Exploratory data analysis of a clinical study group: Development of a procedure for exploring multidimensional data

Area of Science:

• Clinical data analysis
• Exploratory data analysis
• Biostatistics

Background:

• Understanding data structure is crucial for forming scientific hypotheses.
• Exploratory data analysis (EDA) methods reveal data structure.
• Selecting appropriate EDA tools for clinical datasets can be challenging.

Purpose of the Study:

• To present a comprehensive set of tools for exploratory analysis of clinical data.
• To perform a case study analysis on a dataset of 515 elderly patients.
• To demonstrate the utility of EDA in identifying biologically meaningful patient subgroups.

Main Methods:

• Robust data normalization
• Outlier detection using Mahalanobis (MD) and robust Mahalanobis distances (rMD)
• Hierarchical clustering with Ward's algorithm
• Principal Component Analysis (PCA) with biplot vectors

Main Results:

• The dataset of 515 elderly patients (PolSenior project) with over 40 attributes was analyzed.
• Two initial clusters were identified, separated by sex hormone attributes.
• Male patients partitioned into five subgroups, with two linked to diabetes and hypogonadism.
• The female patient set was more homogeneous, showing no pathological subgroups.

Conclusions:

• Outlier detection methods (MD and rMD) can assess dataset heterogeneity beyond identifying outliers.
• The proposed EDA procedure effectively identifies and visualizes biologically relevant patient subgroups.
• This approach facilitates the formation of detailed scientific hypotheses and research questions from clinical data.