Patient characteristics

When analysing a randomized trial, generally the first table to be presented is a table of pre-treatment or patient characteristics. This is presented for two reasons, first, to show the population of patients that have entered into the trial and second to 'display' how effective the randomization has been in achieving balance on these known patient characteristics. For the BA06 bladder cancer trial [3] a subset of the table of patient characteristics are shown in Table 9.6.

It is important that in such a table, for each characteristic each randomized patient is accounted for, even if they have missing data. For example, ten patients have missing data for size of tumour (four in the chemotherapy arm, six in the no chemotherapy arm). This table shows that all these characteristics are well balanced between the two groups. It is of interest to note that the median age of the patients in this trial is sixty four, which is perhaps lower than the population of patients with this disease. This is probably because the nature of the new treatment (chemotherapy) dictates that patients have to be relatively fit, indicated by their generally good performance status, and as a consequence this group of patients are somewhat younger than the population as a whole.

Sometimes it is common to see a hypothesis test being performed to assess whether there is a 'statistically significant' imbalance in patient characteristics between the randomized arms. Further, it is sometimes suggested that any observed 'imbalances' should

Table 9.6 Subset of patient characteristics for the BA06 bladder cancer trial. Reprinted with permission from Elsevier Science (The Lancet, 1999, 354, 533-40)

Patient characteristic

Chemotherapy (n = 491)

No chemotherapy (n = 485)

T-category

T2

1 69 (34%)

1 65 (34%)

T3

285 (58%)

282 (58%)

T4a

37 (8%)

38 (8%)

Age

Median

64

64

Sex

Male

433 (88%)

430 (89%)

Female

58 (12%)

55 (11%)

Histological grade

G1

6 (1%)

2 (0.4%)

G2

52 (1 1 %)

61 (1 3%)

G3

433 (88%)

421 (87%)

Not known

0

1

Tumour size (cm)

<2.5

82 (17%)

93 (19%)

2.6-5.0

306 (62%)

315 (65%)

5.1-6.9

88 (18%)

63 (13%)

^7.0

11 (2%)

8 (2%)

Missing

4

Median 4 4

be allowed for in a stratified analysis. For example, in Table 9.6 interest may focus on whether there is an imbalance in the size of the tumour across the two arms of the trial. As Senn [11] and others have pointed out, it is inappropriate to perform a significance test for the number of reasons. Firstly, in doing this standard advice for hypotheses tests is ignored. In particular, null and alternative hypotheses are not clearly stated, a lack of significant difference between arms is treated as a 'proof' of equivalence and no adjustment is made for multiple testing. Further, if randomization was carried out 'appropriately,' then it is almost impossible to address these issues in any sensible way, and it transpires that all that is tested is that the process of randomization was concealed in the way planned, rather than whether randomization achieved the desired effect of 'balance.' Thus it is inappropriate to perform such tests of significance. Rather, the best use of these characteristics is set out in Box 9.3.

0 0

Post a comment