Biometrical Journal最新文献

Partially linear models are commonly used in observational studies of the causal effect of treatment and/or exposure when there are observed confounding variables. The models are robust and asymptotically distribution-free for testing the causal null hypothesis. In this research, we investigate methods for estimating the partially linear models with data missing at random in all the variables, including the response, the treatment, and the confounding variables. We develop a general estimation method for inference in partially linear models with nonmonotone missing at random data. It proposes using partially linear working models to improve the estimation efficiency of the standard complete case method. It can be shown that the new estimator is consistent, which does not depend on the correctness of the working models. In addition, we recommend bootstrap estimates for the asymptotic variances and semiparametric models for the missing data probabilities. It is computationally simple and can be directly implemented in standard software. Simulation studies are provided to examine its performance. A real data example with sparsely observed missingness patterns is used to illustrate the method.

Estimating optimal adaptive treatment strategies (ATSs) can be done in several ways, including dynamic weighted ordinary least squares (dWOLS). This approach is doubly robust as it requires modeling both the treatment and the response, but only one of those models needs to be correctly specified to obtain a consistent estimator. For estimating an average treatment effect, doubly robust methods have been shown to combine better with machine learning methods than alternatives. However, the use of machine learning within dWOLS has not yet been investigated. Using simulation studies, we evaluate and compare the performance of the dWOLS estimator when the treatment probability is estimated either using machine learning algorithms or a logistic regression model. We further investigate the use of an adaptive $�m$-out-of-$�n$ bootstrap method for producing inferences. SuperLearner performed at least as well as logistic regression in terms of bias and variance in scenarios with simple data-generating models and often had improved performance in more complex scenarios. Moreover, the $�m$-out-of-$�n$ bootstrap produced confidence intervals with nominal coverage probabilities for parameters that were estimated with low bias. We also apply our proposed approach to the data from a breast cancer registry in Québec, Canada, to estimate an optimal ATS to personalize the use of hormonal therapy in breast cancer patients. Our method is implemented in the R software and available on GitHub https://github.com/kosstre20/MachineLearningToControlConfoundingPersonalizedMedicine.git. We recommend routine use of machine learning to model treatment within dWOLS, at least as a sensitivity analysis for the point estimates.

In the hybrid frequentist-Bayesian approach, the probability of success (PoS) of a trial is the expected value of the traditional power function of a test with respect to a design prior assigned to the parameter under scrutiny. However, this definition is not univocal and some of the proposals do not lack of potential drawbacks. These problems are related to the fact that such definitions are all based on the probability of rejecting the null hypothesis rather than on the probability of choosing the correct hypothesis, be it the null or the alternative. In this article, we propose a unifying, decision-theoretic approach that yields a new definition of PoS as the expected utility of the trial (u-PoS), that is, as the expected probability of making the correct choice between the two hypotheses. This proposal shows a conceptual advantage over previous definitions of PoS; moreover, it produces smaller optimal sample sizes whenever the design prior assigns positive probability to the null hypothesis.

In medical, ecological, and psychological research, there is a need for methods to handle multiple testing, for example, to consider group comparisons with more than two groups. Typical approaches that deal with multiple testing are mean- or variance-based which can be less effective in the context of heavy-tailed and skewed data. Here, the median is the preferred measure of location and the interquartile range (IQR) is an adequate alternative to the variance. Therefore, it may be fruitful to formulate research questions of interest in terms of the median or the IQR. For this reason, we compare different inference approaches for two-sided and noninferiority hypotheses formulated in terms of medians or IQRs in an extensive simulation study. We consider multiple contrast testing procedures combined with a bootstrap method as well as testing procedures with Bonferroni correction. As an example of a multiple testing problem based on heavy-tailed data, we analyze an ecological trait variation in early and late breeding in a medium-sized bird of prey.

Longitudinal biomarkers constitute a broad class of potential surrogate endpoints in clinical trials. Several approaches have been proposed for surrogate validation but available methods for validating a longitudinal biomarker as a surrogate of a time-to-event endpoint such as death remain limited. In this work, we propose a method for validating a longitudinal outcome as a surrogate of a time-to-event endpoint using a combination of joint modeling and mediation analysis. The proportion of the total treatment effect on the time-to-event endpoint due to its effect on the biomarker is used as a surrogacy measure. This method is developed to integrate meta-analytic data using a joint model with random effects at both the individual and trial levels. From this model, the indirect treatment effect through the surrogate as well as the direct and total treatment effects is derived using a mediation formula. A simulation study was designed to evaluate the performance of this approach. We applied this method to a multicentric study on prostate cancer to investigate the use of prostate-specific antigen level as a surrogate for disease-free survival.

Motivated by the study of genetic effect modifiers of cancer, we examined weighting approaches to correct for ascertainment bias in survival analysis. Outcome-dependent sampling is common in genetic epidemiology leading to study samples with too many events in comparison to the population and an overrepresentation of young, affected subjects. A usual approach to correct for ascertainment bias in this setting is to use an inverse probability-weighted Cox model, using weights based on external available population-based age-specific incidence rates of the type of cancer under investigation. However, the current approach is not general enough leading to invalid weights in relevant practical settings if oversampling of cases is not observed in all age groups. Based on the same principle of weighting observations by their inverse probability of selection, we propose a new, more general approach, called the generalized weighted approach. We show the advantage of the new generalized weighted cohort method using simulations and two real data sets. In both applications, the goal is to assess the association between common susceptibility loci identified in genome-wide association studies (GWAS) and cancer (colorectal and breast) using data collected through genetic testing in clinical genetics centers.