Statistical Methods in Medical Research最新文献

In clinical trials, we often encounter observations from patients' paired organs. In paired correlated data, there exist various measures to evaluate the therapeutic responses, such as risk difference, relative risk ratio, and odds ratio. These measures are essentially some forms of response rate functions. Based on this point, this article aims to test the equality of response rate functions such that the homogeneity tests of the above measures are special cases. Under an interclass correlation model, the global and constrained maximum likelihood estimations are obtained through algorithms. Furthermore, we construct likelihood ratio, score, and Wald-type statistics and provide the explicit expressions of the corresponding tests based on the risk difference, relative risk ratio, and odds ratio. Monte Carlo simulations are conducted to compare the performance of the proposed methods in terms of the empirical type I error rates and powers. The results show that the score tests perform satisfactorily as their type I error rates are close to the specified nominal level, followed by the likelihood ratio test. The Wald-type tests exhibit poor performance, especially for small sample sizes. A real example is given to illustrate the three proposed test statistics.

Youden index, a linear function of sensitivity and specificity, provides a direct measurement of the highest diagnostic accuracy achievable by a biomarker. It is maximized at the cut-off point that optimizes the biomarker's overall classification rate while assigning equal weight to sensitivity and specificity. In this paper, we consider the problem of estimating the Youden index when only group-tested data are available. The unavailability of individual disease statuses poses a challenge, especially when there is differential false positives and negatives in disease screening. We propose both parametric and nonparametric procedures for estimation of the Youden index, and exemplify our methods by utilizing data from the National Health and Nutrition Examination Survey (NHANES) to evaluate the diagnostic ability of monocyte for predicting chlamydia.

In cluster randomized trials (CRTs) with a binary outcome, intervention effects are usually reported as odds ratios, but the CONSORT statement advocates reporting both a relative and an absolute intervention effect. With a simulation study, we assessed several methods to estimate a risk difference (RD) in the framework of a CRT with adjustment on both individual- and cluster-level covariates. We considered both a conditional approach (with the generalized linear mixed model [GLMM]) and a marginal approach (with the generalized estimating equation [GEE]). For both approaches, we considered the Gaussian, binomial, and Poisson distributions. When considering the binomial or Poisson distribution, we used the g-computation method to estimate the RD. Convergence problems were observed with the GEE approach, especially with low intra-cluster coefficient correlation values, small number of clusters, small mean cluster size, high number of covariates, and prevalences close to 0. All methods reported no bias. The Gaussian distribution with both approaches and binomial and Poisson distributions with the GEE approach had satisfactory results in estimating the standard error. Results for type I error and coverage rates were better with the GEE than GLMM approach. We recommend using the Gaussian distribution because of its ease of use (the RD is estimated in one step only). The GEE approach should be preferred and replaced with the GLMM approach in cases of convergence problems.

The difference between two proportions is the most important parameter in comparing two treatments based on independent two binomials and has garnered widespread application across various fields, particularly in clinical trials. There exists significant interest in devising optimal confidence intervals for the difference. Approximate intervals relying on asymptotic normality may lack reliability, thus calling for enhancements in exact confidence interval construction to bolster reliability and precision. In this paper, we present a novel approach that leverages the most probable test statistic and employs the $h$-function method to construct an optimal exact interval for the difference. We juxtapose the proposed interval against other exact intervals established through methodologies such as the Agresti-Min exact unconditional method, the Wang method, the fiducial method, and the hybrid score method. Our comparative analysis, employing the infimum coverage probability and total interval length as evaluation metrics, underscores the uniformly superior performance of the proposed interval. Additionally, we elucidate the application of these exact intervals using two real datasets.

In a longitudinal randomized study where multiple time-to-event outcomes are collected, the overall treatment effect may be quantified by a composite endpoint defined as the time to the first occurrence of any of the selected events including death. The reverse counting process (RCP) was recently proposed to extend the restricted mean survival time (RMST) approach with an advantage of utilizing observations of events beyond the "first-occurrence" endpoint. However, the interpretation may be questionable because RCP treats all events equally without considering their different associations with the overall survival. In this work, we propose a novel approach, the weighted reverse counting process (WRCP), to construct a weighted composite endpoint to evaluate the overall treatment effect. A multi-state transition model is used to model the association between events, and an adaptive weighting algorithm is developed to determine the weight for individual endpoints based on the association between the nonfatal endpoints and death using the trial data. Simulation studies are presented to compare the performance of WRCP with RCP, log-rank test and RMST approach. The results show that WRCP is a powerful and robust method to detect the overall treatment effect while controlling the clinically false positive rate well across different simulation scenarios.

Instrumental variables (IVs) methods have recently gained popularity since, under certain assumptions, they may yield consistent causal effect estimators in the presence of unmeasured confounding. Existing simulation studies that evaluate the performance of IV approaches for time-to-event outcomes tend to consider either an additive or a multiplicative data-generating mechanism (DGM) and have been limited to an exponential constant baseline hazard model. In particular, the relative merits of additive versus multiplicative IV models have not been fully explored. All IV methods produce less biased estimators than naïve estimators that ignore unmeasured confounding, unless the IV is very weak and there is very little unmeasured confounding. However, the mean squared error of IV estimators may be higher than that of the naïve, biased but more stable estimators, especially when the IV is weak, the sample size is small to moderate, and the unmeasured confounding is strong. In addition, the sensitivity of IV methods to departures from their assumed DGMs differ substantially. Additive IV methods yield clearly biased effect estimators under a multiplicative DGM whereas multiplicative approaches appear less sensitive. All can be extremely variable. We would recommend that survival probabilities should always be reported alongside the relevant hazard contrasts as these can be more reliable and circumvent some of the known issues with causal interpretation of hazard contrasts. In summary, both additive IV and Cox IV methods can perform well in some circumstances but an awareness of their limitations is required in analyses of real data where the true underlying DGM is unknown.

Longitudinal data are frequently encountered in medical research, where participants are followed throughout time. Additional structure and hence complexity occurs when there is pairing between the participants (e.g. matched case-control studies) or within the participants (e.g. analysis of participants' both eyes). Various modelling approaches, identified through a systematic review, are discussed, including (un)paired $t$-tests, multivariate analysis of variance, difference scores, linear mixed models (LMMs), and new or more recent statistical methods. Next, highlighting the importance of selecting appropriate models based on the data's characteristics, the methods are applied to both a real-life case study in ophthalmology and a simulated case-control study. Key findings include the superiority of the conditional LMM and multilevel models in handling paired longitudinal data in terms of precision. Moreover, the article underscores the impact of accounting for intra-pair correlations and missing data mechanisms. Focus will be on discussing the advantages and disadvantages of the approaches, rather than on the mathematical or computational details.

We investigate the optimal allocation design for response adaptive clinical trials, under the average reward criterion. The treatment randomization process is formatted as a Markov decision process and the Bayesian method is used to summarize the information on treatment effects. A span-contraction operator is introduced and the average reward generated by the policy identified by the operator is shown to converge to the optimal value. We propose an algorithm to approximate the optimal treatment allocation using the Thompson sampling and the contraction operator. For the scenario of two treatments with binary responses and a sample size of 200 patients, simulation results demonstrate efficient learning features of the proposed method. It allocates a high proportion of patients to the better treatment while retaining a good statistical power and having a small probability for a trial going in the undesired direction. When the difference in success probability to detect is 0.2, the probability for a trial going in the unfavorable direction is < 1.5%, which decreases further to < 0.9% when the difference to detect is 0.3. For normally distribution responses, with a sample size of 100 patients, the proposed method assigns 13% more patients to the better treatment than the traditional complete randomization in detecting an effect size of difference 0.8, with a good statistical power and a < 0.7% probability for the trial to go in the undesired direction.

Kernel machine regression is a nonparametric regression method widely applied in biomedical and environmental health research. It employs a kernel function to measure the similarities between sample pairs, effectively identifying significant exposures and assessing their nonlinear impacts on outcomes. This article introduces an enhanced framework, the generalized Bayesian kernel machine regression. In comparison to traditional kernel machine regression, generalized Bayesian kernel machine regression provides substantial flexibility to accommodate a broader array of outcome variables, ranging from continuous to binary and count data. Simulations show generalized Bayesian kernel machine regression can successfully identify the nonlinear relationships between independent variables and outcomes of various types. In the real data analysis, we applied generalized Bayesian kernel machine regression to uncover cytosine phosphate guanine sites linked to health-related conditions such as asthma and smoking. The results identify crucial cytosine phosphate guanine sites and provide insights into their complex, nonlinear relationships with outcome variables.

Deviation from the treatment strategy under investigation occurs in many clinical trials. We term this intervention deviation. Per-protocol analyses are widely adopted to estimate a hypothetical estimand without the occurrence of intervention deviation. Per-protocol by censoring is prone to selection bias when intervention deviation is associated with time-varying confounders that also influence counterfactual outcomes. This can be corrected by inverse probability of censoring weighting, which gives extra weight to uncensored individuals who had similar prognostic characteristics to censored individuals. Such weights are computed by modelling selected covariates. Inverse probability of censoring weighting relies on the no unmeasured confounding assumption whose plausibility is not statistically testable. Suboptimal implementation of inverse probability of censoring weighting which violates the assumption will lead to bias. In a simulation study, we evaluated the performance of per-protocol and inverse probability of censoring weighting with different implementations to explore whether inverse probability of censoring weighting is a safe alternative to per-protocol. Scenarios were designed to vary intervention deviation in one or both arms with different prevalences, correlation between two confounders, effect of each confounder, and sample size. Results show that inverse probability of censoring weighting with different combinations of covariates outperforms per-protocol in most scenarios, except for an unusual case where selection bias caused by two confounders is in two directions, and 'cancels' out.