Statistical Methods in Medical Research最新文献

Joint modeling of multiple longitudinal markers and time-to-event outcomes is common in clinical studies. However, as the number of markers increases, estimation becomes computationally challenging or infeasible due to long runtimes and convergence difficulties. We propose a novel two-stage Bayesian approach for estimating joint models involving multiple longitudinal measurements and time-to-event outcomes. The proposed method is related to the standard two-stage approach, which separately estimates longitudinal submodels and then incorporates their outputs as time-dependent covariates in a survival model. Unlike the standard method, our first stage estimates separate one-marker joint models for the event and each longitudinal marker, rather than relying on mixed-effects models. From these models, predictions of expected current values and/or slopes of individual marker trajectories are obtained, thereby avoiding bias due to informative dropout. In the second stage, a proportional hazards model is fitted that includes the predicted current values and/or slopes of all markers as time-dependent covariates. To account for uncertainty in the first-stage predictions, a multiple imputation strategy is employed when estimating the survival model. This approach enables the construction of prediction models based on a large number of longitudinal markers that would otherwise be computationally intractable using conventional multi-marker joint models. The performance of the proposed method is evaluated through simulation studies and an application to the public PBC2 dataset. Additionally, it is applied to predict dementia risk using a real-world dataset with seventeen longitudinal markers. To facilitate practical use, we developed an R package, TSJM, which is freely available on GitHub: https://github.com/tbaghfalaki/TSJM.

When data become increasingly complex, desirable models are required to be more flexible for analyzing survival data. Building upon the existing functional Cox model, we introduce a novel functional varying-coefficient Cox model and the corresponding estimation algorithms are proposed in this article. The proposed model can simultaneously handle survival data with varying-coefficient covariates and functional covariates, thereby significantly enhancing the adaptability of survival models. The model performance is evaluated by simulation studies, and a real application using Alzheimer's disease neuroimaging initiative (ADNI) data is used to illustrate the practicality of the proposed model.

Heterogeneity of the progression of neurodegenerative diseases is one of the main challenges faced in developing therapies. Thanks to the increasing number of clinical databases, progression models have allowed a better understanding of this heterogeneity. Joint models have proven their effectiveness by combining longitudinal and survival data. Nevertheless, they require a reference time, which is ill-defined for neurodegenerative diseases, where biological underlying processes start before the first symptoms. In this work, we propose a joint non-linear mixed-effect model with a latent disease age, to overcome this need for a precise reference time. We used a longitudinal model with a latent disease age as a longitudinal sub-model. We associated it with a survival sub-model that estimates a Weibull distribution from the latent disease age. We validated our model on simulated data and benchmarked it with a state-of-the-art joint model on data from patients with Amyotrophic Lateral Sclerosis (ALS). Finally, we showed how the model could be used to describe ALS heterogeneity. Our model got significantly better results than the state-of-the-art joint model for absolute bias on ALS functional rating scale revised score (4.21(SD 4.41) versus 4.24(SD 4.14)($p$-value$=$1.4$\times {10}^{-17}$)), and mean-cumulative-AUC for right-censored events on death (0.67(0.07) versus 0.61(0.09)($p$-value$=$1.7$\times {10}^{-03}$)). To conclude, we propose a new model better suited in the context of unreliable reference time.

Simultaneous confidence intervals that are compatible with a given closed test procedure are often non-informative. More precisely, for a one-sided null hypothesis, the bound of the simultaneous confidence interval can stick to the border of the null hypothesis, irrespective of how far the point estimate deviates from the null hypothesis. This has been illustrated for the Bonferroni-Holm and fall-back procedures, for which alternative simultaneous confidence intervals have been suggested, that are free of this deficiency. These informative simultaneous confidence intervals are not fully compatible with the initial multiple test, but are close to it and hence provide similar power advantages. They provide a multiple hypothesis test with strong family wise error rate control that can be used in replacement of the initial multiple test. The current paper extends previous work for informative simultaneous confidence intervals to graphical test procedures. The information gained from the newly suggested simultaneous confidence intervals is shown to be always increasing with increasing evidence against a null hypothesis. The new simultaneous confidence intervals provide a compromise between information gain and the goal to reject as many hypotheses as possible. The simultaneous confidence intervals are defined via a family of dual graphs and the projection method. A simple iterative algorithm for the computation of the intervals is provided. A simulation study illustrates the results for a complex graphical test procedure.

Dual agent dose-finding trials study the effect of a combination of more than one agent, where the objective is to find the Maximum Tolerated Dose Combination, the combination of doses of the two agents that is associated with a pre-specified risk of being unsafe. In a Phase I/II setting, the objective is to find a dose combination that is both safe and active, the Optimal Biological Dose, that optimises a criterion based on both safety and activity. Since Oncology treatments are typically given over multiple cycles, both the safety and activity outcome can be considered as late-onset, potentially occurring in the later cycles of treatment. This work proposes two model-based designs for dual-agent dose finding studies with late-onset activity and late-onset toxicity outcomes, the Joint time-to-event (TITE) partial order continual reassessment method and the Joint TITE Bayesian logistic regression model. Their performance is compared alongside a model-assisted comparator in a comprehensive simulation study motivated by a real trial example, with an extension to consider alternative sized dosing grids. It is found that both model-based methods outperform the model-assisted design. Whilst on average the two model-based designs are comparable, this comparability is not consistent across scenarios.

Inverse probability of censoring weighting is an approach used to estimate the hypothetical treatment effect that would have been observed in a clinical trial if certain intercurrent events had not occurred. Despite the unbiased estimates obtained by inverse probability of censoring weighting when its key assumptions are satisfied, large standard errors and wide confidence intervals can be potential concerns. Inverse probability of censoring weighting with unstabilised weights can be simply implemented by calculating the reciprocal of the probability of being uncensored by the intercurrent events. To improve precision, stabilisation can be realised by replacing the numerator in the unstabilised weights with functions of the time and baseline covariates. Here, we aim to investigate whether stabilised weight is a preferred choice and if so how we should specify the numerator. In a simulation study, we assessed the performance of inverse probability of censoring weighting implementations with unstabilised weights and with different forms of stabilisation when the outcome analysis model was correctly specified or mis-specified. Scenarios were designed to vary the prevalence of the intercurrent event in one or both randomised arms, the existence of a deterministic intercurrent event, the indirect effect through baseline covariates and overall treatment effect, the existence and the pattern of time-varying effect and sample size. Results show that compared with unstabilised weights, stabilisation improves the efficiency of the inverse probability of censoring weighting estimator in most cases and the improvement is obvious when we stabilise for the baseline covariates. However, stabilisation risks increasing the bias when the outcome analysis model is mis-specified.

Continuous-time Markov chain (CTMC) models and latent classification methods are commonly used to analyze longitudinal categorical outcomes in medical research. While CTMC models are popular for their simplicity and effectiveness, their assumption of constant transition rates presents limitations in capturing dynamic behaviors. To address this, non-homogeneous continuous-time Markov chains (NH-CTMCs) have been developed, incorporating time-varying transition rates to enhance model flexibility. In this study, we leverage closed-form transition probabilities for a fully ergodic two-state NH-CTMC model and propose a latent class clustering approach to identify heterogeneous transition rate patterns within the population. We emphasize the potential advantages of these models in health sciences, particularly for longitudinal studies where transition rates vary over time and across subgroups. Additionally, we demonstrate the practical application of our model using data from an ambulatory hypertension monitoring study.

Accurate epidemic forecasting is critical for effective public health interventions. This study compares Bayesian and Frequentist estimation frameworks within deterministic compartmental epidemic models, focusing on nonlinear least squares (NLS) optimization versus Bayesian inference assuming a normal likelihood and using MCMC sampling in Stan. Rather than evaluating all methodological variants, we assess forecasting performance under a shared modeling structure and error assumption. The findings apply to specific implementations of both approaches. Performance is evaluated using simulated datasets (with $R_0=2$ and 1.5) and historical outbreaks, including the 1918 influenza pandemic, the 1896-1897 Bombay plague epidemic, and the COVID-19 pandemic. Metrics include mean absolute error (MAE), root mean squared error (RMSE), weighted interval score (WIS), and 95% prediction interval coverage. Forecasting performance varies by epidemic phase and dataset; no method consistently dominates. The Frequentist method performs well at the peak in simulations and in the post-peak phases of real outbreaks but is less accurate pre-peak. Bayesian methods, especially those with uniform priors, offer higher predictive accuracy early in epidemics and stronger uncertainty quantification when data are sparse or noisy. Frequentist methods often yield more accurate point forecasts with lower MAE, RMSE, and WIS, though their interval estimates are less robust. We also discuss the influence of prior choice and the effects of longer forecasting horizons on convergence and computational efficiency. These findings provide practical guidance for selecting estimation strategies suited to epidemic phase and data quality, aiding forecast-based decision-making.

We study spline-based efficient estimation of frailty models for panel count data using a penalization technique. An easy-to-implement and computationally efficient two-stage iterative expectation-maximization algorithm is proposed for the analysis. A general quasi-likelihood estimation that does not specify the stochastic model of the underlying counting process is developed to provide flexibility for model fitting. A powerful score test is discussed to detect the presence of overdispersion in count data. The proposed methods are assessed via an extensive simulation and further illustrated by analyzing data from a non-melanoma skin cancer chemoprevention study.

In several clinical areas, traditional clinical trials often use a responder outcome, a composite endpoint that involves dichotomising a continuous measure. An augmented binary method that improves power while retaining the original responder endpoint has previously been proposed. The method leverages information from the undichotomised component to improve power. We extend this method for basket trials, which are gaining popularity in many clinical areas. For clinical areas where response outcomes are used, we propose the new augmented binary method for basket trials that enhances efficiency by borrowing information on the treatment effect between subtrials. The method is developed within a latent variable framework using a Bayesian hierarchical modelling approach. We investigate the properties of the proposed methodology by analysing point estimates and high-density intervals in various simulation scenarios, comparing them to the standard analysis for basket trials that assumes binary outcomes. Our method results in a reduction of 95% high-density interval of the posterior distribution of the log odds ratio and an increase in power when the treatment effect is consistent across subtrials. We illustrate our approach using real data from two clinical trials in rheumatology.