Data Treatment Process

• It was counted the variables with null values and the percentage of null in each variable.
• The variables with more than 75% of null values were deleted for the analysis. They were considered like garbage data.
• The remaining variables were filled out based if they were numerical or categorical. The numerical variables were filled out with the mean. The categorical variables were filled out with a K-Nearest Neighbors Imputer.
• The categorical variables were transformed from text into numbers.
• It was analyzed if they data set was unbalanced or not. Roughly 60% data from pre-pandemic time and 40% from pandemic time. It was considered like a balanced dataset.
• It was determined the coefficient of linear correlation.
• It was generated a scatter plot for all variables with a hue to discard the Simpson paradox. It was not obtained any pattern in the data.
• It was carried out a one-dimensional analysis of the variables. It was analyzed the tumor’s size, presentation mode, the age of the patient, gender, an underlying cirrhosis disease, and a previous known cirrhosis.
• It was carried out a bidimensional analysis. The pairs of variables analyzed were mode of presentation vs age, treatment_grps vs survival_fromMDM, etiology vs age, etiology vs size, size vs alive_dead, and size vs age.

Bidimensional Analysis

• The size vs the age of the patient did not seem to follow any pattern.

• The size of tumors is sightly important to liver or not. The median of tumor size for live is roughly 25 mm, but the tumor size for dead is under 50 mm.