count data, but this approach can’t accommodate overdispersion—when the variance exceeds the mean. To address this issue, the negative binomial (NB) distribution (NB-D) and, by extension, NB regression provide a well-documented alternative. However, real-data applications present additional challenges that must be considered. Two such challenges are (i) the presence of (mild) outliers that can influence the performance of the NB-D and (ii) the availability of covariates that can enhance inference about the mean of the count variable of interest. To jointly address these issues, we propose the contaminated NB (cNB) distribution that exhibits the necessary flexibility to accommodate mild outliers. This model is shown to be simple and intuitive in interpretation. In addition to the parameters of the NB-D, our proposed model has a parameter describing the proportion of mild outliers and one specifying the degree of contamination. To allow available covariates to improve the estimation of the mean of the cNB distribution, we propose the cNB regression model. An expectation-maximization algorithm is outlined for parameter estimation, and its performance is evaluated through a parameter recovery study. The effectiveness of our model is demonstrated via a sensitivity analysis and on two health datasets, where it outperforms well-known count models. The methodology proposed is implemented in an R package which is available at https://github.com/arnootto/cNB .

JRSSB High-Dimensional · 发表于 2025-01-17

Prediction sets for high-dimensional mixture of experts models

Adel Javanmard, Simeng Shao, Jacob Bien

摘要

Large datasets make it possible to build predictive models that can capture heterogenous relationships between the response variable and features. The mixture of high-dimensional linear experts model posits that observations come from a mixture of high-dimensional linear regression models, where the mixture weights are themselves feature-dependent. In this article, we show how to construct valid prediction sets for an ℓ1-penalized mixture of experts model in the high-dimensional setting. We make use of a debiasing procedure to account for the bias induced by the penalization and propose a novel strategy for combining intervals to form a prediction set with coverage guarantees in the mixture setting. Synthetic examples and an application to the prediction of critical temperatures of superconducting materials show our method to have reliable practical performance.

Biometrics Genetics & Omics · 发表于 2025-01-17

Change surface regression for nonlinear subgroup identification with application to warfarin pharmacogenomics data

Pan Liu, Yaguang Li, Jialiang Li

摘要

Pharmacogenomics stands as a pivotal driver toward personalized medicine, aiming to optimize drug efficacy while minimizing adverse effects by uncovering the impact of genetic variations on inter-individual outcome variability. Despite its promise, the intricate landscape of drug metabolism introduces complexity, where the correlation between drug response and genes can be shaped by numerous nongenetic factors, often exhibiting heterogeneity across diverse subpopulations. This challenge is particularly pronounced in datasets such as the International Warfarin Pharmacogenetic Consortium (IWPC), which encompasses diverse patient information from multiple nations. To capture the between-patient heterogeneity in dosing requirement, we formulate a novel change surface model as a model-based approach for multiple subgroup identification in complex datasets. A key feature of our approach is its ability to accommodate nonlinear subgroup divisions, providing a clearer understanding of dynamic drug-gene associations. Furthermore, our model effectively handles high-dimensional data through a doubly penalized approach, ensuring both interpretability and adaptability. We propose an iterative 2-stage method that combines a change point detection technique in the first stage with a smoothed local adaptive majorize-minimization algorithm for surface regression in the second stage. Performance of the proposed methods is evaluated through extensive numerical studies. Application of our method to the IWPC dataset leads to significant new findings, where 3 subgroups subject to different pharmacogenomic relationships are identified, contributing valuable insights into the complex dynamics of drug-gene associations in patients.

SMMR Survival Analysis Causal Inference · 发表于 2025-01-17

Adjusting for switches to multiple treatments: Should switches be handled separately or combined?

Helen Bell Gorrod, Shahrul Mt-Isa, Jingyi Xuan, Kristel Vandormael, William Malbecq, Victoria Yorke-Edwards, Ian R White, Nicholas Latimer

摘要

Treatment switching is common in randomised controlled trials (RCTs). Participants may switch onto a variety of different treatments, all of which may have different treatment effects. Adjustment analyses that target hypothetical estimands – estimating outcomes that would have been observed in the absence of treatment switching – have focused primarily on a single type of switch. In this study, we assess the performance of applications of inverse probability of censoring weights (IPCW) and two-stage estimation (TSE) which adjust for multiple switches by either (i) adjusting for each type of switching separately (‘treatments separate’) or (ii) adjusting for switches combined without differentiating between switched-to treatments (‘treatments combined’). We simulate 48 scenarios in which RCT participants may switch to multiple treatments. Switch proportions, treatment effects, number of switched-to treatments and censoring proportions were varied. Method performance measures included mean percentage bias in restricted mean survival time and the frequency of model convergence. Similar levels of bias were produced by treatments combined and treatments separate in both TSE and IPCW applications. In the scenarios examined, there was no demonstrable advantage associated with adjusting for each type of switch separately, compared with adjusting for all switches together.

Biometrics High-Dimensional Causal Inference · 发表于 2025-01-15

Penalized G-estimation for effect modifier selection in a structural nested mean model for repeated outcomes

Ajmery Jaman, Guanbo Wang, Ashkan Ertefaie, Michèle Bally, Renée Lévesque, Robert W Platt, Mireille E Schnitzer

摘要

Effect modification occurs when the impact of the treatment on an outcome varies based on the levels of other covariates known as effect modifiers. Modeling these effect differences is important for etiological goals and for purposes of optimizing treatment. Structural nested mean models (SNMMs) are useful causal models for estimating the potentially heterogeneous effect of a time-varying exposure on the mean of an outcome in the presence of time-varying confounding. A data-adaptive selection approach is necessary if the effect modifiers are unknown a priori and need to be identified. Although variable selection techniques are available for estimating the conditional average treatment effects using marginal structural models or for developing optimal dynamic treatment regimens, all of these methods consider a single end-of-follow-up outcome. In the context of an SNMM for repeated outcomes, we propose a doubly robust penalized G-estimator for the causal effect of a time-varying exposure with a simultaneous selection of effect modifiers and prove the oracle property of our estimator. We conduct a simulation study for the evaluation of its performance in finite samples and verification of its double-robustness property. Our work is motivated by the study of hemodiafiltration for treating patients with end-stage renal disease at the Centre Hospitalier de l’Université de Montréal. We apply the proposed method to investigate the effect heterogeneity of dialysis facility on the repeated session-specific hemodiafiltration outcomes.

Biometrics Survival Analysis High-Dimensional Functional Data · 发表于 2025-01-14

High-dimensional partially linear functional Cox models

Xin Chen, Hua Liu, Jiaqi Men, Jinhong You

摘要

As a commonly employed method for analyzing time-to-event data involving functional predictors, the functional Cox model assumes a linear relationship between the functional principal component (FPC) scores of the functional predictors and the hazard rates. However, in practical scenarios, such as our study on the survival time of kidney transplant recipients, this assumption often fails to hold. To address this limitation, we introduce a class of high-dimensional partially linear functional Cox models, which accommodates the non-linear effects of functional predictors on the response and allows for diverging numbers of scalar predictors and FPCs as the sample size increases. We employ the group smoothly clipped absolute deviation method to select relevant scalar predictors and FPCs, and use B-splines to obtain a smoothed estimate of the non-linear effect. The finite sample performance of the estimates is evaluated through simulation studies. The model is also applied to a kidney transplant dataset, allowing us to make inferences about the non-linear effects of functional predictors on patients’ hazard rates, as well as to identify significant scalar predictors for long-term survival time.

Biometrics Causal Inference · 发表于 2025-01-13

Causal inference with cross-temporal design

Yi Cao, Pedro L Gozalo, Roee Gutman

摘要

When many participants in a randomized trial do not comply with their assigned intervention, the randomized encouragement design is a possible solution. In this design, the causal effects of the intervention can be estimated among participants who would have experienced the intervention if encouraged. For many policy interventions, encouragements cannot be randomized and investigators need to rely on observational data. To address this, we propose a cross-temporal design, which uses time to mimic a randomized encouragement experiment. However, time may be confounded with temporal trends that influence the outcomes. To disentangle these trends from the intervention effects, we replace the commonly used exclusion restrictions with temporal assumptions. We develop Bayesian procedures to estimate the causal effects and compare it to instrumental variables and matching approaches in simulations. The Bayesian approach outperforms the other 2 approaches in terms of estimation accuracy, and it is relatively robust to various violations of the common trends assumption. Taking advantage of the expansion of the Medicare Advantage (MA) program between 2011 and 2017, we implement the proposed method to estimate the effects of MA enrollment on the risk of skilled nursing facility residents being re-hospitalized within 30 days after discharge from the hospital.

Biometrics Networks & Graphs Bayesian Methods · 发表于 2025-01-11

Robust Bayesian graphical regression models for assessing tumor heterogeneity in proteomic networks

Tsung-Hung Yao, Yang Ni, Anindya Bhadra, Jian Kang, Veerabhadran Baladandayuthapani

摘要

Graphical models are powerful tools to investigate complex dependency structures in high-throughput datasets. However, most existing graphical models make one of two canonical assumptions: (i) a homogeneous graph with a common network for all subjects or (ii) an assumption of normality, especially in the context of Gaussian graphical models. Both assumptions are restrictive and can fail to hold in certain applications such as proteomic networks in cancer. To this end, we propose an approach termed robust Bayesian graphical regression (rBGR) to estimate heterogeneous graphs for non-normally distributed data. rBGR is a flexible framework that accommodates non-normality through random marginal transformations and constructs covariate-dependent graphs to accommodate heterogeneity through graphical regression techniques. We formulate a new characterization of edge dependencies in such models called conditional sign independence with covariates, along with an efficient posterior sampling algorithm. In simulation studies, we demonstrate that rBGR outperforms existing graphical regression models for data generated under various levels of non-normality in both edge and covariate selection. We use rBGR to assess proteomic networks in lung and ovarian cancers to systematically investigate the effects of immunogenic heterogeneity within tumors. Our analyses reveal several important protein–protein interactions that are differentially associated with the immune cell abundance; some corroborate existing biological knowledge, whereas others are novel findings.

Biometrics Spatial & Time Series Meta-Analysis & Data Integration · 发表于 2025-01-11

Distributed model building and recursive integration for big spatial data modeling

Emily C Hector, Brian J Reich, Ani Eloyan

摘要

Motivated by the need for computationally tractable spatial methods in neuroimaging studies, we develop a distributed and integrated framework for estimation and inference of Gaussian process model parameters with ultra-high-dimensional likelihoods. We propose a shift in viewpoint from whole to local data perspectives that is rooted in distributed model building and integrated estimation and inference. The framework’s backbone is a computationally and statistically efficient integration procedure that simultaneously incorporates dependence within and between spatial resolutions in a recursively partitioned spatial domain. Statistical and computational properties of our distributed approach are investigated theoretically and in simulations. The proposed approach is used to extract new insights into autism spectrum disorder from the autism brain imaging data exchange.

JRSSB Spatial & Time Series · 发表于 2025-01-09

Spatial effect detection regression for large-scale spatio-temporal covariates

Chenlin Zhang, Ling Zhou, Bin Guo, Huazhen Lin

摘要

We develop a Spatial Effect Detection Regression (SEDR) model to capture the nonlinear and irregular effects of high-dimensional spatio-temporal predictors on a scalar outcome. Specifically, we assume that both the component and the coefficient functions in the SEDR are unknown smooth functions of location and time. This allows us to leverage spatially and temporally correlated information, transforming the curse of dimensionality into a blessing, as confirmed by our theoretical and numerical results. Moreover, we introduce a set of 0–1 regression coefficients to automatically identify the boundaries of the spatial effect, implemented via a novel penalty. A simple iterative algorithm, with explicit forms at each update step, is developed, and we demonstrate that it converges from the initial values given in the paper. Furthermore, we establish the convergence rate and selection consistency of the proposed estimator under various scenarios involving dimensionality and the effect space. Through simulation studies, we thoroughly evaluate the superior performance of our method in terms of bias and empirical efficiency. Finally, we apply the method to analyse and forecast data from environmental monitoring and Alzheimer’s Disease Neuroimaging Initiative study, revealing interesting findings and achieving smaller out-of-sample prediction errors compared to existing methods.

Biometrics Optimization & Learning Theory Causal Inference Missing Data & Measurement Error · 发表于 2025-01-07

Weighted Q-learning for optimal dynamic treatment regimes with nonignorable missing covariates

Jian Sun, Bo Fu, Li Su

摘要

Dynamic treatment regimes (DTRs) formalize medical decision-making as a sequence of rules for different stages, mapping patient-level information to recommended treatments. In practice, estimating an optimal DTR using observational data from electronic medical record (EMR) databases can be complicated by nonignorable missing covariates resulting from informative monitoring of patients. Since complete case analysis can provide consistent estimation of outcome model parameters under the assumption of outcome-independent missingness, Q-learning is a natural approach to accommodating nonignorable missing covariates. However, the backward induction algorithm used in Q-learning can introduce challenges, as nonignorable missing covariates at later stages can result in nonignorable missing pseudo-outcomes at earlier stages, leading to suboptimal DTRs, even if the longitudinal outcome variables are fully observed. To address this unique missing data problem in DTR settings, we propose 2 weighted Q-learning approaches where inverse probability weights for missingness of the pseudo-outcomes are obtained through estimating equations with valid nonresponse instrumental variables or sensitivity analysis. The asymptotic properties of the weighted Q-learning estimators are derived, and the finite-sample performance of the proposed methods is evaluated and compared with alternative methods through extensive simulation studies. Using EMR data from the Medical Information Mart for Intensive Care database, we apply the proposed methods to investigate the optimal fluid strategy for sepsis patients in intensive care units.

Biometrics Spatial & Time Series · 发表于 2025-01-07

Gaussian processes for time series with lead–lag effects with applications to biology data

Wancen Mu, Jiawen Chen, Eric S Davis, Kathleen Reed, Douglas Phanstiel, Michael I Love, Didong Li

摘要

Investigating the relationship, particularly the lead–lag effect, between time series is a common question across various disciplines, especially when uncovering biological processes. However, analyzing time series presents several challenges. Firstly, due to technical reasons, the time points at which observations are made are not at uniform intervals. Secondly, some lead–lag effects are transient, necessitating time-lag estimation based on a limited number of time points. Thirdly, external factors also impact these time series, requiring a similarity metric to assess the lead–lag relationship. To counter these issues, we introduce a model grounded in the Gaussian process, affording the flexibility to estimate lead–lag effects for irregular time series. In addition, our method outputs dissimilarity scores, thereby broadening its applications to include tasks such as ranking or clustering multiple pairwise time series when considering their strength of lead–lag effects with external factors. Crucially, we offer a series of theoretical proofs to substantiate the validity of our proposed kernels and the identifiability of kernel parameters. Our model demonstrates advances in various simulations and real-world applications, particularly in the study of dynamic chromatin interactions, compared to other leading methods.

JASA Other · 发表于 2025-01-03

When Frictions Are Fractional: Rough Noise in High-Frequency Data

Carsten H. Chong, Thomas Delerue, Guoying Li

摘要

The analysis of high-frequency financial data is often impeded by the presence of noise. This article is motivated by intraday return data in which market microstructure noise appears to be rough, that is, best captured by a continuous-time stochastic process that locally behaves as fractional Brownian motion. Assuming that the underlying efficient price process follows a continuous Itô semimartingale, we derive consistent estimators and asymptotic confidence intervals for the roughness parameter of the noise and the integrated price and noise volatilities, in all cases where these quantities are identifiable. In addition to desirable features such as serial dependence of increments, compatibility between different sampling frequencies and diurnal effects, the rough noise model can further explain divergence rates in volatility signature plots that vary considerably over time and between assets. 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 2025-01-03

Unlocking Retrospective Prevalent Information in EHRs—A Revisit to the Pairwise Pseudolikelihood

Nir Keret, Malka Gorfine

摘要

Electronic health records offer abundant data on various diseases and health conditions, enabling researchers to explore the relationship between disease onset age and underlying risk factors.Unlike mortality data, the event of interest is nonterminal, hence, individuals can retrospectively report their disease-onset-age upon recruitment to the study.These individuals, diagnosed with the disease before entering the study, are termed "prevalent." The ascertainment imposes a left truncation condition, also known as a "delayed entry, " because individuals had to survive a certain period before being eligible for enrollment.The standard method to accommodate delayed entry conditions on the entire history up to recruitment, hence, the retrospective prevalent failure times are conditioned upon and cannot participate in estimating the disease-onset-age distribution.Other methods that condition on less information and allow the incorporation of the prevalent observations either bring about numerical and computational difficulties or require statistical assumptions that are violated by most biobanks.This work presents a novel estimator of the coefficients in a regression model for the age-at-onset, successfully using the prevalent data.Asymptotic results are provided, and simulations are conducted to showcase the substantial efficiency gain.In particular, the method is highly useful in leveraging large-scale repositories for replication analysis of genetic variants.Indeed, analysis of urinary bladder cancer data reveals that the proposed approach yields about twice as many replicated discoveries compared to the popular approach.Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work. *(abstract via OpenAlex)*

JASA Testing & Uncertainty · 发表于 2025-01-03

Simulation-Based, Finite-Sample Inference for Privatized Data

Jordan Awan, Zhanyu Wang

摘要

Privacy protection methods, such as differentially private mechanisms, introduce noise into resulting statistics which often produces complex and intractable sampling distributions. In this article, we propose a simulation-based “repro sample” approach to produce statistically valid confidence intervals and hypothesis tests, which builds on the work of Xie and Wang. We show that this methodology is applicable to a wide variety of private inference problems, appropriately accounts for biases introduced by privacy mechanisms (such as by clamping), and improves over other state-of-the-art inference methods such as the parametric bootstrap in terms of the coverage and Type I error of the private inference. We also develop significant improvements and extensions for the repro sample methodology for general models (not necessarily related to privacy), including (a) modifying the procedure to ensure guaranteed coverage and Type I errors, even accounting for Monte Carlo error, and (b) proposing efficient numerical algorithms to implement the confidence intervals and p-values. 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 Bayesian Methods Nonparametric & Semiparametric · 发表于 2025-01-03

Bayesian Clustering via Fusing of Localized Densities

Alexander Dombowsky, David B. Dunson

摘要

Bayesian clustering typically relies on mixture models, with each component interpreted as a different cluster. After defining a prior for the component parameters and weights, Markov chain Monte Carlo (MCMC) algorithms are commonly used to produce samples from the posterior distribution of the component labels. The data are then clustered by minimizing the expectation of a clustering loss function that favors similarity to the component labels. Unfortunately, although these approaches are routinely implemented, clustering results are highly sensitive to kernel misspecification. For example, if Gaussian kernels are used but the true density of data within a cluster is even slightly non-Gaussian, then clusters will be broken into multiple Gaussian components. To address this problem, we develop Fusing of Localized Densities (FOLD), a novel clustering method that melds components together using the posterior of the kernels. FOLD has a fully Bayesian decision theoretic justification, naturally leads to uncertainty quantification, can be easily implemented as an add-on to MCMC algorithms for mixtures, and favors a small number of distinct clusters. We provide theoretical support for FOLD including clustering optimality under kernel misspecification. In simulated experiments and real data, FOLD outperforms competitors by minimizing the number of clusters while inferring meaningful group structure. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work. *(abstract via PubMed)*

Biometrics Networks & Graphs Genetics & Omics · 发表于 2025-01-02

Structured feature ranking for genomic marker identification accommodating multiple types of networks

Yeheng Ge, Tao Li, Xingdong Feng, Mengyun Wu, Hailong Liu

摘要

Numerous statistical methods have been developed to search for genomic markers associated with the development, progression, and response to treatment of complex diseases. Among them, feature ranking plays a vital role due to its intuitive formulation and computational efficiency. However, most of the existing methods are based on the marginal importance of molecular predictors and share the limitation that the dependence (network) structures among predictors are not well accommodated, where a disease phenotype usually reflects various biological processes that interact in a complex network. In this paper, we propose a structured feature ranking method for identifying genomic markers, where such network structures are effectively accommodated using Laplacian regularization. The proposed method innovatively investigates multiple network scenarios, where the networks can be known a priori and data-dependently estimated. In addition, we rigorously explore the noise and uncertainty in the networks and control their impacts with proper selection of tuning parameters. These characteristics make the proposed method enjoy especially broad applicability. Theoretical result of our proposal is rigorously established. Compared to the original marginal measure, the proposed network structured measure can achieve sure screening properties with a faster convergence rate under mild conditions. Extensive simulations and analysis of The Cancer Genome Atlas melanoma data demonstrate the improvement of finite sample performance and practical usefulness of the proposed method.

Biometrics Spatial & Time Series · 发表于 2025-01-02

Spatially adaptive variable screening in presurgical functional magnetic resonance imaging data analysis

Yifei Hu, Xinge Jessie Jeng

摘要

Accurate delineation of functional brain regions adjacent to tumors is imperative for planning neurosurgery that preserves critical functions. Functional magnetic resonance imaging (fMRI) plays an increasingly pivotal role in presurgical counseling and planning. In the analysis of presurgical fMRI data, the impact of false negatives on patients surpasses that of false positives because failure to identify functional regions and unintentionally resecting critical tissues can result in severe harm to patients. This paper introduces a novel metric, the Bayesian missed discovery rate (BMDR), designed for controlling false negatives within the voxel-specific mixture model. Building on the BMDR metric, we propose a new variable screening procedure that not only ensures effective control of false negatives but also capitalizes on the spatial structure of fMRI data. In comparison to existing statistical methods in fMRI data analysis, our new procedure directly regulates false negatives at a desirable level and is entirely data-driven. Moreover, it significantly differs from current false-negative control procedures by incorporating spatial information. Numerical examples demonstrate that the new method outperforms several state-of-the-art methods in retaining signal voxels, particularly the subtle ones at the boundaries of functional regions, while achieving a cleaner separation of functional regions from background noise. These findings hold promising implications for planning function-preserving neurosurgery.

JASA Meta-Analysis & Data Integration High-Dimensional · 发表于 2025-01-02

Robust Personalized Federated Learning with Sparse Penalization

Weidong Liu, Xiaojun Mao, Xiaofei Zhang, Xin Zhang

摘要

Federated learning (FL) is an emerging topic due to its advantage in collaborative learning with distributed data. Due to the heterogeneity in the local data-generating mechanism, it is important to consider personalization when developing federated learning methods. In this work, we propose a personalized federated learning (PFL) method to address the robust regression problem. Specifically, we aim to learn the regression weight by solving a Huber loss with the sparse fused penalty. Additionally, we designed our personalized federated learning for robust and sparse regression (PerFL-RSR) algorithm to solve the estimation problem in the federated system efficiently. Theoretically, we show that the proposed PerFL-RSR reaches a convergence rate of O(1/T), and the proposed estimator is statistically consistent. Thorough experiments and real data analysis are conducted to corroborate the theoretical results of our proposed personalized federated learning 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 Networks & Graphs Spatial & Time Series · 发表于 2025-01-02

Modeling Recurrent Failures on Large Directed Networks

Qingqing Zhai, Zhisheng Ye, Cheng Li, Matthew Revie, David B. Dunson

摘要

Many lifeline infrastructure systems consist of thousands of components configured in a complex directed network. Disruption of the infrastructure constitutes a recurrent failure process over a directed network. Statistical inference for such network recurrence data is challenging because of the large number of nodes with irregular connections among them. Motivated by 16 years of Scottish Water operation records, we propose a network Gamma-Poisson Autoregressive NHPP (GPAN) model for recurrent failure data from large-scale directed physical networks. The model consists of two layers: the temporal layer applies a Non-Homogeneous Poisson Process (NHPP) with node-specific frailties, and the spatial layer uses a well-orchestrated gamma-Poisson autoregressive scheme to establish correlations among the frailties. Under the network-GPAN model, we develop a sum-product algorithm to compute the marginal distribution for each frailty conditional on the recurrence data. The marginal conditional frailty distributions are useful for predicting future failures based on historical data. In addition, the ability to rapidly compute these marginal distributions allows adoption of an EM type algorithm for estimation. Through a Bethe approximation, the output from the sum-product algorithm is used to compute maximum log-likelihood estimates. Applying the methods to the Scottish Water network, we demonstrate utility in aiding operation management and risk assessment of the water utility. Supplementary materials for this article are available online including a standardized description of the materials available for reproducing the work. *(abstract via OpenAlex)*

JMLR High-Dimensional · 发表于 2025

skglm: Improving scikit-learn for Regularized Generalized Linear Models

Badr Moufad, Pierre-Antoine Bannier, Quentin Bertrand, Quentin Klopfenstein, Mathurin Massias

摘要

We introduce skglm, an open-source Python package for regularized Generalized Linear Models. Thanks to its composable nature, it supports combining datafits, penalties, and solvers to fit a wide range of models, many of them not included in scikit-learn (e.g. Group Lasso and variants). It uses state-of-the-art algorithms to solve problems involving high-dimensional datasets, providing large speed-ups compared to existing implementations. It is fully compliant with the scikit-learn API and acts as a drop-in replacement for its estimators. Finally, it abides by the standards of open source development and is integrated in the scikit-learn-contrib GitHub organization.

JMLR Other · 发表于 2025

gsplat: An Open-Source Library for Gaussian Splatting

Vickie Ye, Ruilong Li, Justin Kerr, Matias Turkulainen, Brent Yi, Zhuoyang Pan, Otto Seiskari, Jianbo Ye, Jeffrey Hu, Matthew Tancik, Angjoo Kanazawa

摘要

gsplat is an open-source library designed for training and developing Gaussian Splatting methods. It features a front-end with Python bindings compatible with the PyTorch library and a back-end with highly optimized CUDA kernels. gsplat offers numerous features that enhance the optimization of Gaussian Splatting models, which include optimization improvements for speed, memory, and convergence times. Experimental results demonstrate that gsplat achieves up to 10% less training time and 4x less memory than the original implementation. Utilized in several research projects, gsplat is actively maintained on GitHub. Source code is available at https://github.com/nerfstudio-project/gsplat under Apache License 2.0. We welcome contributions from the open-source community.

JMLR Machine Learning · 发表于 2025

depyf: Open the Opaque Box of PyTorch Compiler for Machine Learning Researchers

Kaichao You, Runsheng Bai, Meng Cao, Jianmin Wang, Ion Stoica, Mingsheng Long

摘要

PyTorch 2.x introduces a compiler designed to accelerate deep learning programs. However, for machine learning researchers, fully leveraging the PyTorch compiler can be challenging due to its operation at the Python bytecode level, making it appear as an opaque box. To address this, we introduce depyf, a tool designed to demystify the inner workings of the PyTorch compiler. depyf decompiles the bytecode generated by PyTorch back into equivalent source code and establishes connections between the code objects in the memory and their counterparts in source code format on the disk. This feature enables users to step through the source code line by line using debuggers, thus enhancing their understanding of the underlying processes. Notably, depyf is non-intrusive and user-friendly, primarily relying on two convenient context managers for its core functionality. The project is openly available at https://github.com/thuml/depyf and is recognized as a PyTorch ecosystem project at https://pytorch.org/blog/introducing-depyf.

JMLR Testing & Uncertainty · 发表于 2025

Zono-Conformal Prediction: Zonotope-Based Uncertainty Quantification for Regression and Classification Tasks

Laura Lützow, Michael Eichelbeck, Mykel J. Kochenderfer, Matthias Althoff

摘要

Conformal prediction is a popular uncertainty quantification method that augments a base predictor to return sets of predictions with statistically valid coverage guarantees. However, current methods are often computationally expensive and data-intensive, as they require constructing an uncertainty model before calibration. Moreover, existing approaches typically represent the prediction sets with intervals, which limits their ability to capture dependencies in multi-dimensional outputs. We address these limitations by introducing zono-conformal prediction, a novel approach inspired by interval predictor models and reachset-conformant identification that constructs prediction zonotopes with assured coverage. By placing zonotopic uncertainty sets directly into the model of the base predictor, zono-conformal predictors can be identified via a single, data-efficient linear program. While we can apply zono-conformal prediction to arbitrary nonlinear base predictors, we focus on feed-forward neural networks in this work. Aside from regression tasks, we also construct optimal zono-conformal predictors in classification settings where the output of an uncertain predictor is a set of possible classes. We provide probabilistic coverage guarantees and present methods for detecting outliers in the identification data. In extensive numerical experiments, we show that zono-conformal predictors are less conservative than interval predictor models and standard conformal prediction methods, while achieving a similar coverage over the test data.

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