2022 年的期刊文章 · 第 3 页

共 1481 篇,按发表时间排,新的在前,每页 40 篇。这一页是 2022-11-30 到 2022-11-17。

按期刊、按主题看 →

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

SMMR Survival Analysis Functional Data · 发表于 2022-11-30

Additive hazards model with time-varying coefficients and imaging predictors

Qi Yang, Chuchu Wang, Haijin He, Xiaoxiao Zhou, Xinyuan Song

摘要

Conventional hazard regression analyses frequently assume constant regression coefficients and scalar covariates. However, some covariate effects may vary with time. Moreover, medical imaging has become an increasingly important tool in screening, diagnosis, and prognosis of various diseases, given its information visualization and quantitative assessment. This study considers an additive hazards model with time-varying coefficients and imaging predictors to examine the dynamic effects of potential scalar and imaging risk factors for the failure of interest. We develop a two-stage approach that comprises the high-dimensional functional principal component analysis technique in the first stage and the counting process-based estimating equation approach in the second stage. In addition, we construct the pointwise confidence intervals for the proposed estimators and provide a significance test for the effects of scalar and imaging covariates. Simulation studies demonstrate the satisfactory performance of the proposed method. An application to the Alzheimer’s disease neuroimaging initiative study further illustrates the utility of the methodology.

JASA Other · 发表于 2022-11-29

Statistical Learning for Individualized Asset Allocation

Yi Ding, Yingying Li, Rui Song

摘要

We establish a high-dimensional statistical learning framework for individualized asset allocation. Our proposed methodology addresses continuous-action decision-making with a large number of characteristics. We develop a discretization approach to model the effect of continuous actions and allow the discretization frequency to be large and diverge with the number of observations. We estimate the value function of continuous-action using penalized regression with our proposed generalized penalties that are imposed on linear transformations of the model coefficients. We show that our proposed Discretization and Regression with generalized fOlded concaVe penalty on Effect discontinuity (DROVE) approach enjoys desirable theoretical properties and allows for statistical inference of the optimal value associated with optimal decision-making. Empirically, the proposed framework is exercised with the Health and Retirement Study data in finding individualized optimal asset allocation. The results show that our individualized optimal strategy improves the financial well-being of the population. Supplementary materials for this article are available online. *(abstract via OpenAlex)*

JASA Spatial & Time Series · 发表于 2022-11-29

Modeling Point Referenced Spatial Count Data: A Poisson Process Approach

Diego Morales-Navarrete, Moreno Bevilacqua, Christian Caamaño-Carrillo, Luis M. Castro

摘要

Random fields are useful mathematical tools for representing natural phenomena with complex dependence structures in space and/or time. In particular, the Gaussian random field is commonly used due to its attractive properties and mathematical tractability. However, this assumption seems to be restrictive when dealing with counting data. To deal with this situation, we propose a random field with a Poisson marginal distribution considering a sequence of independent copies of a random field with an exponential marginal distribution as “inter-arrival times” in the counting renewal processes framework. Our proposal can be viewed as a spatial generalization of the Poisson counting process. Unlike the classical hierarchical Poisson Log-Gaussian model, our proposal generates a (non)-stationary random field that is mean square continuous and with Poisson marginal distributions. For the proposed Poisson spatial random field, analytic expressions for the covariance function and the bivariate distribution are provided. In an extensive simulation study, we investigate the weighted pairwise likelihood as a method for estimating the Poisson random field parameters. Finally, the effectiveness of our methodology is illustrated by an analysis of reindeer pellet-group survey data, where a zero-inflated version of the proposed model is compared with zero-inflated Poisson Log-Gaussian and Poisson Gaussian copula models. Supplementary materials for this article, including technical proofs and R code for reproducing the work, are available as an online supplement. *(abstract via OpenAlex)*

SiM Survival Analysis Diagnostics & Prediction Testing & Uncertainty · 发表于 2022-11-29

A nonparametric simultaneous confidence band for biomarker effect on the restricted mean survival time

Wen Teng, Wenyu Jiang, Bingshu E. Chen

摘要

Study of prognostic and predictive biomarkers plays an important role in the design and analysis of clinical trials. The Cox proportional hazards model is often used to study the biomarker main effect and the treatment‐biomarker interaction effect for survival data. The estimated effects can be biased if the proportional hazards assumption is violated. The restricted mean survival time is becoming popular in clinical studies for having a clear intuitive interpretation. In this article, we first propose nonparametric methods to make statistical inference for the one‐sample problem of the biomarker effect on the restricted mean survival time; we then extend the methods to the two‐sample problem for studying the difference in the biomarker effects between treatment groups in clinical trials. For a given biomarker, the restricted mean survival time is estimated by kernel smoothing methods with the inverse probability of censoring weights. We prove the consistency for the estimates and develop simultaneous confidence bands for the biomarker effects on the restricted mean survival time. The simultaneous confidence bands are evaluated in extensive simulation studies and are found to have good finite sample performance. We then apply the proposed methods to a breast cancer study conducted by the Breast International Group (BIG) to illustrate how the Ki67 biomarker, a protein marker of cell proliferation, affects the survival time of patients, compared between the treatment groups.

SiM Spatial & Time Series Bayesian Methods Machine Learning · 发表于 2022-11-28

Spatially informed Bayesian neural network for neurodegenerative diseases classification

David Payares‐Garcia, Jorge Mateu, Wiebke Schick

摘要

Magnetic resonance imaging (MRI) plays an increasingly important role in the diagnosis and prognosis of neurodegenerative diseases. One field of extensive clinical use of MRI is the accurate and automated classification of degenerative disorders. Most of current classification studies either do not mirror medical practice where patients may exhibit early stages of the disease, comorbidities, or atypical variants, or they are not able to produce probabilistic predictions nor account for uncertainty. Also, the spatial heterogeneity of the brain alterations caused by neurodegenerative processes is not usually considered, despite the spatial configuration of the neuronal loss is a characteristic hallmark for each disorder. In this article, we propose a classification technique that incorporates uncertainty and spatial information for distinguishing between healthy subjects and patients from four distinct neurodegenerative diseases: Alzheimer's disease, mild cognitive impairment, Parkinson's disease, and Multiple Sclerosis. We introduce a spatially informed Bayesian neural network (SBNN) that combines a three‐dimensional neural network to extract neurodegeneration features from MRI, Bayesian inference to account for uncertainty in diagnosis, and a spatially informed MRI image using hidden Markov random fields to encode cerebral spatial information. The SBNN model demonstrates that classification accuracy increases up to 25% by including a spatially informed MRI scan. Furthermore, the SBNN provides a robust probabilistic diagnosis that resembles clinical decision‐making and can account for the heterogeneous medical presentations of neurodegenerative disorders.

SiM Longitudinal Data Survival Analysis Diagnostics & Prediction · 发表于 2022-11-28

Gaussian variational approximate inference for joint models of longitudinal biomarkers and a survival outcome

Jieqi Tu, Jiehuan Sun

摘要

The shared random effects joint model is one of the most widely used approaches to study the associations between longitudinal biomarkers and a survival outcome and make dynamic risk predictions using the longitudinally measured biomarkers. Various types of joint models have been developed under different settings in the past decades. One major limitation of joint models is that they could be computationally expensive for complex models where the number of the shared random effects is large. Moreover, the inferential accuracy of joint models could also be diminished for complex models due to approximation errors. However, complex models are frequently needed in practice, for example, when the longitudinal biomarkers have nonlinear trajectories over time or the number of longitudinal biomarkers of interest is large. In this article, we propose a novel Gaussian variational approximate inference approach for fitting joint models, which significantly improves computational efficiency while maintaining inferential accuracy. We conduct extensive simulation studies to evaluate the performance of our proposed method and compare it to existing methods. The performance of our proposed method is further demonstrated on a dataset of patients with primary biliary cirrhosis.

JASA Bayesian Methods Spatial & Time Series High-Dimensional · 发表于 2022-11-28

Bayesian Spatial Blind Source Separation via the Thresholded Gaussian Process

Ben Wu, Ying Guo, Jian Kang

摘要

Blind source separation (BSS) aims to separate latent source signals from their mixtures. For spatially dependent signals in high dimensional and large-scale data, such as neuroimaging, most existing BSS methods do not take into account the spatial dependence and the sparsity of the latent source signals. To address these major limitations, we propose a Bayesian spatial blind source separation (BSP-BSS) approach for neuroimaging data analysis. We assume the expectation of the observed images as a linear mixture of multiple sparse and piece-wise smooth latent source signals, for which we construct a new class of Bayesian nonparametric prior models by thresholding Gaussian processes. We assign the vMF priors to mixing coefficients in the model. Under some regularity conditions, we show that the proposed method has several desirable theoretical properties including the large support for the priors, the consistency of joint posterior distribution of the latent source intensity functions and the mixing coefficients, and the selection consistency on the number of latent sources. We use extensive simulation studies and an analysis of the resting-state fMRI data in the Autism Brain Imaging Data Exchange (ABIDE) study to demonstrate that BSP-BSS outperforms the existing method for separating latent brain networks and detecting activated brain activation in the latent sources. *(abstract via PubMed)*

Biometrics Survival Analysis Nonparametric & Semiparametric · 发表于 2022-11-28

An Information Ratio-Based Goodness-of-Fit Test for Copula Models on Censored Data

Tao Sun, Yu Cheng, Ying Ding

摘要

Copula is a popular method for modeling the dependence among marginal distributions in multivariate censored data. As many copula models are available, it is essential to check if the chosen copula model fits the data well for analysis. Existing approaches to testing the fitness of copula models are mainly for complete or right-censored data. No formal goodness-of-fit (GOF) test exists for interval-censored or recurrent events data. We develop a general GOF test for copula-based survival models using the information ratio (IR) to address this research gap. It can be applied to any copula family with a parametric form, such as the frequently used Archimedean, Gaussian, and D-vine families. The test statistic is easy to calculate, and the test procedure is straightforward to implement. We establish the asymptotic properties of the test statistic. The simulation results show that the proposed test controls the type-I error well and achieves adequate power when the dependence strength is moderate to high. Finally, we apply our method to test various copula models in analyzing multiple real datasets. Our method consistently separates different copula models for all these datasets in terms of model fitness.

SiM Survival Analysis · 发表于 2022-11-27

Generalized mean residual life models for survival data with missing censoring indicators

Wenwen Li, Huijuan Ma, David Faraggi, Gregg E. Dinse

摘要

The mean residual life (MRL) function is an important and attractive alternative to the hazard function for characterizing the distribution of a time‐to‐event variable. In this article, we study the modeling and inference of a family of generalized MRL models for right‐censored survival data with censoring indicators missing at random. To estimate the model parameters, augmented inverse probability weighted estimating equation approaches are developed, in which the non‐missingness probability and the conditional probability of an uncensored observation are estimated by parametric methods or nonparametric kernel smoothing techniques. Asymptotic properties of the proposed estimators are established and finite sample performance is evaluated by extensive simulation studies. An application to brain cancer data is presented to illustrate the proposed methods.

Biometrics Clinical Trials Diagnostics & Prediction · 发表于 2022-11-27

Design Considerations for Two-Stage Enrichment Clinical Trials

Rosamarie Frieri, William Fisher Rosenberger, Nancy Flournoy, Zhantao Lin

摘要

When there is a predictive biomarker, enrichment can focus the clinical trial on a benefiting subpopulation. We describe a two-stage enrichment design, in which the first stage is designed to efficiently estimate a threshold and the second stage is a “phase III-like” trial on the enriched population. The goal of this paper is to explore design issues: sample size in Stages 1 and 2, and re-estimation of the Stage 2 sample size following Stage 1. By treating these as separate trials, we can gain insight into how the predictive nature of the biomarker specifically impacts the sample size. We also show that failure to adequately estimate the threshold can have disastrous consequences in the second stage. While any bivariate model could be used, we assume a continuous outcome and continuous biomarker, described by a bivariate normal model. The correlation coefficient between the outcome and biomarker is the key to understanding the behavior of the design, both for predictive and prognostic biomarkers. Through a series of simulations we illustrate the impact of model misspecification, consequences of poor threshold estimation, and requisite sample sizes that depend on the predictive nature of the biomarker. Such insight should be helpful in understanding and designing enrichment trials.

SiM Survival Analysis · 发表于 2022-11-25

RECeUS: Ratio estimation of censored uncured subjects, a different approach for assessing cure model appropriateness in studies with long‐term survivors

Subodh Selukar, Megan Othus

摘要

The need to model a cure fraction, the proportion of a cohort not susceptible to the event of interest, arises in a variety of contexts including tumor relapse in oncology. Existing methodology assumes that follow‐up is long enough for all uncured subjects to have experienced the event of interest at the time of analysis, and researchers have demonstrated that fitting cure models without sufficient follow‐up leads to bias. Few statistical methods exist to evaluate sufficient follow‐up, and they can exhibit poor performance and lead users to falsely conclude sufficient follow‐up, leading to bias, or to falsely claim insufficient follow‐up, possibly leading to additional, costly data collection. We propose a new quantitative statistic (RECeUS) to evaluate whether cure models may be appropriate to apply to censored data. Specifically, we propose that the estimated proportion of censored uncured subjects in a study can be used to evaluate cure model appropriateness. We evaluated the performance of RECeUS against existing methods via simulation and with two data examples, and we observe that RECeUS displays superior performance. In simulated and real‐world settings, RECeUS correctly identifies both situations in which data appear appropriate for cure modeling and when data seem inappropriate for fitting cure models.

SiM Clinical Trials · 发表于 2022-11-25

Practical implementation of the partial ordering continual reassessment method in a Phase I combination‐schedule dose‐finding trial

Pavel Mozgunov, Thomas Jaki, Ioannis Gounaris, Thomas Goddemeier, Anja Victor, Marianna Grinberg

摘要

There is a growing medical interest in combining several agents and optimizing their dosing schedules in a single trial in order to optimize the treatment for patients. Evaluating at doses of several drugs and their scheduling in a single Phase I trial simultaneously possess a number of statistical challenges, and specialized methods to tackle these have been proposed in the literature. However, the uptake of these methods is slow and implementation examples of such advanced methods are still sparse to date. In this work, we share our experience of proposing a model‐based partial ordering continual reassessment method (POCRM) design for three‐dimensional dose‐finding in an oncology trial. In the trial, doses of two agents and the dosing schedule of one of them can be escalated/de‐escalated. We provide a step‐by‐step summary on how the POCRM design was implemented and communicated to the trial team. We proposed an approach to specify toxicity orderings and their a‐priori probabilities, and developed a number of visualization tools to communicate the statistical properties of the design. The design evaluation included both a comprehensive simulation study and considerations of the individual trial behavior. The study is now enrolling patients. We hope that sharing our experience of the successful implementation of an advanced design in practice that went through evaluations of several health authorities will facilitate a better uptake of more efficient methods in practice.

SiM Bayesian Methods Nonparametric & Semiparametric · 发表于 2022-11-25

Bayesian additive regression trees for multivariate skewed responses

Seungha Um, Antonio R. Linero, Debajyoti Sinha, Dipankar Bandyopadhyay

摘要

This paper introduces a nonparametric regression approach for univariate and multivariate skewed responses using Bayesian additive regression trees (BART). Existing BART methods use ensembles of decision trees to model a mean function, and have become popular recently due to their high prediction accuracy and ease of use. The usual assumption of a univariate Gaussian error distribution, however, is restrictive in many biomedical applications. Motivated by an oral health study, we provide a useful extension of BART, theskewBARTmodel, to address this problem. We then extendskewBARTto allow for multivariate responses, with information shared across the decision trees associated with different responses within the same subject. The methodology accommodates within‐subject association, and allows varying skewness parameters for the varying multivariate responses. We illustrate the benefits of our multivariateskewBARTproposal over existing alternatives via simulation studies and application to the oral health dataset with bivariate highly skewed responses. Our methodology is implementable via theRpackageskewBART, available on GitHub.

SMMR Causal Inference · 发表于 2022-11-25

An overview of propensity score matching methods for clustered data

Benjamin Langworthy, Yujie Wu, Molin Wang

摘要

Propensity score matching is commonly used in observational studies to control for confounding and estimate the causal effects of a treatment or exposure. Frequently, in observational studies data are clustered, which adds to the complexity of using propensity score techniques. In this article, we give an overview of propensity score matching methods for clustered data, and highlight how propensity score matching can be used to account for not just measured confounders, but also unmeasured cluster level confounders. We also consider using machine learning methods such as generalized boosted models to estimate the propensity score and show that accounting for clustering when using these methods can greatly reduce the performance, particularly when there are a large number of clusters and a small number of subjects per cluster. In order to get around this we highlight scenarios where it may be possible to control for measured covariates using propensity score matching, while using fixed effects regression in the outcome model to control for cluster level covariates. Using simulation studies we compare the performance of different propensity score matching methods for clustered data across a number of different settings. Finally, as an illustrative example we apply propensity score matching methods for clustered data to study the causal effect of aspirin on hearing deterioration using data from the conservation of hearing study.

Biometrika Causal Inference · 发表于 2022-11-24

Assessing time-varying causal effect moderation in the presence of cluster-level treatment effect heterogeneity and interference

J Shi, Z Wu, W Dempsey

摘要

Summary: The micro-randomized trial is a sequential randomized experimental design to empirically evaluate the effectiveness of mobile health intervention components that may be delivered at hundreds or thousands of decision points. Micro-randomized trials have motivated a new class of causal estimands, termed causal excursion effects, for which semiparametric inference can be conducted via a weighted, centred least-squares criterion (Boruvka et al., 2018). Causal excursion effects allow health scientists to answer important scientific questions about how intervention effectiveness may change over time or may be moderated by individual characteristics, time-varying context or past responses. Existing definitions and associated methods assume between-subject independence and noninterference. Deviations from these assumptions often occur. In this paper, causal excursion effects are revisited under potential cluster-level treatment effect heterogeneity and interference, where the treatment effect of interest may depend on cluster-level moderators. Utility of the proposed methods is shown by analysing data from a multi-institution cohort of first-year medical residents in the United States.

Biometrics Clinical Trials Causal Inference · 发表于 2022-11-23

Assessing Exposure-Time Treatment Effect Heterogeneity in Stepped-Wedge Cluster Randomized Trials

Lara Maleyeff, Fan Li, Sebastien Haneuse, Rui Wang

摘要

A stepped-wedge cluster randomized trial (CRT) is a unidirectional crossover study in which timings of treatment initiation for clusters are randomized. Because the timing of treatment initiation is different for each cluster, an emerging question is whether the treatment effect depends on the exposure time, namely, the time duration since the initiation of treatment. Existing approaches for assessing exposure-time treatment effect heterogeneity either assume a parametric functional form of exposure time or model the exposure time as a categorical variable, in which case the number of parameters increases with the number of exposure-time periods, leading to a potential loss in efficiency. In this article, we propose a new model formulation for assessing treatment effect heterogeneity over exposure time. Rather than a categorical term for each level of exposure time, the proposed model includes a random effect to represent varying treatment effects by exposure time. This allows for pooling information across exposure-time periods and may result in more precise average and exposure-time-specific treatment effect estimates. In addition, we develop an accompanying permutation test for the variance component of the heterogeneous treatment effect parameters. We conduct simulation studies to compare the proposed model and permutation test to alternative methods to elucidate their finite-sample operating characteristics, and to generate practical guidance on model choices for assessing exposure-time treatment effect heterogeneity in stepped-wedge CRTs.

SiM Clinical Trials · 发表于 2022-11-23

Adjusting for treatment selection in phase II/III clinical trials with time to event data

Josephine N. Khan, Peter K. Kimani, Ekkehard Glimm, Nigel Stallard

摘要

Phase II/III clinical trials are efficient two‐stage designs that test multiple experimental treatments. In stage 1, patients are allocated to the control and all experimental treatments, with the data collected from them used to select experimental treatments to continue to stage 2. Patients recruited in stage 2 are allocated to the selected treatments and the control. Combined data of stage 1 and stage 2 are used for a confirmatory phase III analysis. Appropriate analysis needs to adjust for selection bias of the stage 1 data. Point estimators exist for normally distributed outcome data. Extending these estimators to time to event data is not straightforward because treatment selection is based on correlated treatment effects and stage 1 patients who do not get events in stage 1 are followed‐up in stage 2. We have derived an approximately uniformly minimum variance conditional unbiased estimator (UMVCUE) and compared its biases and mean squared errors to existing bias adjusted estimators. In simulations, one existing bias adjusted estimator has similar properties as the practically unbiased UMVCUE while the others can have noticeable biases but they are less variable than the UMVCUE. For confirmatory phase II/III clinical trials where unbiased estimators are desired, we recommend the UMVCUE or the existing estimator with which it has similar properties.

SiM Meta-Analysis & Data Integration · 发表于 2022-11-22

Two‐sample test withg‐modeling and its applications

Jingyi Zhai, Hui Jiang

摘要

Many real data analyses involve two‐sample comparisons in location or in distribution. Most existing methods focus on problems where observations are independently and identically distributed in each group. However, in some applications the observed data are not identically distributed but associated with some unobserved parameters which are identically distributed. To address this challenge, we propose a novel two‐sample testing procedure as a combination of the ‐modeling density estimation introduced by Efron and the two‐sample Kolmogorov‐Smirnov test. We also propose efficient bootstrap algorithms to estimate the statistical significance for such tests. We demonstrate the utility of the proposed approach with two biostatistical applications: the analysis of surgical nodes data with binomial model and differential expression analysis of single‐cell RNA sequencing data with zero‐inflated Poisson model.

SMMR Meta-Analysis & Data Integration Testing & Uncertainty · 发表于 2022-11-22

Standard error estimation in meta-analysis of studies reporting medians

Sean McGrath, Stephan Katzenschlager, Alexandra J Zimmer, Alexander Seitel, Russell Steele, Andrea Benedetti

摘要

We consider the setting of an aggregate data meta-analysis of a continuous outcome of interest. When the distribution of the outcome is skewed, it is often the case that some primary studies report the sample mean and standard deviation of the outcome and other studies report the sample median along with the first and third quartiles and/or minimum and maximum values. To perform meta-analysis in this context, a number of approaches have recently been developed to impute the sample mean and standard deviation from studies reporting medians. Then, standard meta-analytic approaches with inverse-variance weighting are applied based on the (imputed) study-specific sample means and standard deviations. In this article, we illustrate how this common practice can severely underestimate the within-study standard errors, which results in poor coverage for the pooled mean in common effect meta-analyses and overestimation of between-study heterogeneity in random effects meta-analyses. We propose a straightforward bootstrap approach to estimate the standard errors of the imputed sample means. Our simulation study illustrates how the proposed approach can improve the estimation of the within-study standard errors and consequently improve coverage for the pooled mean in common effect meta-analyses and estimation of between-study heterogeneity in random effects meta-analyses. Moreover, we apply the proposed approach in a meta-analysis to identify risk factors of a severe course of COVID-19.

SMMR Survival Analysis Latent Variables & Mixtures Clinical Trials · 发表于 2022-11-22

Simulating time-to-event data subject to competing risks and clustering: A review and synthesis

Can Meng, Denise Esserman, Fan Li, Yize Zhao, Ondrej Blaha, Wenhan Lu, Yuxuan Wang, Peter Peduzzi, Erich J. Greene

摘要

Simulation studies play an important role in evaluating the performance of statistical models developed for analyzing complex survival data such as those with competing risks and clustering. This article aims to provide researchers with a basic understanding of competing risks data generation, techniques for inducing cluster-level correlation, and ways to combine them together in simulation studies, in the context of randomized clinical trials with a binary exposure or treatment. We review data generation with competing and semi-competing risks and three approaches of inducing cluster-level correlation for time-to-event data: the frailty model framework, the probability transform, and Moran’s algorithm. Using exponentially distributed event times as an example, we discuss how to introduce cluster-level correlation into generating complex survival outcomes, and illustrate multiple ways of combining these methods to simulate clustered, competing and semi-competing risks data with pre-specified correlation values or degree of clustering.

SiM Survival Analysis · 发表于 2022-11-22

Measures of explained variation under the mixture cure model for survival data

Yingwei Peng, Yuyao Wang, Ronghui Xu

摘要

Explained variation is well understood under linear regression models and has been extended to models for survival data. In this article, we consider the mixture cure models. We propose two approaches to define explained variation under the mixture cure models, one based on the Kullback‐Leibler information gain and the other based on residual sum of squares. We show that the proposed measures have desired properties as measures of explained variation, similar to those under other regression models. A simulation study is conducted to demonstrate the properties of the proposed measures. They are also applied to real data analyses to illustrate the use of explained variation.

Biometrika Optimization & Learning Theory Bayesian Methods · 发表于 2022-11-22

Existence of matching priors on compact spaces

Haosui Duanmu, Daniel M Roy, Aaron Smith

摘要

Summary: A matching prior at level $1-\alpha$ is a prior such that an associated $1-\alpha$ credible region is also a $1-\alpha$ confidence set. We study the existence of matching priors for general families of credible regions. Our main result gives topological conditions under which matching priors for specific families of credible regions exist. Informally, we prove that, on compact parameter spaces, a matching prior exists if the so-called rejection-probability function is jointly continuous when we adopt the Wasserstein metric on priors. In light of this general result, we observe that typical families of credible regions, such as credible balls, highest-posterior density regions, quantiles, etc., fail to meet this topological condition. We show how to design approximate posterior credible balls and highest-posterior density regions that meet these topological conditions, yielding matching priors. Finally, we evaluate a numerical scheme for computing approximately matching priors based on discretization and iteration. The proof of our main theorem uses tools from nonstandard analysis and establishes new results about the nonstandard extension of the Wasserstein metric that may be of independent interest.

Biometrics Causal Inference Nonparametric & Semiparametric · 发表于 2022-11-22

Efficient Targeted Learning of Heterogeneous Treatment Effects for Multiple Subgroups

Waverly Wei, Maya Petersen, Mark J van der Laan, Zeyu Zheng, Chong Wu, Jingshen Wang

摘要

In biomedical science, analyzing treatment effect heterogeneity plays an essential role in assisting personalized medicine. The main goals of analyzing treatment effect heterogeneity include estimating treatment effects in clinically relevant subgroups and predicting whether a patient subpopulation might benefit from a particular treatment. Conventional approaches often evaluate the subgroup treatment effects via parametric modeling and can thus be susceptible to model mis-specifications. In this paper, we take a model-free semiparametric perspective and aim to efficiently evaluate the heterogeneous treatment effects of multiple subgroups simultaneously under the one-step targeted maximum-likelihood estimation (TMLE) framework. When the number of subgroups is large, we further expand this path of research by looking at a variation of the one-step TMLE that is robust to the presence of small estimated propensity scores in finite samples. From our simulations, our method demonstrates substantial finite sample improvements compared to conventional methods. In a case study, our method unveils the potential treatment effect heterogeneity of rs12916-T allele (a proxy for statin usage) in decreasing Alzheimer's disease risk.

JRSSB Spatial & Time Series Causal Inference · 发表于 2022-11-22

Causal Inference with Spatio-Temporal Data: Estimating the Effects of Airstrikes on Insurgent Violence in Iraq

Georgia Papadogeorgou, Kosuke Imai, Jason Lyall, Fan Li

摘要

Many causal processes have spatial and temporal dimensions. Yet the classic causal inference framework is not directly applicable when the treatment and outcome variables are generated by spatio-temporal point processes. We extend the potential outcomes framework to these settings by formulating the treatment point process as a stochastic intervention. Our causal estimands include the expected number of outcome events in a specified area under a particular stochastic treatment assignment strategy. Our methodology allows for arbitrary patterns of spatial spillover and temporal carryover effects. Using martingale theory, we show that the proposed estimator is consistent and asymptotically normal as the number of time periods increases. We propose a sensitivity analysis for the possible existence of unmeasured confounders, and extend it to the Hájek estimator. Simulation studies are conducted to examine the estimators' finite sample performance. Finally, we illustrate the proposed methods by estimating the effects of American airstrikes on insurgent violence in Iraq from February 2007 to July 2008. Our analysis suggests that increasing the average number of daily airstrikes for up to 1 month may result in more insurgent attacks. We also find some evidence that airstrikes can displace attacks from Baghdad to new locations up to 400 km away.

JRSSB Latent Variables & Mixtures Causal Inference · 发表于 2022-11-22

A Statistical Test to Reject the Structural interpretation of a Latent Factor Model

Tyler J. VanderWeele, Stijn Vansteelandt

摘要

Factor analysis is often used to assess whether a single univariate latent variable is sufficient to explain most of the covariance among a set of indicators for some underlying construct. When evidence suggests that a single factor is adequate, research often proceeds by using a univariate summary of the indicators in subsequent research. Implicit in such practices is the assumption that it is the underlying latent, rather than the indicators, that is causally efficacious. The assumption that the indicators do not have effects on anything subsequent, and that they are themselves only affected by antecedents through the underlying latent is a strong assumption, effectively imposing a structural interpretation on the latent factor model. In this paper, we show that this structural assumption has empirically testable implications, even though the latent variable itself is unobserved. We develop a statistical test to potentially reject the structural interpretation of a latent factor model. We apply this test to data concerning associations between the Satisfaction with Life Scale and subsequent all-cause mortality, which provides strong evidence against a structural interpretation for a univariate latent underlying the scale. Discussion is given to the implications of this result for the development, evaluation and use of measures, and for the use of factor analysis itself.

Biometrika Testing & Uncertainty · 发表于 2022-11-21

Gaussian universal likelihood ratio testing

Robin Dunn, Aaditya Ramdas, Sivaraman Balakrishnan, Larry Wasserman

摘要

Summary: The classical likelihood ratio test based on the asymptotic chi-squared distribution of the log-likelihood is one of the fundamental tools of statistical inference. A recent universal likelihood ratio test approach based on sample splitting provides valid hypothesis tests and confidence sets in any setting for which we can compute the split likelihood ratio statistic, or, more generally, an upper bound on the null maximum likelihood. The universal likelihood ratio test is valid in finite samples and without regularity conditions. This test empowers statisticians to construct tests in settings for which no valid hypothesis test previously existed. For the simple, but fundamental, case of testing the population mean of $d$-dimensional Gaussian data with an identity covariance matrix, the classical likelihood ratio test itself applies. Thus, this setting serves as a perfect test bed to compare the classical likelihood ratio test against the universal likelihood ratio test. This work presents the first in-depth exploration of the size, power and relationships between several universal likelihood ratio test variants. We show that a repeated subsampling approach is the best choice in terms of size and power. For large numbers of subsamples, the repeated subsampling set is approximately spherical. We observe reasonable performance even in a high-dimensional setting, where the expected squared radius of the best universal likelihood ratio test’s confidence set is approximately 3/2 times the squared radius of the classical likelihood ratio test’s spherical confidence set. We illustrate the benefits of the universal likelihood ratio test through testing a nonconvex doughnut-shaped null hypothesis, where a universal inference procedure can have higher power than a standard approach.

SiM Causal Inference · 发表于 2022-11-21

Double robust estimation of optimal partially adaptive treatment strategies: An application to breast cancer treatment using hormonal therapy

Denis Talbot, Erica E.M. Moodie, Caroline Diorio

摘要

Precision medicine aims to tailor treatment decisions according to patients' characteristics. G‐estimation and dynamic weighted ordinary least squares are double robust methods to identify optimal adaptive treatment strategies. It is underappreciated that they require modeling all existing treatment‐confounder interactions to be consistent. Identifying optimal partially adaptive treatment strategies that tailor treatments according to only a few covariates, ignoring some interactions, may be preferable in practice. Building on G‐estimation and dWOLS, we propose estimators of such partially adaptive strategies and demonstrate their double robustness. We investigate these estimators in a simulation study. Using data maintained by the Centre des Maladies du Sein, we estimate a partially adaptive treatment strategy for tailoring hormonal therapy use in breast cancer patients. R software implementing our estimators is provided.

JRSSB Missing Data & Measurement Error High-Dimensional · 发表于 2022-11-20

High-Dimensional Principal Component Analysis with Heterogeneous Missingness

Ziwei Zhu, Tengyao Wang, Richard J. Samworth

摘要

We study the problem of high-dimensional Principal Component Analysis (PCA) with missing observations. In a simple, homogeneous observation model, we show that an existing observed-proportion weighted (OPW) estimator of the leading principal components can (nearly) attain the minimax optimal rate of convergence, which exhibits an interesting phase transition. However, deeper investigation reveals that, particularly in more realistic settings where the observation probabilities are heterogeneous, the empirical performance of the OPW estimator can be unsatisfactory; moreover, in the noiseless case, it fails to provide exact recovery of the principal components. Our main contribution, then, is to introduce a new method, which we call primePCA, that is designed to cope with situations where observations may be missing in a heterogeneous manner. Starting from the OPW estimator, primePCA iteratively projects the observed entries of the data matrix onto the column space of our current estimate to impute the missing entries, and then updates our estimate by computing the leading right singular space of the imputed data matrix. We prove that the error of primePCA converges to zero at a geometric rate in the noiseless case, and when the signal strength is not too small. An important feature of our theoretical guarantees is that they depend on average, as opposed to worst-case, properties of the missingness mechanism. Our numerical studies on both simulated and real data reveal that primePCA exhibits very encouraging performance across a wide range of scenarios, including settings where the data are not Missing Completely At Random.

SiM Survival Analysis · 发表于 2022-11-20

Comparing and combining data from immune assays based on left‐censored multivariate normal model assuming common assay differences across settings

Ying Huang, Yunda Huang

摘要

In vaccine research towards the prevention of infectious diseases, immune response biomarkers serve as an important tool for comparing and ranking vaccine candidates based on their immunogenicity and predicted protective effect. However, analyses of immune response outcomes can be complicated by differences across assays when immune response data are acquired from multiple groups/laboratories. Motivated by a real‐world problem to accommodate the use of two different neutralization assays in COVID‐19 vaccine trials, we propose methods based on left‐censored multivariate normal model assuming common assay differences across settings, to adjust for differences between assays with respect to measurement error and the lower limit of detection. Our proposed methods integrate external paired‐sample data with bridging assumptions to achieve two objectives, both using pooled data acquired from different assays: (i) comparing immunogenicity between vaccine regimens, and (ii) evaluating correlates of risk. In simulation studies, for the first objective, our method leads to unbiased calibrated assay mean with good coverage of bootstrap confidence interval, as well as valid test for immunogenicity comparison, while the alternative method assuming constant calibration model between assays leads to biased estimate of assay mean with undercoverage problem and invalid test with inflated type‐I error; for the second objective, in the presence of noticeable left‐censoring rate, our proposed method can drastically outperform the existing method that ignores left‐censoring, in terms of reduced bias and improved precision. We apply the proposed methods to SARS‐CoV‐2 spike‐pseudotyped virus neutralization assay data generated in vaccine and convalescent samples by two different laboratories.

Biometrics Bayesian Methods · 发表于 2022-11-19

Hospital Profiling Using Bayesian Decision Theory

Johannes Hengelbrock, Johannes Rauh, Jona Cederbaum, Maximilian Kähler, Michael Höhle

摘要

For evaluating the quality of care provided by hospitals, special interest lies in the identification of performance outliers. The classification of healthcare providers as outliers or non-outliers is a decision under uncertainty, because the true quality is unknown and can only be inferred from an observed result of a quality indicator. We propose to embed the classification of healthcare providers into a Bayesian decision theoretical framework that enables the derivation of optimal decision rules with respect to the expected decision consequences. We propose paradigmatic utility functions for two typical purposes of hospital profiling: the external reporting of healthcare quality and the initiation of change in care delivery. We make use of funnel plots to illustrate and compare the resulting optimal decision rules and argue that sensitivity and specificity of the resulting decision rules should be analyzed. We then apply the proposed methodology to the area of hip replacement surgeries by analyzing data from 1,277 hospitals in Germany which performed over 180,000 such procedures in 2017. Our setting illustrates that the classification of outliers can be highly dependent upon the underlying utilities. We conclude that analyzing the classification of hospitals as a decision theoretic problem helps to derive transparent and justifiable decision rules. The methodology for classifying quality indicator results is implemented in an R package (iqtigbdt) and is available on GitHub.

Biometrika Networks & Graphs Causal Inference Nonparametric & Semiparametric · 发表于 2022-11-18

Variable elimination, graph reduction and the efficient g-formula

F Richard Guo, Emilija Perković, Andrea Rotnitzky

摘要

Summary: We study efficient estimation of an interventional mean associated with a point exposure treatment under a causal graphical model represented by a directed acyclic graph without hidden variables. Under such a model, a subset of the variables may be uninformative, in that failure to measure them neither precludes identification of the interventional mean nor changes the semiparametric variance bound for regular estimators of it. We develop a set of graphical criteria that are sound and complete for eliminating all the uninformative variables, so that the cost of measuring them can be saved without sacrificing estimation efficiency, which could be useful when designing a planned observational or randomized study. Further, we construct a reduced directed acyclic graph on the set of informative variables only. We show that the interventional mean is identified from the marginal law by the g-formula (Robins, 1986) associated with the reduced graph, and the semiparametric variance bounds for estimating the interventional mean under the original and the reduced graphical model agree. The g-formula is an irreducible, efficient identifying formula in the sense that the nonparametric estimator of the formula, under regularity conditions, is asymptotically efficient under the original causal graphical model, and no formula with this property exists that depends only on a strict subset of the variables.

JRSSB Networks & Graphs · 发表于 2022-11-18

Structure Learning for Extremal Tree Models

Sebastian Engelke, Stanislav Volgushev

摘要

Extremal graphical models are sparse statistical models for multivariate extreme events. The underlying graph encodes conditional independencies and enables a visual interpretation of the complex extremal dependence structure. For the important case of tree models, we develop a data-driven methodology for learning the graphical structure. We show that sample versions of the extremal correlation and a new summary statistic, which we call the extremal variogram, can be used as weights for a minimum spanning tree to consistently recover the true underlying tree. Remarkably, this implies that extremal tree models can be learned in a completely non-parametric fashion by using simple summary statistics and without the need to assume discrete distributions, existence of densities or parametric models for bivariate distributions.

Biometrika High-Dimensional Testing & Uncertainty · 发表于 2022-11-17

Testing Kronecker product covariance matrices for high-dimensional matrix-variate data

Long Yu, Jiahui Xie, Wang Zhou

摘要

Summary: The Kronecker product covariance structure provides an efficient way to model the inter-correlations of matrix-variate data. In this paper, we propose test statistics for the Kronecker product covariance matrix based on linear spectral statistics of renormalized sample covariance matrices. A central limit theorem is proved for the linear spectral statistics, with explicit formulas for the mean and covariance functions, thereby filling a gap in the literature. We then show theoretically that the proposed test statistics have well-controlled size and high power. We further propose a bootstrap resampling algorithm to approximate the limiting distributions of the associated linear spectral statistics. Consistency of the bootstrap procedure is guaranteed under mild conditions. The proposed test procedure is also applicable to the Kronecker product covariance model with additional random noise. In our simulations, the empirical sizes of the proposed test procedure and its bootstrapped version are close to the corresponding theoretical values, while the power converges to $1$ quickly as the dimension and sample size increase.

SiM Clinical Trials · 发表于 2022-11-17

Target estimands for population‐adjusted indirect comparisons

Antonio Remiro‐Azócar

摘要

The author declares no potential conflict of interests. Data sharing is not applicable to this article because no new data were analyzed. *(abstract via OpenAlex)*

SiM Clinical Trials Meta-Analysis & Data Integration Econometrics & Economics · 发表于 2022-11-17

Some considerations on target estimands for health technology assessment

Antonio Remiro‐Azócar

摘要

First and foremost, I would like to thank an anonymous Associate Editor and Prof. Nigel Stallard for arranging a fascinating discussion around my article, "Target estimands for population-adjusted indirect comparisons."1 This is based on a prior exchange with Phillippo et al.2-4 I extend my gratitude to Russek-Cohen,5 Schiel,6 Senn,7 Spieker,8 and Van Lancker et al,9 for their additional contributions. This rejoinder discusses the potential development of an estimands framework in the context of evidence synthesis and health technology assessment (HTA). I consider the following base-case scenario. An evidence synthesis (eg, an indirect treatment comparison or a network meta-analysis) is required for HTA. The evidence synthesis combines the results of multiple randomized controlled trials (RCTs).10-12 Each RCT has been designed for regulatory approval in the premarketing authorization setting, and has target estimands of its own. There are exceptions to this scenario; regulatory decisions are not exclusively based on RCTs and HTA decisions are not exclusively based on evidence synthesis. Nevertheless, it is a common scenario. Two challenges arise. First, different policy decisions will require different research questions; the estimands targeted for regulatory approval will likely not align with those targeted for HTA. Second, within the evidence synthesis, there may be misalignment between the estimands targeted by individual studies. Finally, I reassert that marginal estimands should be preferred in the context of the original discussion.2-4 For RCTs in the regulatory setting, the use of estimands has been stimulated by the recent publication of the E9 (R1) Addendum, issued by the International Council of Harmonisation (ICH) and recognized by agencies such as the United States Food and Drug Administration.13 According to the ICH guidance, an estimand should be defined on the basis of five attributes: (1) treatment(s); (2) target population; (3) clinical outcome of interest; (4) population-level summary effect measure; and (5) strategy for intercurrent (post-randomization) events. My original commentary1 places focus on the second and fourth elements. It does not intend to cover other dimensions, which are also important. The application of the ICH guidance to individual RCTs has been addressed in depth elsewhere.14, 15 In evidence synthesis, and more generally in HTA reporting, the term estimand is rarely used. This does not imply a lack of interest in clarifying the research question prior to the analysis. In HTA, the PICO framework is typically used to translate policy questions into research questions. PICO consists of four components: (1) population; (2) intervention; (3) comparator(s); and (4) outcome. To obtain reimbursement for a given treatment, the manufacturer must submit an evidence dossier that addresses the research question(s) included in the scope. Current best-practice guidelines recommend prespecifying the relevant PICO question(s) in the HTA scoping process.16-18 In evidence synthesis, PICO questions are often formulated prior to the analysis in order to guide the data extraction required for systematic literature reviews.19 The PICO framework is a valuable tool. Nevertheless, it needs refinement to more precisely characterize the inferential target in HTA; that is, to map the HTA research question into a formal estimand. The current framework does not discuss the population-level measure used to summarize the treatment effect. It also lacks a strategy for intercurrent events, which as stated by Van Lancker et al,9 "are at the heart of the ICH E9 (R1) addendum." I would propose a PICOSI framework for describing the HTA research question: (1) population; (2) intervention; (3) comparator(s); (4) outcome; (5) summary effect measure; and (6) intercurrent events. This would make the summary measure used to compare treatments and the intercurrent event strategy explicit. It would also allow stakeholders to translate regulatory language to HTA terminology. My original commentary1 and prior exchange with Phillippo et al2-4 highlight the importance of prespecifying the summary effect measure. It is necessary to make a distinction between marginal and conditional effect measures. I argue that the former are more relevant to the HTA setting.1, 3 Another question that arises is whether relative estimands (treatment effects) or absolute estimands (mean outcomes) are more appropriate. When the goal is to estimate comparative effectiveness, relative estimands are the logical choice. Nevertheless, when the aim is to estimate comparative cost-effectiveness, absolute measures (eg, event probabilities) are often input to health economic models.4 The transportability of relative and absolute measures between populations depends on different conditions.20 Another important matter is that the evaluation of effect modifier status, necessary to assess generalizability, transportability and external validity, is tied to the scale used to measure the relative effect.21 Effect modifiers cannot be selected prior to the specification of a summary effect measure of interest. Fundamentally, the estimands required for regulatory and HTA decision-making are expected to diverge. Regulatory decisions are made prior to HTA decisions. Individual studies will likely be completed at the time of undertaking the evidence synthesis, and estimand definitions for the former will precede estimand definitions for the latter. Moreover, regulatory and HTA decision-makers have different perspectives. The HTA paradigm places emphasis on evaluating "real-world" effectiveness, whereas regulators typically prioritize internal over external validity when assessing clinical efficacy. In terms of intercurrent events, estimands based on the "treatment policy" and the "composite" strategies seem more relevant for HTA purposes. Nevertheless, regulators may prefer other strategies in certain scenarios. In terms of comparator(s), HTA demands comparing the intervention under study versus all alternative treatments for a specific disease state. Conversely, regulators typically require a head-to-head comparison against placebo or standard of care. In addition, there is a multiplicity of decision-makers within HTA. Different agencies will target different populations and, therefore, different estimands. HTA policy decisions typically consider external validity, but external validity depends on the population that is targeted by the payer. Given the large number of regulatory bodies and national and regional HTA agencies, I coincide with Schiel: we should avoid focusing exclusively on the viewpoint of a single stakeholder when designing studies.6 Traditionally, HTA objectives have played a limited role in the design of RCTs. Potential HTA research questions should be posed in the early stages of drug development, so that these can be answered at the time of HTA decision-making.22, 23 Apart from mismatches in estimands between decision-making contexts, there may also be inconsistencies between the trial-specific estimands combined in an evidence synthesis. As expressed by Russek-Cohen, "individual studies were not planned with similar estimands nor they were necessarily planned in anticipation of a meta-analysis."5 There may be mismatches in endpoint definitions, treatment implementations, study target populations, summary effect measures or intercurrent event strategies, that invalidate the results of an evidence synthesis. The availability of standardized outcome and treatment definitions depends on the clinical indication and the interventions being evaluated. Differences in endpoints may arise if outcomes are measured using different methods, at different times, or if the trials have different duration. When a treatment is assessed in multiple studies, there may be dissimilarities between the versions of the intervention being administered, for example, in dosing formulation, delivery mechanism, co-treatment regimens, or concomitant background medications. Clinical input is critical to assess the potential impact of these sources of misalignment on the conclusions of an evidence synthesis. The studies being synthesized may target populations that differ in terms of their baseline characteristics, due to different inclusion/exclusion criteria. Even if inclusion/exclusion criteria are identical, the covariate distribution of the sampled patients will likely vary from study to study. These differences may give rise to heterogeneous treatment effect sizes across studies, which threaten the validity of the evidence synthesis. Random effects models have traditionally been used to account for effect heterogeneity.24-26 Nevertheless, the conventional approaches do not explicitly produce estimates that are directly interpretable in any particular population.27-29 This is problematic, given the central role of the population in the estimands framework, and in formulating research questions in evidence synthesis and HTA. Here lies the value of covariate-adjusted indirect comparisons20, 30, 31 and network meta-regression approaches,32, 33 such as those examined by Phillippo et al2 in their original simulation study. These methods explicitly account for differences in patient characteristics between studies, thereby targeting estimands in specific samples or populations. As a result, they are being increasingly considered by payers.34 In many practical scenarios, patient-level data will be unavailable for certain trials, and the evidence synthesis will be based on published aggregate-level data to some degree. Therefore, in the words of Russek-Cohen, "there needs to be an incentive for summarizing individual trials so they can be more easily studied in meta-analyses that utilize covariates."5 Conflicts in the summary effect measures that are pooled in an evidence synthesis are also a cause for concern. When the effect measure is non-collapsible, combining marginal and conditional estimates will produce bias.12 Non-collapsible effect measures such as (log) odds ratios and (log) hazard ratios are commonly used in network meta-analysis.35, 36 Finally, there is a need to align within-study intercurrent event strategies prior to an evidence synthesis. To conclude, I wish to examine some of the considerations made by Senn7 and Van Lancker et al9 about conditional and marginal estimands. Senn7 argues that "we usually treat individuals not populations." I am in agreement. That being so, reimbursement decisions in HTA are typically made on populations not single patients. At the time of decision-making, the target population for the therapeutic indication is "stable" and well-defined, having been prespecified in the PICO question on the scope of the HTA. The target population of interest is not necessarily broad. Often, it is highly selected and takes into account specific patient characteristics. For instance, the final scope of a health technology appraisal for abemaciclib (TA810), recently published by the National Institute for Health and Care Excellence, describes the target population as "adults with hormone receptor-positive, HER2-negative, node-positive early breast cancer after definitive surgery of the primary breast tumor at high-risk of recurrence."37 A preference for marginal estimands does not necessarily imply that an "all or nothing" decision is required or that "homogeneity" is assumed. When subsets of the target population are considered to benefit differently from the reimbursement of a drug, relevant sub-populations are defined during the scoping process, in which case separate PICO questions can be formulated for each. In this case, the covariates used to define the subsets are binary or categorical, and the distinction between marginal and conditional estimands is blurred. A conditional estimand at the population level can be viewed as a marginal estimand at the sub-population level. Nevertheless, the specific context—covariate-adjusted indirect comparisons in HTA—is central to the debate. Consider the situation described by Phillippo et al in their original simulation study:2 (1) covariates are continuous; (2) model-based covariate adjustment is required for transportability; (3) valid estimation relies on correct model specification. As suggested by Spieker,8 a "modeling stage" is necessary to select the covariate adjustment model. When the adjustment model is an outcome regression, analysts will seek goodness-of-fit to the observed data. In this setting, the conditional estimand would align to the selection of the estimator. However, the estimands framework demands that the estimand is articulated first, prior to any estimation. Here, the definition of a conditional estimand would depend on statistical modeling assumptions. Conversely, the marginal estimand can be unambiguously interpreted without reference to a particular adjustment model. I believe that this point is implicit throughout Van Lancker et al's discussion.9 Ultimately, the marginal estimand is more compatible with the estimands framework, because it is the inferential target that allows one to separate the research question from the method of statistical analysis. *(abstract via OpenAlex)*

JASA High-Dimensional · 发表于 2022-11-17

Multi-Task Learning with High-Dimensional Noisy Images

Xin Ma, Suprateek Kundu, for the Alzheimer’s Disease Neuroimaging Initiative

摘要

Recent medical imaging studies have given rise to distinct but inter-related datasets corresponding to multiple experimental tasks or longitudinal visits. Standard scalar-on-image regression models that fit each dataset separately are not equipped to leverage information across inter-related images, and existing multi-task learning approaches are compromised by the inability to account for the noise that is often observed in images. We propose a novel joint scalar-on-image regression framework involving wavelet-based image representations with grouped penalties that are designed to pool information across inter-related images for joint learning, and which explicitly accounts for noise in high-dimensional images via a projection-based approach. In the presence of non-convexity arising due to noisy images, we derive non-asymptotic error bounds under non-convex as well as convex grouped penalties, even when the number of voxels increases exponentially with sample size. A projected gradient descent algorithm is used for computation, which is shown to approximate the optimal solution via well-defined non-asymptotic optimization error bounds under noisy images. Extensive simulations and application to a motivating longitudinal Alzheimer's disease study illustrate significantly improved predictive ability and greater power to detect true signals, that are simply missed by existing methods without noise correction due to the attenuation to null phenomenon. *(abstract via PubMed)*

Biometrics Testing & Uncertainty · 发表于 2022-11-17

Identifying Alert Concentrations Using a Model-Based Bootstrap Approach

Kathrin Möllenhoff, Kirsten Schorning, Franziska Kappenberg

摘要

The determination of alert concentrations, where a pre-specified threshold of the response variable is exceeded, is an important goal of concentration–response studies. The traditional approach is based on investigating the measured concentrations and attaining statistical significance of the alert concentration by using a multiple t-test procedure. In this paper, we propose a new model-based method to identify alert concentrations, based on fitting a concentration–response curve and constructing a simultaneous confidence band for the difference of the response of a concentration compared to the control. In order to obtain these confidence bands, we use a bootstrap approach which can be applied to any functional form of the concentration–response curve. This particularly offers the possibility to investigate also those situations where the concentration–response relationship is not monotone and, moreover, to detect alerts at concentrations which were not measured during the study, providing a highly flexible framework for the problem at hand.

SiM Other · 发表于 2022-11-17

Explicit underdose control based on toxicity: Four points to consider

Paul H. Frankel, Elizabeth Garrett‐Mayer, Mark D. Krailo

摘要

There are many design options for toxicity-driven phase 1 dose-finding in oncology drug development. There are methods that are simple, traditional and transparent, such as the 3 + 31 or rolling 6 design2 that do not target a dose limiting toxicity (DLT)-rate and are focused on exploring escalating doses subject to limits on patient risk. On the other end of the spectrum are designs that focus on a DLT target-rate and require sophisticated tools throughout the conduct of the study, are less transparent, and require statistical support. Two examples are the CRM design3, 4 and the BLRM design.5 There are also designs between these two extremes. Closest to the 3 + 3 and rolling 6 are queue-based variations of those designs6 or other A + B designs7 that also explore doses subject to pre-specified rules to limit patient risk. Closer to the CRM or BLRM are statistically motivated methods such as the mTPI,8 TEQR,9 and BOIN,10 where there is a target DLT rate and/or range, but simple decision rules and tables reduce the absolute requirement for sophisticated tools or a statistician except in the design and final analysis. Entering into this challenging area is the current effort11 to provide a new twist to the BLRM. The key change is to augment the BLRM with an underdose control rule to be pitted against overdose control. The popularity of this method will likely be limited, as this method is being introduced exactly at the time when the “more is better” paradigm is less accepted.12 Project Optimus, led by the FDA, is specifically concerned with the assumption that the maximum tolerated dose provides the best toxicity-efficacy trade-off, leading to a focus on selecting doses with relatively high toxicity. Instead, low toxicity is not generally an issue to be used to eliminate certain doses for possible consideration for future study. Nevertheless, there are increasingly rare, but important, clinical settings where underdose control concerns persists and this is exactly where such a tool can be considered. Primarily, these are short-term therapies with curative intent. Once restricted to such a setting, is this BLRM with both overdose control and underdose control superior to other designs that seek to find a DLT rate within a pre-specified range? The authors make the reasonable case (see Table 7 of Reference 11) that in some settings, depending on the details of the dose-toxicity curve and the dose increment, this method may have some advantages over some alternative DLT-target rate designs. In other settings, it may not. Knowing when and how to use a new tool is critical for statisticians in the field, and to that end, we will consider 4 separate points. The first point is the concern expressed in Table 1 (see Reference 11), where 0/3 DLTs are observed at each dose level of 10 mg, 25 mg, 50 mg and 100 mg, and yet 200 mg exceeds the overdose control rule so the next patients are not treated at that dose. At first view, this might appear to convey a problem since no escalation is permitted with 0 DLTs on multiple dose levels. However, it is very uncommon in a traditional 3 + 3 design to see dose doubling between a fourth and fifth dose level. The BLRM with overdose control may be pointing to this concern in the design when encountering the 100 mg increment in this case. The authors did not present the full BLRM model behind this hypothetical data scenario, but would a lesser increment, such as a 40 mg increment, representing a 40% increase, have been allowed? This is not to detract from the main role of underdose control, but simply to highlight that the concern about the conservative nature of the BLRM design with overdose control may have been overstated and this may influence the choice of method. The second point to highlight relates to the details on how to use this modified BLRM method. The BLRM method is usually written as a guide while providing certain constraints on decision-making. It presents a permissible range, and a guide within that range, and otherwise, the physicians use all the information at their disposal to make a dosing decision for the next patient or cohort. BLRM with underdose control should be implemented in a manner that benefits from this history and inherits that culture. The BLRM designs are generally implemented as proscriptive, not prescriptive, which is a considerable advantage. This type of flexibility can be written into other design choices, but the BLRM designs lend themselves to permitting physician judgment within certain limits, and modifications of the BLRM should inherit this culture. The third point relates to the interaction with physicians. Physician judgment, and especially the principal investigator (PI), is key to the choice of the design and design details. For the BLRM or related methods, this includes the DLT definition, the DLT-target, and the DLT-target acceptable range which are all very related. There is also the selection of priors and dose increments and other choices required from the clinical team that drive such models. Usually, a DLT is declared when a patient encounters toxicities that are considered at least a) severe (eg, grade 3 per CTCAE) and beyond readily reversible events that can be managed by pre-planned dose modifications or b) life-threatening (eg, grade 4) adverse events, or c) adverse events such that the patient cannot tolerate continued treatment (eg, delays>2 weeks). The primary patient-level quantity observed in a phase 1 trial is a simple binary DLT (Y/N) call usually based on the first cycle of therapy. Consideration of near DLTs or particularly devastating DLTs, inconsistent DLTs across patients with debatable attribution to treatment, later cycle toxicity and biological correlatives (eg, PK/PD) all require PI judgment (or more generally the dose level review committee/study team). Statistical rules provide a guide and are also required by IRBs to restrict those decisions so no PI can use their judgment to put patients at unbounded risk. This raises an interesting question: When is it necessary to prevent underdosing by a statistical rule? If the PI, who may have been concerned with underdosing a priori prefers not to escalate based on their judgment, is anything gained by a statistical underdosing rule based on a limited Bayesian framework created by abstract queries of the PI? The PI can review the actual data at hand and use their judgment which is always layered upon their prior experience and knowledge. If there are no DLTs, but every patient experienced moderate toxicity, that may be too high a dose in some settings. Beyond this consideration, the designation of underdosing presumes that the hypothesized clinical efficacy increases with increased dose levels; this generally cannot be confirmed in a phase 1 setting. Ultimately, if the target DLT-rate is 25%, the most commonly selected target, do the physicians need to have an acceptable interval of 0.16 to 0.33 or would it be just as reasonable to set the range as 0% to 33% DLT rate for the permissible doses? By identifying a set of doses that are tolerable, the doses to be tested for future investigation can be chosen by supplementary information, such as a measure of the putative biological mechanism of action and number of cycles tolerated. The candidate dose (or doses) for further study would then not be restricted to a mathematically framed underdosing rule. As a result, this trade-off between underdosing and overdosing as proposed may find more applications where it is reframed to be a concern for an excessive risk of two competing adverse events such as DLT vs graft-rejection that are expected to be present on different ends of the dose ladder on a graft-rejection drug and where simulation studies are often used to help modify rule-based designs.13 The clinical settings where a minimum DLT percentage must be pre-specified are very limited, and this should be carefully discussed. The fourth point to address is the statement11 that “Simulation results reveal that the new designs have better accuracy and treat more patients at the MTD.” This is a common statement seen in DLT-target methods, but why is it good to treat more patients at the MTD? The MTD is not necessarily the recommended dose, especially if targeting a 25% DLT rate. This can be compared to the traditional 3 + 3 rules where empirically 0 or 1 in 6 experience a DLT (0%-16.7%) at the MTD. And if the target was thought to be the recommended phase 2 dose (RP2D), how would the PI have any expectation that any pre-specified target DLT rate identifies a good RP2D? Does accuracy on an arbitrary target matter? There are no clinical data to support any target DLT-rate.14 Today, the BLRM original first author prefers to specify a “range of acceptable toxicity, for example a 0-15%” DLT rate, depending on the clinical setting (Beat Neuenschwander, personal communication), and tutorials from more than a decade ago when there was less targeted therapy than today already suggested 10%-20% might be more appropriate for treatments given over many cycles.4 This is also consistent with a physician survey raising concerns about DLT-rate selection and related simulations.15 Users of these methods need to make sure not to conflate the MTD with candidate recommended phase 2 doses. That any such designs are appropriate in only some very select settings does not detract from the effort. The method is both thought provoking and may provide some advantages over competing designs in certain situations. Inheriting the culture and flexibility of the BLRM method is one of its key strengths. Clinical judgment and flexibility can improve the actual reproducibility of phase I studies and improve the selection of the candidate dose (or doses) to further explore, even if mathematical reproducibility suffers. This is because the same patients, coming in the same order is never possible. It is challenging to convey such fuzzy logic, so at this point it will be left as just an opinion. Data sharing is not applicable to this article as no new data were created or analyzed in this study. *(abstract via OpenAlex)*

Biometrics Diagnostics & Prediction · 发表于 2022-11-17

Estimating the Area under the ROC Curve When Transporting a Prediction Model to a Target Population

Bing Li, Constantine Gatsonis, Issa J. Dahabreh, Jon A. Steingrimsson

摘要

We propose methods for estimating the area under the receiver operating characteristic (ROC) curve (AUC) of a prediction model in a target population that differs from the source population that provided the data used for original model development. If covariates that are associated with model performance, as measured by the AUC, have a different distribution in the source and target populations, then AUC estimators that only use data from the source population will not reflect model performance in the target population. Here, we provide identification results for the AUC in the target population when outcome and covariate data are available from the sample of the source population, but only covariate data are available from the sample of the target population. In this setting, we propose three estimators for the AUC in the target population and show that they are consistent and asymptotically normal. We evaluate the finite-sample performance of the estimators using simulations and use them to estimate the AUC in a nationally representative target population from the National Health and Nutrition Examination Survey for a lung cancer risk prediction model developed using source population data from the National Lung Screening Trial.

Biometrics Causal Inference Survival Analysis · 发表于 2022-11-17

A Sensitivity Analysis Approach for the Causal Hazard Ratio in Randomized and Observational Studies

Rachel Axelrod, Daniel Nevo

摘要

The hazard ratio (HR) is often reported as the main causal effect when studying survival data. Despite its popularity, the HR suffers from an unclear causal interpretation. As already pointed out in the literature, there is a built-in selection bias in the HR, because similarly to the truncation by death problem, the HR conditions on post-treatment survival. A recently proposed alternative, inspired by the Survivor Average Causal Effect, is the causal HR, defined as the ratio between hazards across treatment groups among the study participants that would have survived regardless of their treatment assignment. We discuss the challenge in identifying the causal HR and present a sensitivity analysis identification approach in randomized controlled trials utilizing a working frailty model. We further extend our framework to adjust for potential confounders using inverse probability of treatment weighting. We present a Cox-based and a flexible non-parametric kernel-based estimation under right censoring. We study the finite-sample properties of the proposed estimation methods through simulations. We illustrate the utility of our framework using two real-data examples.

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