Philip Buonadonna/Jason Hill
IS-247, Assignment 4
Oct. 10, 2000
"Efficacy of Treatments for Chapped Lips" (source)
This dataset was collected from a study to evaluate different methods of treatment for chapped lips. Patients participating in the study were divided into three treatment groups: stick balm, liquid balm and a control group (no treatment). Each patient's progress was tracked on a daily basis for one week. On each day, the patients took note of their perceived improvement. On days 1, 3, 5 and eight, an "expert grader" evaluated their treatment and took water loss measurements of the affected areas.
Variables in the dataset included the following (numbers in parenthesis indicate a numeric value assigned to nominal data):
The following variables were noted by the patient on a daily basis.
1) The patients receiving either form of liquid or stick balm treatments will have significantly better results compared to those that receive no treatment at all. Specifically, patients who received treatment will have better expert grader assessments, more water loss (i.e. "wetter" lips), less dryness, soreness flaking or cracking, and higher relief factors.
2) The stick and liquid balm treatments will have equivalent results through the study. The only difference between these treatments is that one comes in a waxy base (stick) while the other does not. Given this, the effectiveness of each treatment should be the same.
3) Patients using liquid treatments will use more of the product than those using liquid treatments. The waxy nature of the stick treatments will limit the amount applied per use and prevent over dispensing the product.
Two tools were used to help analyze the data: Inxight's Eurkea (table lens) and XMDV (parallel axis). The results of the entire dataset can be seen in the table lens (Figure 1). NOTE: Due to the density of the table, only the first letter of each data category appears. The categories appear in the same order as listed above. The last 5 categories have one column for each day.
Figure 1: Table Lens
From the overall visualization, it is immediately apparent that patients using treatments observe better relief on all days and less dryness, soreness, flaking and cracking between day 1 and day 7. Also, the expert grader assessments show improved results for the treated group over the control group. This would tend to support the first hypothesis regarding the treated vs. untreated groups.
Also, the visualization data does not show any noticeable difference in the results between the stick-balm and liquid-balm treatment groups. This supports the second hypothesis that the different treatment methods produce the same results.
Support for the third hypothesis is found in the table lens and highlighted using the parallel axis analysis of XMDV (Figure 2). NOTE: Due to tool limitations, the five daily categories are not portrayed on the parallel axis visualization. Using the brush tool to highlight the highest usage cases, we observed that all of these patients were using liquid-balm treatment and that their results were no better or worse than the stick-balm group. The stick-balm treated patients used comparatively little of the product.
Figure 2: Analysis of balm usage.
The data visualization also revealed an unexpected result. The water loss (WL) measurements did not coincide with the other treatment results for any group. Figure 3 is a parallel axis view with the highest WL measurements taken on day 8 brushed. The WL data points do no converge to a particular grader assessment, usage amount or even the treatment group! The table lens also confirms the discontinuity with the patients daily notes. Since the divergence is evident in all the data points, it suggests that the WL analysis is not an effective measure of the severity of chapped lips.
Figure 3: Analysis of Water Loss measurement and overall treatment.
The visualizations also showed no relation between age, height, weight or sex and the effectiveness of the treatment used. This was an expected result.
The Inxight tool was the most useful for analyzing this dataset. The ability of the table lens to present a large amount of data and the overall robustness of the vendor's software made it the tool of choice. The data could easily be sorted by a particular category to search for relations and the tool had no problem digesting the size of the dataset. The only feature that we found missing was the ability to select and hide a group of data categories at the same time. With a data set of this size, individually selecting and hiding categories became tedious and sometime error prone.
The parallel axis view of XMDV supplemented the table lens by permitting better trend analysis for a collection of data points from a particular category. The ability to brush a set of points and view the data trace was extremely useful. The robustness of the software left much to be desired, however. The tool could only accept about 15 data categories before crashing. Also, the need to pre-specify minimum, maximum and bin-size was tedious and prone to error.