Score equivalence of paper-, tablet-, and interactive voice response system-based versions of PROMIS, PRO-CTCAE, and numerical rating scales among cancer patients

Journal of Patient-Reported Outcomes

Table 2 Differential item functioning by modes of administration for anxiety

Anxiety	McFadden pseudo R² change (\({R}_{2}^{2}-{R}_{1}^{2}\))	McFadden pseudo R² change (\({R}_{3}^{2}-{R}_{2}^{2}\))	McFadden pseudo R² change (\({R}_{3}^{2}-{R}_{1}^{2}\))
I felt fearful	0.0001	0.0007	0.0005
I found it hard to focus on anything other than my anxiety	0.0041	0.0051	0.0010
My worries overwhelmed me	0.0019	0.0028	0.0008
I felt uneasy	0.0007	0.0015	0.0008
I felt nervous	0.0017	0.0023	0.0006
I felt like I needed help for my anxiety	0.0015	0.0029	0.0014
I felt anxious	0.0002	0.0004	0.0001
I felt tense	0.0011	0.0015	0.0003
(NRS) describe the level of anxiety on average	0.0005	0.0005	0.0000
(PRO-CTCAE) How often did you feel anxiety?	0.0049	0.0058	0.0009
(PRO-CTCAE) What was the severity of your anxiety at the WORST?	0.0025	0.0025	0.0000
(PRO-CTCAE) How much did anxiety interfere with usual/daily activities?	0.0009	0.0017	0.0008

A base model (model 1) posits that only the trait level predicts responses. A second model (model 2) has both trait level and group as independent variables. If model 2 predicts item responses statistically significantly better than model 1 (i.e., McFadden pseudo R² change (\({R}_{2}^{2}-{R}_{1}^{2}\)) ≥ 0.02), then there is uniform DIF. In uniform DIF, DIF has a consistent impact across trait levels. If the model that includes an interaction term between trait and group (model 3) fits significantly better than model 2 (i.e., McFadden pseudo R² change (\({R}_{3}^{2}-{R}_{2}^{2}\)) ≥ 0.02), then the impact of DIF varies by trait level (nonuniform DIF). If model 3 fits significantly better than model 1 (i.e., McFadden pseudo R² change (\({R}_{3}^{2}-{R}_{1}^{2}\)) ≥ 0.02), there is overall or total DIF