patients who had little or no evidence of retinopathy at baseline. Retinopa-thies are
abnormalities of the retina that sometimes occur among diabetic patients and can result in
substantial losses of vision in advanced stages. Patients were randomized to either Sorbinil,
an aldose-reductase inhibitor, or placebo and were seen at 1 year and then every 9 months up
to 48 months after randomization. In addition, all subjects had a final visit at the end of the
trial (max=56 months). Sixteen of the patients provided no follow-up. The primary endpoint
of the trial was based on change in retinopathy severity level from baseline to maximum
follow-up (i.e., severity level at maximum follow-up - severity level at baseline). An ordinal
grading scale was used to evaluate change: 2+ levels better, 1 level better, no change, 1 level
worse, ..., 5+ levels worse. The outcome data by treatment group are given in Table 1.
Table 1. Outcome Data for the Sorbinil Retinopathy Trial (n=481) change in retinopathy level
Better
Worse
Group
2+
1
No
1
2
3
4
5+
Total
Levels
Level
change Level
Levels
Levels
Levels
Levels
Placebo 5
17
84
59
37
18
9
14
243
Sorbinil 4
21
97
50
22
16
14
14
238
The primary outcome for the study was worsening by 2 or more levels.
a) What test can be used to compare the two treatment groups on the primary endpoint?
b) Implement the test in Problem a), and provide a p-value (two-tailed).
A more efficient method of analysis would leave the change in retinopathy level in its raw
form without grouping the data but would take into account the ordinal nature of the change
scores.
c) What test can be used to compare the two groups if this more efficient method is used?
d) Implement the test in Problem c), and provide a p-value (two-tailed).
e) Reanalyze the data in Table 1 using the Wilcoxon rank sum test.
f) Compare your results in Problems c) and d). Which do you think is a better method of
analysis?