Biometrical Journal最新文献

We propose a novel method to adjust for unmeasured time-stable confounding when the time between consecutive treatment administrations is fixed. We achieve this by focusing on a new-user cohort. Furthermore, we envisage that all time-stable confounding goes through the potential time on treatment as dictated by the disease condition at the initiation of treatment. Following this logic, we may eliminate all unmeasured time-stable confounding by adjusting for the potential time on treatment. A challenge with this approach is that right censoring of the potential time on treatment occurs when treatment is terminated at the time of the event of interest, for example, if the event of interest is death. We show how this challenge may be solved by means of the expectation-maximization algorithm without imposing any further assumptions on the distribution of the potential time on treatment. The usefulness of the methodology is illustrated in a simulation study. We also apply the methodology to investigate the effect of depression/anxiety drugs on subsequent poisoning by other medications in the Danish population by means of national registries. We find a protective effect of treatment with selective serotonin reuptake inhibitors on the risk of poisoning by various medications (1- year risk difference of approximately $��-�3�\%�$) and a standard Cox model analysis shows a harming effect (1-year risk difference of approximately $��2�\%�$), which is consistent with what we would expect due to confounding by indication. Unmeasured time-stable confounding can be entirely adjusted for when the time between consecutive treatment administrations is fixed.

Assessment of covariate balance is a key step when performing comparisons between groups particularly in real-world data. We generally evaluate it on baseline covariates, but rarely on longitudinal ones prior to a management decision. We could use pointwise standardized mean differences, standardized differences of slopes, or weights from the model for such purpose. Pointwise differences could be cumbersome for densely sampled longitudinal markers and/or measured at different points. Slopes are suitable for linear or transformable models but not for more complex curves. Weights do not identify the specific covariate(s) responsible for imbalances. This work presents the Fréchet distance as a viable alternative to assess balance of time-dependent covariates. A set of linear and nonlinear curves for which their standardized difference or differences in functional parameters were within 10% sought to identify the Fréchet distance equivalent to this threshold. This threshold is dependent on the level of noise present and thus within group heterogeneity and error variance are needed for its interpretation. Applied to a set of real curves representing the monthly trajectory of hemoglobin A1c from diabetic patients showed that the curves in the two groups were not balanced at the 10% mark. A Beta distribution represents the Fréchet distance distribution reasonably well in most scenarios. This assessment of covariate balance provides the following advantages: It can handle curves of different lengths, shapes, and arbitrary time points. Future work includes examining the utility of this measure under within-series missingness, within-group heterogeneity, its comparison with other approaches, and asymptotics.

In biomedical studies, investigators often encounter clustered data. The cluster sizes are said to be informative if the outcome depends on the cluster size. Ignoring informative cluster sizes in the analysis leads to biased parameter estimation in marginal and mixed-effect regression models. Several methods to analyze data with informative cluster sizes have been proposed; however, methods to test the informativeness of the cluster sizes are limited, particularly for the marginal model. In this paper, we propose a score test and a Wald test to examine the informativeness of the cluster sizes for a generalized linear model, a Cox model, and a proportional subdistribution hazards model. Statistical inference can be conducted through weighted estimating equations. The simulation results show that both tests control Type I error rates well, but the score test has higher power than the Wald test for right-censored data while the power of the Wald test is generally higher than the score test for the binary outcome. We apply the Wald and score tests to hematopoietic cell transplant data and compare regression analysis results with/without adjusting for informative cluster sizes.

The selection of best variables is a challenging problem in supervised and unsupervised learning, especially in high-dimensional contexts where the number of variables is usually much larger than the number of observations. In this paper, we focus on two multivariate statistical methods: principal components analysis and partial least squares. Both approaches are popular linear dimension-reduction methods with numerous applications in several fields including in genomics, biology, environmental science, and engineering. In particular, these approaches build principal components, new variables that are combinations of all the original variables. A main drawback of principal components is the difficulty to interpret them when the number of variables is large. To define principal components from the most relevant variables, we propose to cast the best subset solution path method into principal component analysis and partial least square frameworks. We offer a new alternative by exploiting a continuous optimization algorithm for best subset solution path. Empirical studies show the efficacy of our approach for providing the best subset solution path. The usage of our algorithm is further exposed through the analysis of two real data sets. The first data set is analyzed using the principle component analysis while the analysis of the second data set is based on partial least square framework.

Adjustment of statistical significance levels for repeated analysis in group-sequential trials has been understood for some time. Adjustment accounting for testing multiple hypotheses is also well understood. There is limited research on simultaneously adjusting for both multiple hypothesis testing and repeated analyses of one or more hypotheses. We address this gap by proposing adjusted-sequential p-values that reject when they are less than or equal to the family-wise Type I error rate (FWER). We also propose sequential $�p$-values for intersection hypotheses to compute adjusted-sequential $�p$-values for elementary hypotheses. We demonstrate the application using weighted Bonferroni tests and weighted parametric tests for inference on each elementary hypothesis tested.