Hotelling's test, multiple t‐tests, and uterine artery Doppler evaluation

Abstract

I read with interest the paper by Lima et al. in the September 2009 edition of Ultrasound in Obstetrics and Gynecology1. The authors assessed pregnant women showing evidence of premature labor who were given oral nifedipine as a tocolytic agent, performing Doppler ultrasound evaluation before drug administration and after 5 and 24 h. Two Doppler ultrasound parameters, resistance index (RI) and peak systolic velocity (PSV), were recorded from three vessels: maternal uterine, fetal middle cerebral (MCA) and umbilical arteries. Significant changes were observed only in the fetal MCA; a reduction in PSV was observed 5 h after nifedipine (but not after 24 h) and a reduction in RI was observed after 24 h (but not after 5 h). The objective of this paper was to evaluate drug safety. Several issues in this paper arouse concern. Firstly the statistical analysis: the authors compared the ultrasound parameters among the three different times using Hotelling's test and performed pair-wise comparisons using multiple paired t-tests. Hotelling's test is usually used in multivariate analyses, but the authors evaluated Doppler parameters by a univariate approach where repeated measures ANOVA would usually be used. Additionally, two problems are expected when using multiple t-tests: the interpretation becomes cognitively more difficult as multiple results are obtained, and the probability of at least one type I error occurring somewhere in the analysis greatly increases. This increase in type I error resulted in findings of additional significant differences in the PSV of the umbilical and right uterine arteries in the pair-wise comparison that were not observed when the sets of measurements were assessed using Hotelling's test. These problems could be avoided by performing repeated measures ANOVA and a post-hoc test, such as Tukey's test. When describing how they determined their sample size, the authors made a small mistake. They reported an SD for MCA-RI of 0.005 without any reference, but the SD values for this parameter reported for their study population were 0.04 (evaluation before drug and after 5 h) and 0.06 (after 24 h) so I presume that the SD used by the authors to estimate sample size was 0.05 instead of 0.005. However, considering SD = 0.05, the estimated sample size to determine a difference of 0.03 would be 44 and not 33, as reported. Regarding Doppler evaluation, the authors reported use of the triplex system in all scans. When using triplex, spectral Doppler waves are interleaved with color Doppler and B-mode ultrasound waves, which results in noisy signals and limitation of the pulse repetition frequency that can be used. This should be avoided, especially when evaluating uterine arteries, when high velocities are expected. Moreover, the reported measurements of uterine artery PSV are far below the values expected; when evaluating these vessels I usually notice PSV between 100 cm/s and 200 cm/s, sometimes reaching 300 cm/s, whilst the reported means were below 70 cm/s in this paper. This difference was probably due to use of triplex or improper angle correction. Finally, in the discussion section the authors explained that the reduction in MCA-RI could be related to a decrease in PSV, which is a component of the numerator in the RI formula. However, this explanation does not fit with the observed results as the decrease in PSV was observed only 5 h after drug administration, when RI remained unchanged, and the decrease in MCA-RI was observed only 24 h after drug administration, when PSV had already increased. W. P. Martins*, * Faculdade de Medicina de Ribeirão Preto da Universidade de São Paulo, Departamento de Ginecologia e Obstetrícia and Escola de Ultra-sonografia e Reciclagem Médica de Ribeirão Preto (EURP), Rua Casemiro de Abreu, 660, Ribeirão Preto, São Paulo, Brasil CEP: 14020-060