Testes múltiplos para comparação de dados funcionais

Abstract

The recording of human motion is an essential requirement for biomechanical studies aimed at understanding normal and altered movement patterns to propose preventive or rehabilitation programs. In these studies, the analysis of angular data from joints over time is common. In this work, we focus on data obtained from the recording of knee flexion-extension angles during a gait cycle, which is the movement of the knee between two consecutive heel strikes on the ground. Therefore, the analyzed data consists of curves of knee flexion-extension angles plotted against time, constructed for different strides performed.

The objective of the study is to investigate muscular balance between the left and right legs of the same individual by comparing bilateral knee flexion-extension curves. Specifically, we aim to test the null hypotheses that there is no difference between the distributions of angular values in different intervals across the gait cycle. To test each hypothesis, we use a permutation-based p-value derived from different sample versions of the Szekelly (2004) proposal for the e-distance between distributions. As multiple null hypotheses are considered simultaneously, we apply Holm's and Bonferroni's multiplicity correction procedures to obtain adjusted p-values.

For the actual data considered, a significant reduction in the number of rejected null hypotheses was observed when using multiplicity corrections, which is more appropriate for problems with a large number of hypotheses being tested. Specifically, when combining intervals associated with rejected hypotheses, the Bonferroni and Holm methods showed a reduction of 35% and 34%, respectively, in the length of regions associated with rejected hypotheses compared to when no multiplicity correction is applied.