Statistics in Medicine最新文献

Mediation analysis aims to identify and estimate the effect of an exposure on an outcome that is mediated through one or more intermediate variables. In the presence of multiple intermediate variables, two pertinent methodological questions arise: estimating mediated effects when mediators are correlated, and performing high-dimensional mediation analyses when the number of mediators exceeds the sample size. This paper presents a two-step procedure for high-dimensional mediation analyses. The first step selects a reduced number of candidate mediators using an ad-hoc lasso penalty. The second step applies a procedure we previously developed to estimate the mediated effects, accounting for the correlation structure among the retained candidate mediators. We compare the performance of the proposed two-step procedure with state-of-the-art methods using simulated data. Additionally, we demonstrate its practical application by estimating the causal role of DNA methylation (DNAm) in the pathway between smoking and rheumatoid arthritis (RA) using real data.

Cancer prevention is recognized as a key strategy for reducing disease incidence, mortality, and the overall burden on individuals and society. However, determining when to begin preventive interventions presents a significant challenge: starting too early may lead to more interventions and increased lifetime burdens due to repeated administrations, while delaying may miss opportunities to prevent cancer. Evidence-based recommendations require a benefit-burden analysis that weighs life-years gained against the burden of interventions. With the growing availability of large-scale observational data, there is now an opportunity to empirically evaluate these trade-offs. In this paper, we propose a causal framework for assessing the benefit and burden of cancer prevention, using an illness-death model with semi-competing risks. Extensive simulations demonstrate that the proposed estimators are unbiased, with robust inference across realistic scenarios. We apply this approach to a benefit-burden analysis of the preventive screening for colorectal cancer, utilizing data from the large-scale Women's Health Initiative. Our findings suggest that initiating screening at age 50 years achieves the highest life-year gains with an acceptable incremental burden-to-benefit ratio compared to no screening, contributing valuable real-world evidence to the field of preventive cancer interventions.

Untreated periodontitis causes inflammation within the supporting tissue of the teeth and can ultimately lead to tooth loss. Modeling periodontal outcomes is beneficial as they are difficult and time-consuming to measure, but disparities in representation between demographic groups must be considered. There may not be enough participants to build group-specific models, and it can be ineffective, and even dangerous, to apply a model to participants in an underrepresented group if demographic differences were not considered during training. We propose an extension to the RECaST Bayesian transfer learning framework. Our method jointly models multivariate outcomes, exhibiting significant improvement over the previous univariate RECaST method. Further, we introduce an online approach to model sequential data sets. Negative transfer is mitigated to ensure that the information shared from the other demographic groups does not negatively impact the modeling of the underrepresented participants. The Bayesian framework naturally provides uncertainty quantification on predictions. Especially important in medical applications, our method does not share data between domains. We demonstrate the effectiveness of our method in both predictive performance and uncertainty quantification on simulated data and on a database of dental records from the HealthPartners Institute.

Genomics and other studies encounter many features and a selection of essential features with high accuracy is desired. In recent years, there has been a significant advancement in the use of Bayesian inference for variable (or feature) selection. However, there needs to be more practical information regarding their implementation and assessment of their relative performance. Our goal in this paper is to perform a comparative analysis of approaches, mainly from different Bayesian genres that apply to genomic analysis. In particular, we are examining how well shrinkage, global-local, and mixture priors, SUSIE, and a simple two-step procedure-namely, RFSFS, which we propose-perform in terms of several metrics: FDR, FNR, F-score, and mean squared prediction error under various simulation scenarios. There is no single method that outperforms others uniformly across all scenarios and in terms of variable selection and prediction performance metrics. So, we order the methods based on the average ranking across different scenarios. We found LASSO, spike-and-slab prior with normal slab (SN), and RFSFS are the most competitive methods for FDR and F-score when features are uncorrelated. When features are correlated, SN, SuSIE, and RFSFS are the most competitive methods for FDR whereas LASSO has an edge over SuSIE in terms of F-score. For illustration, we have applied these methods to analyzed The Cancer Genome Atlas Program (TCGA) renal cell carcinoma (RCC) data and have offered methodological direction.

Randomized controlled trials (RCTs) often include subgroup analyses to assess whether treatment effects vary across prespecified patient populations. However, these analyses frequently suffer from small sample sizes, which limit the power to detect heterogeneous effects. Power can be improved by leveraging predictors of the outcome-that is, through covariate adjustment-as well as by borrowing external data from similar RCTs or observational studies. The benefits of covariate adjustment may be limited when the trial sample is small. Borrowing external data can increase the effective sample size and improve power, but it introduces two key challenges: (i) integrating data across sources can lead to model misspecification, and (ii) practical violations of the positivity assumption-where the probability of receiving the target treatment is near zero for some covariate profiles in the external data-can lead to extreme inverse-probability weights and unstable inferences, ultimately negating potential power gains. To account for these shortcomings, we present an approach to improving power in preplanned subgroup analyses of small RCTs that leverages both baseline predictors and external data. We propose de-biased estimators that accommodate parametric, machine learning (ML), and nonparametric Bayesian methods. To address practical positivity violations (PPVs), we introduce three estimators: A covariate-balancing approach, an automated de-biased machine learning (DML) estimator, and a calibrated-DML estimator. We show improved power in various simulations and offer practical recommendations for the application of the proposed methods. Finally, we apply them to evaluate the effectiveness of citalopram for negative symptoms in first-episode schizophrenia (FES) patients across subgroups defined by duration of untreated psychosis (DUP), using data from two small RCTs.

Multi-cancer testing with localization aims to detect signals from any of a set of targeted cancer types and predict the cancer signal origin from a single biological sample. Such tests have the potential to aid clinical decisions and significantly improve health outcomes. When used for multi-cancer screening in an asymptomatic population, these tests are referred to as multi-cancer early detection (MCED) tests. MCED testing has not yet achieved regulatory approval, reimbursement or broad clinical adoption. Some major reasons for this are that the clinical benefits and harms are not well understood, including the risk of unnecessary work-ups and false reassurance from a negative test that could reduce uptake of standard-of-care screening. Part of this uncertainty stems from the use of clinically obtuse metrics to assess the test's clinical validity. Traditionally, performance of MCED tests has been quantified using aggregate measures, disregarding the joint distribution of cancer type, stage (both at intended-use incidence rates) and predicted cancer signal origin, thereby obscuring biological variability and underlying differences in the test's behavior and limiting insight into true effectiveness. Clinically informative evaluation of an MCED test's performance requires metrics that are specific to cancer type, stage and predicted cancer origin at expected incidence rates in the intended-use population. In the context of a case-control sampling design, this paper derives analytical methods that allow for unbiased estimation of cancer-specific intrinsic accuracy, predicted cancer signal origin-specific predictive value and the marginal test classification distribution, each with corresponding valid confidence interval formulae. A simulation study is presented that evaluates performance of the proposed methodology and provides guidance for implementation. An application to a published MCED test dataset is given. The derived statistical analysis framework in general allows for estimation and inference for pointed metrics of a multi-category test that enables precisely informed decision-making, supports optimized trial designs across classical, digital, AI-driven, and hybrid stratified diagnostic screening platforms, and facilitates informed healthcare decisions by clinicians, policymakers, regulators, scientists, and patients.

Medical research often involves the study of composite endpoints that combine multiple clinical events to assess the efficacy of treatments. When constructing composite endpoints, it is a common practice to analyze the time to the first event. However, this approach overlooks outcomes that occur after the first event, resulting in information loss. Furthermore, the terminal event can not only be of interest but also, be a competing risk for other types of outcomes. While existing semi-parametric regression models can be used to analyze both fatal (terminal) and non-fatal composite events, potential nonlinear covariate effects on the logarithm of the rate function have not been addressed. To address this important issue, we introduce random forests for composite endpoints (Rforce) consisting of non-fatal composite events and terminal events. Rforce utilizes generalized estimating equations to build trees and handles the dependent censoring due to the terminal events with the concept of pseudo-at-risk duration. Simulation studies and real data analysis are conducted to demonstrate the performance of Rforce.

In this paper, we investigate the selection of minimal and efficient covariate adjustment sets for the imputation-based regression calibration method, which corrects for bias due to continuous exposure measurement error. We use directed acyclic graphs to illustrate how subject-matter knowledge aids in selecting these sets. For unbiased measurement error correction, researchers must collect, in both main and validation studies, (I) common causes of both the true exposure and the outcome, and (II) common causes of both measurement error and the outcome. For regression calibration under linear models, at minimum, covariate set (I) must be adjusted for in both the measurement error model (MEM) and the outcome model, while set (II) should be adjusted for in at least the MEM. Adjusting for non-risk factors that are correlates of true exposure or measurement error within the MEM alone improves efficiency. We apply this covariate selection approach to the Health Professionals Follow-up Study, assessing fiber intake's effect on cardiovascular disease. We also highlight potential pitfalls in data-driven MEM building that ignores structural assumptions. Additionally, we extend existing estimators to allow for effect modification. Finally, we caution against using regression calibration to estimate the effect of true nutritional intake through calibrating biomarkers.

Network meta-analysis of randomized controlled trials is traditionally conducted on a single outcome measured at one time point. However, many trials also feature a secondary outcome and both outcomes may have been reported at multiple time points. Existing network meta-analysis methods for synthesizing continuous outcome data from such trials focus on either the longitudinal data aspect or the multiple outcomes aspect, but not on both simultaneously. In this paper, we present two Bayesian network meta-analysis models that account for the correlation of outcome measurements over time using Gaussian random walks with drift. The first model is suitable for a single continuous outcome measured at multiple time points, while the second model extends the first model to allow incorporation of a second outcome through cointegration of random walks. A simulation study to evaluate several statistical properties of these models is conducted. The results indicate that both proposed models produce unbiased estimates of relative treatment effect and drift parameters, as well as reasonable coverage. Furthermore, in some scenarios, using the cointegration model yields small gains in precision over using the single outcome model. Based on various performance measures, both proposed models also outperform an existing random walk network meta-analysis model previously used by investigators to synthesize osteoarthritis trials data. The proposed models are illustrated with an application to trials evaluating treatments for knee and hip osteoarthritis. Both models are useful additions to existing tools available to investigators undertaking a network meta-analysis of continuous outcome data at multiple time points.

In pharmacoepidemiology, safety and effectiveness are frequently evaluated using readily available administrative and electronic health records data. In these settings, detailed confounder data are often not available in all data sources and therefore missing on a subset of individuals. Multiple imputation (MI) and inverse-probability weighting (IPW) are go-to analytical methods to handle missing data and are dominant in the biomedical literature. Doubly-robust methods, which are consistent under fewer assumptions, can be more efficient with respect to mean-squared error. We discuss two practical-to-implement doubly-robust estimators, generalized raking and inverse probability-weighted targeted maximum likelihood estimation (TMLE), which are both currently under-utilized in biomedical studies. We compare their performance to IPW and MI in a detailed numerical study for a variety of synthetic data-generating and missingness scenarios, including scenarios with rare outcomes and a high missingness proportion. Further, we consider plasmode simulation studies that emulate the complex data structure of a large electronic health records cohort in order to compare anti-depressant therapies in a rare-outcome setting where a key confounder is prone to more than 50% missingness. We provide guidance on selecting a missing data analysis approach, based on which methods excelled with respect to the bias-variance trade-off across the different scenarios studied.