Applying Permanova for Multivariate Analysis of Variance in Health Studies

Health data often do not meet the normality assumption, which limits the applicability of traditional analysis of variance methods. The aim of this study is to propose a methodological framework for analyzing such data by examining PERMANOVA (Permutational Multivariate Analysis of Variance), a method that does not require the normality assumption and is particularly suitable for complex datasets, within the context of maternal health data. In the context of maternal healthcare in Bangladesh, the effects of two independent variables—risk and age factors—on multivariate response variables such as Systolic Blood Pressure (mmHg), Diastolic Blood Pressure (mmHg), Blood Sugar (mmol/L), Body Temperature (Fahrenheit), and Heart Rate (beats per minute) were examined using the PERMANOVA method. The first independent variable represents the risk factor, comprising three different risk levels (low, mid, high), while the second independent variable represents the age factor, divided into four age groups (young, adolescent, middle-aged, menopausal). The dependent variables did not follow a normal distribution, as confirmed by the Anderson-Darling test and Mardia’s multivariate normality test. As a result of the PERMANOVA analysis, it was determined that at least two mean differences between the groups of the risk factor and the age factor were statistically significant in terms of the response variables (P<0.01). Furthermore, pairwise comparisons of the factor groups revealed that the mean differences between the low, mid, and high levels of the risk factor, as well as the mean differences among the young, adolescent, and middle-aged groups of the age factor, were statistically significant (P<0.01). However, the mean difference between the middle-aged and menopausal groups for the age factor was found to be statistically insignificant (P>0.01). The PERMANOVA method is recommended for researchers to accurately determine whether the mean differences in factor levels are statistically significant or to identify threshold values of the groups by using multiple response variables simultaneously and performing pairwise comparisons of factor groups.