Power of the statistical graphs - an example of different ways for a Bar Chart

For the same set of data from the clinical trial, it is important to choose the appropriate statistical analysis methods. It is also important to choose the appropriate plot to display and visualize the data. The example below demonstrate the power of the statistical graphs (even the simple bar charts). 

A clinical trial is designed as a randomized, double-blinded, parallel three-arm study to evaluate the effect of high-dose and low-dose of an investigational drug in comparison with the Placebo. The outcome measure is NYHA Functional Class with four grades (I, II, III, and IV). At each post-baseline visit, function class is improved if it is shifted at least one grade toward the lower end, and functional class deteriorates if it is shifted at least one grade toward the high end. Suppose that at the end of the study, the outcome of the last visit can be organized as the following: 

	High-Dose (n = 200)	Low-Dose (n = 198)	Placebo (n = 202)
Improved	49 (24.5%)	37 (18.7%)	26 (12.9%)
No Change	139 (69.5%)	142 (71.7%)	148 (73.3%)
Deteriorated	12 (6.0%)	19 (9.6%)	28 (13.9%)

Both the dose (high-dose, low-dose, and placebo) and outcome (improved, no change, deteriorated) are considered to be the ordinal data. The statistical test with CMH gives a p-value of 0.0010 indicating the association between the dose groups and the outcome measures. 

The interesting thing is the data visualization - there are different ways to plot the data in the table above. What is the best way to plot the results in the table above?

If we are only interested in the proportion of subjects who have 'improved' functional class:

If we are only interested in the proportion of subjects who have 'deteriorated' functional class:

If we are interested in both the proportion of subjects who have 'improved' functional class and the proportion of subjects who have 'deteriorated' functional class: 

The bar chart can be arranged by outcome (improved and deteriorated) and the dose group (high-dose, low-dose, and placebo)

or the bar chart can be arranged by the dose group (high-dose, low-dose, and placebo) and then the outcome (improved and deteriorated). 

A stacked bar chart can be used to include all three outcome categories (improved, no change, and deteriorated):

The best approach is to place the outcome of 'improved' and 'deteriorated' functional class on the same vertical bar, but one above the 0-line and one below the 0-line. This approach assigns the 'improved' category as a positive value and the 'deteriorated' category as a negative value. 

These bar charts can be created using existing software such as SAS/Graphs, Microsoft Excel, and Graphpad Prism. It seems to be easier to use SAS to manipulate the data or obtain the aggregate data and then use Excel or Prism to create the charts. Prism is better and easier to use for creating charts for publications.