Clinical Trials

Clinical Trials: A Practical Guide ■❚❙❘Table 6. Number (%) of patients in NYHA classes by treatment in a cardiovascular trial.NYHAActive drugTreatmentPlaceboClass I 276 (7.6) 267 (7.5)Class II 1946 (53.8) 1932 (54.4)Class III 1334 (36.9) 1280 (36.0)Class IV 58 (1.6) 74 (2.1)Total 3614 (100.0) 3553 (100.0)Comparing endpoints with more than two categoriesby treatment groupsIn some clinical trials, a discrete efficacy or safety point has more than twocategories (ordered or unordered). For example, in cardiovascular trials dealingwith patients with heart failure, the New York Heart Association (NYHA)classification (four classes: I, II, III, and IV) is often used as a measure of apatient’s functional capacity with respect to heart failure. We might also need tocompare more than two groups (eg, drug 1, drug 2, and placebo), which givesa r × c contingency table (r = rows and c = columns). This can be analyzed bya chi-squared test in a similar way to a 2 × 2 contingency table [2,3].Table 6 presents some data on NYHA class, measured at the end of acardiovascular trial. This is a 4 × 2 contingency table with four rows standingfor the four NYHA classes and two columns representing the two treatments.For this table, the chi-squared test can be used to assess whether the percentagedistributions in NYHA class are statistically significantly different between thetwo treatments. Following the procedures described above for a 2 × 2 contingencytable, we obtained χ 2 = 2.74. For an r × c contingency table, the number of thedegrees of freedom is (r – 1) × (c – 1) = (4 – 1) × (2 – 1) = 3. From Table 4,we have χ 2 = 7.81. Since 0.05,3 χ2 < χ 2 , P > 0.05. We can therefore conclude that0.05,3there is insufficient evidence to suggest that the two population percentagedistributions of NYHA class are significantly different.It should be noted that the chi-squared test compares the distribution of fourNYHA classes, not taking into account the ordinal nature of NYHA class.For an r × c contingency table with an ordered categorical variable like NYHAclass, we can also use the nonparametric Wilcoxon rank sum (Mann–Whitney)test to assess whether there is a statistically significant difference in the outcomedistribution between treatment groups (see Chapter 19). In the case of the NYHAdata in Table 6, the Wilcoxon rank sum test yields a P-value of >0.05, suggesting231