Multivariate approach in analyzing medical data with correlated multiple outcomes: An exploration using ACCORD trial data

Abstract

Background: In clinical setting, to answer a research question, very often more than one outcome variables are used which are statistically correlated. Univariate analysis approach, which is commonly used in such context, violates the assumption of independence for correlated variables, while multivariate approach could give more robust and precise clinical decision by accounting this correlation. This paper is a demonstration of the change in statistical decision in multivariate approach compared to univariate approach in analysing medical data with multiple correlated outcome variables. Method: ology: The data used in this paper was the ACCORD trial data and the variables were Systolic blood pressure (SBP) and Diastolic blood pressure (DBP) which are correlated. The condition of bivariate normality was checked after removing the outliers. We compared the differences in means of SBP and DBP separately between the groups at different follow up time points by univariate approach using unpaired 't' test and thereafter together by bivariate approach using Hotelling 'T 2' . Further the role of varied levels of correlation between SBP and DBP coupled with the variations in sample size in changing the statistical inference has been investigated. Robustness of these results were demonstrated through 1000 samples of simulated data from Accord trial data. Results: The statistical decision regarding difference in means of either SBP or DBP between the two groups at baseline, 12th, 36th and 84th month in both univariate and bivariate approaches were similar. At baseline and 84th month the hypotheses were accepted and at 12th and 36th month the hypotheses were rejected with both the approaches. However at 60th and 72nd month the univariate and multivariate approach results contradicted each other which reflect the significance of correlation in modifying the inference. The analysis results further indicated that varied levels of correlation and sample size too modify the inference in bivariate approach and contradict univariate results. Conclusion: Studies with multiple correlated outcomes, the multivariate approach which accounts this statistical correlation has potential to change statistical decisions and provide more precise results. Hence multivariate approach is recommended in such situations.

Psychosocial factors and hippocampal subfields: The Medea‐7T study

“…In these models, all hippocampal subfields are entered as one outcome, resulting in a single model per each psychosocial factor (see Code S1 ). Previous literature has shown that multivariate approaches increase the power of the model as well as reduce type I error compared with univariate approaches that ignore the correlation between outcomes (Mishra et al, 2021 ). While in univariate analyses, one can adjust the p value, the assumption of independence between outcomes is violated when they are correlated.…”

Application of bivariate meta-analytic approach for pooling effect measures of correlated multiple outcomes in medical research

“…9 Multivariate analysis which accounts for these correlations in making inferences is the correct method for analysing such data and various authors have attempted this in medical data analysis. 10,11 Further, the above mentioned correlation of the primary outcome variables is likely to exist while pooling such effect measures in metaanalysis. Hence, not adjusting for this correlation in the meta-analysis might produce effect measures with compromised precision.…”

Network Analysis for Uncovering the Relationship between Host Response and Clinical Factors to Virus Pathogen: Lessons from SARS-CoV-2

“…Therefore, multivariate analyses are a preferable approach when analysing complex datasets, providing a more realistic basis for robust and accurate clinical decisions [6]. Moreover, multivariate analysis enables the assessment of the contribution of multiple factors concerning one or more clinical factors to reflect reality, reveal relationships between the factors analysed, and reduce the bias of univariate patient characteristics across studies [6][7][8][9]. Nowadays, several multivariate approaches are available that consider complex, multidimensional relationships between factors.…”