The PROPr can be measured using different PROMIS domain item sets

The Patient-Reported Outcomes Measurement Information System (PROMIS) Preference Score (PROPr) is estimated from descriptive health assessments within the PROMIS framework. The underlying item response theory (IRT) allows researchers to measure PROMIS health domains with any subset of items that are calibrated to this domain. Consequently, this should also be true for the PROPr. We aimed to test this assumption using both an empirical and a simulation approach.

Empirically, we estimated 3 PROMIS Pain inference (PI) scores from 3 different item subsets in a sample of n=199 cancer patients: 4 PROMIS-29 items (estimate: θ₄), the 2 original PROPr items (θ₂), and 10 different items (θ₁₀). We calculated mean differences and agreement between θ₄, and θ₂ and θ₁₀, respectively, and between their resulting PROPr₄, PROPr₂, PROPr₁₀, using intraclass correlation coefficients (ICC) and Bland-Altman (B-A) plots with 95 %-Limits of Agreement (LoA). For the simulation, we used the IRT-model to calculate all item responses of the entire 7 PROPr domain item banks from the empirically observed PROMIS-29+cognition θ. From these simulated item banks, we chose the 2 original PROPr items per domain to calculate PROPr_sim and compared it to PROPr₄ again using ICC and B-A plots.

θ₄ vs θ₁₀ showed smaller bias (-0.012, 95 %-LoA -0.88;0.85) than θ₄ vs θ₂ (0.025, 95 %-LoA -0.95;1.00. ICC>0.85 (p<0.001) in both θ-comparisons. PROPr₄ vs PROPr₁₀ showed lower bias (0.0012, 95 %-LoA -0.039;0.042) than PROPr₄ vs PROPr₂ (-0.0029, 95 %-LoA -0.049;0.044). ICC>0.98 (p<0.0001) on both PROPr-comparisons. Mean PROPr_sim was larger than mean PROPr₄ (0.0228, 95 %-LoA -0.1103; 0.1558) and ICC was 0.95 (95 %CI 0.93; 0.97).

Different item subsets can be used to estimate the PROMIS PI for calculation of the PROPr. Reduction to 2 items per domain rather than 4 does not significantly change the PROPr estimate on average. Agreements differ across the spectrum and in individual comparisons.