Defining responders to therapies by a statistical modeling approach applied to randomized clinical trial data

Table 2 List of the five best models obtained defining the models on the training set and testing them on the validation set, ordered according to the p value for treatment by score interaction on the merged training and validation datasets

Model	p* training set	p* validation set	p* training + validation set	Area under the AD(q) curve
Response score 1 = − 0.38 × Age3 + 0.65 × sex + 0.39 × Rel − 0.002 × NBV − 0.20 × Gd+	0.026	0.004	0.00027	0.358
Response score 2 = 0.61 × sex + 0.37 × Rel − 0.002 × NBV − 0.23 × Gd+ − 0.02 × age + 0.02 × EDSS	0.037	0.003	0.00031	0.368
Response score 3 = 0.05 × EDSS4 + 0.61 × sex + 0.37 × Rel − 0.002 × NBV − 0.22 × Gd+ − 0.02 × age	0.040	0.004	0.00032	0.368
Response score 4 = 0.61 × sex + 0.37 × Rel − 0.002 × NBV − 0.22 × Gd+ − 0.02 × age	0.042	0.003	0.00033	0.369
Response score 5 = 0.63 × sex + 0.36 × Rel − 0.01 × age + 0.02 × EDSS	0.037	0.004	0.00044	0.356

Area under the AD(q) curve represents the curve generated by plotting the cumulative distribution of patients ranked by individual treatment response score and the overall treatment effect relative to a given proportion of patients. The lower is the curve, the higher the heterogeneity of treatment effect. Age3: Age3 = 1 if age < 20 years, Age3 = 2 if age in the range 20–50, Age3 = 3 if age ≥ 50 years; sex = 1 if sex = male and sex = 0 if sex = female, EDSS4 = 0 if EDSS 4, EDSS4 = 1 if EDSS ≥ 4
Rel number of relapses in the previous year; NBV normalized brain volume, in cubic centimeter; Gd+ gadolinium-enhancing; EDSS Expanded Disability Status Scale; age age in years
*p for treatment by score interaction

ISSN: 1741-7015