An end-to-end workflow for statistical analysis and inference of large-scale biomedical datasets

Throughout time, as medical and epidemiological studies have grown larger in scale, the challenges associated with extracting useful and relevant information from these data has mounted. General health surveys provide a good example for such studies as they usually cover large populations and are conducted throughout long periods in multiple locations. The challenges associated with interpreting the results of such studies include: the presence of both categorical and continuous variables and the need to compare them within a single statistical framework; the presence of variations in data resulting from the technical limitations in data collection; the danger of selection and information biases in hypothesis-directed study design and implementation; and the complete inadequacy of p values in identifying significant relationships. As a solution to these challenges, we propose an end-to-end analysis workflow using the MUltivariate analysis and VISualization (MUVIS) package within R statistical software. MUVIS consists of a comprehensive set of statistical tools that follow the basic tenet of unbiased exploration of associations within a dataset. We validate its performance by applying MUVIS to data from the Yazd Health Study (YaHS). YaHS is a prospective cohort study consisting of a general health survey of more than 30 health-related measurements and a questionnaire with over 300 questions acquired from 10050 participants. Given the nature of the YaHS dataset, most of the identified associations are corroborated by a large body of medical literature. Nevertheless, some more interesting and less investigated connections were also found which are presented here. We conclude that MUVIS provides a robust statistical framework for extraction of useful and relevant information from medical datasets and their visualization in easily comprehensible ways.