Statistics in Medicine最新文献

The flexibility of finite mixture models makes them suitable candidates for analyzing survival data with complex, multimodal distributions. Such data is often available if the event of interest occurs due to multiple failure modes. Here, we explore the modeling of competing risks time-to-event data with covariates in the presence of long-term survivors in the population using finite mixture models. The mixture cure rate model is used to describe the uncertainty in the population, where the susceptible part of the population is modeled using a finite mixture of Weibull distributions with different shape and scale parameters. Moreover, if information on covariates is available, the cure rate may be modeled using a binary regression model on the covariates. Here, we use the logistic function to relate covariates to the cure rate. The distribution corresponding to the susceptible part may also depend on covariates. To explore such dependency, we model the scale parameter of the Weibull distribution using covariates. Then, we discuss the classical parametric inference for the constructed model based on random and non-informative right-censored competing risks time-to-event data. An efficient method based on the expectation-maximization algorithm is proposed to estimate model parameters, thereby avoiding the complexity of directly maximizing the likelihood function. Additionally, a method for constructing confidence intervals for all model parameters is addressed. A simulation study is performed in the presence of two competing causes to investigate the finite sample properties of the proposed estimation methodologies. Finally, the methods are illustrated by analyzing a real data set on malignant melanoma cancer. Predicting the conditional survival function of an alive patient is of natural interest to an experimenter or medical researcher. A method for estimating such a conditional survival probability is also discussed.

Measurement error is a common challenge in statistical analysis, often leading to incorrect parameter estimation. To address measurement error effects, the simulation and extrapolation (SIMEX) method is one of the widely used approaches because of its flexibility in model specification and generic scope of application. Key concerns of the SIMEX method include the number of repetitions in generating synthetic data and the choice of extrapolation function to recover the corrected estimates from the error-prone ones. In most of the existing developments, the quadratic function is frequently adopted as the extrapolation function. However, when measurement error effects are tremendously severe, quadratic functions may be suboptimal. In addition, the development of theoretical results of existing methods requires an unrealistic assumption that the true extrapolation function is known. To address those concerns, we propose GSIMEX, extending the SIMEX method by considering a higher-order polynomial function as the extrapolation function, which enables us to approximate the unknown and nonlinear extrapolation function. In addition, to improve the accuracy of the corrected estimator, we integrate subset selection and model averaging strategies. The theoretical results of GSIMEX, including the measurement of the approximation and asymptotic normality of the estimator, are rigorously established. Numerical studies are conducted for justification of validation, which show that GSIMEX is valid for dealing with severe measurement error effects and is flexible in handling different types of data structures and regression models. We analyze the simulated and spatial transcriptomics data to illustrate the usage of GSIMEX.

We develop a new framework specifically for early Phase I clinical trials called Bayesian Ordered Lattice Design (BOLD). This study is motivated by two key factors. First, Phase I clinical trials typically involve relatively small sample sizes, which can make the use of prior information on dose-limiting toxicity (DLT) rates highly significant. To address this challenge, the proposed Bayesian methodology incorporates prior information and posterior updating to guide dose selection, toxicity monitoring, early stopping, and identification of the maximum tolerable dose (MTD). Second, a natural ordering among toxicity probabilities across different dose levels can be utilized, with the idea being that analysis of dose-level posterior probabilities can and should acquire insights from data obtained at other dose levels, by leveraging their order relationship. Our proposed approach employs straightforward dose-level Bayesian specifications and relies on intuitive and clinically interpretable DLT rate posterior probabilities for decision-making. Importantly, we show that it can often outperform popular methods in terms of accuracy in determining the MTD. This Bayesian approach is also computationally simple and avoids simulation.

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

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

Understanding the causal effect of a treatment in randomized experiments with noncompliance is of fundamental interest in many domains. Within the instrumental variable (IV) framework, the causal treatment effect can only be reliably assessed for compliers, as they are the only subpopulation whose treatment assignment is influenced by the instrument. In this article, we study the conditional complier quantile treatment effect based on individual characteristics through stratified quantile regression models for compliers with and without treatment, which are flexible in capturing the interaction between treatment and covariates and include the past unified model as a special case. We introduce a tuning parameter-free method that directly utilizes the mixture structure in the compiler problem, departing from past approaches that relied on minimizing a weighted check function with nonparametric method-estimated weights. A novel iterated algorithm is proposed to solve discontinuous equations that involve unknown parameters in a complicated way. The consistency and asymptotic normality of the proposed estimators are established. Numerical results, including extensive simulation studies and real data analysis of the Oregon health insurance experiment and a job training study, show the practical utility of the proposed approach.

Patient-reported outcomes (PROs) are collected daily in clinical trials to measure patients' quality of life, for example by capturing symptoms. These data are often reported on a small-range ordinal scale and analyzed without consideration of their longitudinal characteristics. The emergence of electronic data collection methods for home-based measurements has enabled routine, daily capture of various symptom scores, highlighting the need for statistical methods to analyze frequent ordinal longitudinal data. Both their mean structure over time and variability, which are known to be linked to disease progression, are of interest and can be affected by treatment. To model the dynamics of ordinal PROs, we propose a location-scale latent process model that includes two types of variability across patients: individual underlying level flexibly modeled over time (e.g., with splines) using random effects and covariates, and individual short-term variability with the error variance expressed as a linear structure of covariates (e.g., treatment) and a patient-specific random intercept. The model is estimated in a maximum likelihood framework with an interface in R. The multidimensional intractable integrals in the optimization are approximated using a Quasi-Monte Carlo method. The estimation procedure is validated by a simulation study and we apply the methodology to data from two clinical trials, one in asthma and one in chronic obstructive pulomonary disease (COPD), to evaluate the effect of treatment on the dynamics of various respiratory symptoms and their variability.

In recent years, cancer clinical trials have increasingly encountered non proportional hazards (NPH) scenarios, particularly with the emergence of immunotherapy. In randomized controlled trials comparing immunotherapy with conventional chemotherapy or placebo, late difference and early crossing survival curves scenarios are commonly observed. In such cases, window mean survival time (WMST), the area under the survival curve within a pre-specified interval $\left({\tau }_0,,,{\tau }_1\right)$ , has gained increasing attention due to its superior power compared to restricted mean survival time (RMST), the area under the survival curve up to a pre-specified time point. Considering the increasing use of progression-free survival as a co-primary endpoint alongside overall survival, there is a critical need to establish a WMST estimation method for interval-censored data; however, sufficient research has yet to be conducted. To bridge this gap, this study proposes a WMST inference method utilizing one-point imputations and Turnbull's method. Extensive numerical simulations demonstrate that the WMST estimation method using mid-point imputation for interval-censored data exhibits comparable performance to that using Turnbull's method. Since the former facilitates standard error calculation, we adopt it as the standard method. Numerical simulations on two-sample tests confirm that the proposed WMST testing method have higher power than RMST in late difference and early crossing survival curves scenarios, while having compatible power to the log-rank test under the PH. Furthermore, even when pre-specified ${\tau }_0$ deviated from the clinically desirable time point, WMST consistently maintains higher power than RMST in late difference and early crossing survival curves scenarios.

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

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