Multivariate permutation tests for two sample testing in presence of nondetects with application to microarray data.

Very often, data collected in medical research are characterized by censored observations and/or data with mass on the value zero. This happens for example when some measurements fall below the detection limits of the specific instrument used. This type of left censored observations is called "nondetects". Such a situation of an excessive number of zeros in a data set is also referred to as zero-inflated data. In the present work, we aim at comparing different multivariate permutation procedures in two-sample testing for data with nondetects. The effect of censoring is investigated with regard to the different values that may be attributed to nondetected values, both under the null hypothesis and under alternative. We motivate the problem using data from allergy research.

Nonparametric combination tests for comparing two survival curves with informative and non-informative censoring.

This paper looks at permutation methods used to deal with hypothesis testing within the survival analysis framework. In the literature, several attempts have been made to deal with the comparison of survival curves and, depending on the survival and hazard functions of two groups, they can be more or less efficient in detecting differences. Furthermore, in some situations, censoring can be informative in that it depends on treatment effect. Our proposal is based on the nonparametric combination approach and has proven to be very effective under different configurations of survival and hazard functions. It allows the practitioner to test jointly on primary and censoring events and, by using multiple testing methods, to assess the significance of the treatment effect separately on the survival and the censoring process.

[Statistical tests in medical research: traditional methods vs. multivariate NPC permutation tests]. / Test statistici nella ricerca medica: metodi tradizionali vs test multivariati di permutazione NPC.

Statistical tests in medical research: traditional methods vs. multivariate npc permutation tests.Within medical research, a useful statistical tool is based on hypotheses testing in terms of the so-called null, that is the treatment has no effect, and alternative hypotheses, that is the treatment has some effects. By controlling the risks of wrong decisions, empirical data are used in order to possibly reject the null hypotheses in favour of the alternative, so that demonstrating the efficacy of a treatment of interest. The multivariate permutation tests, based on the nonparametric combination - NPC method, provide an innovative, robust and effective hypotheses testing solution to many real problems that are commonly encountered in medical research when multiple end-points are observed. This paper discusses the various approaches to hypothesis testing and the main advantages of NPC tests, which consist in the fact that they require much less stringent assumptions than traditional statistical tests. Moreover, the related results may be extended to the reference population even in case of selection-bias, that is non-random sampling. In this work, we review and discuss some basic testing procedures along with the theoretical and practical relevance of NPC tests showing their effectiveness in medical research. Within the non-parametric methods, NPC tests represent the current "frontier" of statistical research, but already widely available in the practice of analysis of clinical data.

Conditional tests in a competing risks model.

Testing for equality of competing risks based on their cumulative incidence functions (CIFs) or their cause specific hazard rates (CSHRs) has been considered by many authors. The finite sample distributions of the existing test statistics are in general complicated and the use of their asymptotic distributions can lead to conservative tests. In this paper we show how to perform some of these tests using the conditional distributions of their corresponding test statistics instead (conditional on the observed data). The resulting conditional tests are initially developed for the case of k = 2 and are then extended to k > 2 by performing a sequence of two sample tests and by combining several risks into one. A simulation study to compare the powers of several tests based on their conditional and asymptotic distributions shows that using conditional tests leads to a gain in power. A real life example is also discussed to show how to implement such conditional tests.