The PROPr can be measured using different PROMIS domain item sets.

Area of Science:

• Health Outcomes Research
• Psychometrics
• Item Response Theory (IRT)

Background:

• The Patient-Reported Outcomes Measurement Information System (PROMIS) Preference Score (PROPr) is derived from health assessments within the PROMIS framework.
• Item Response Theory (IRT) enables measurement of PROMIS health domains using various item subsets.
• This study investigates if PROPr can also be accurately estimated using different item subsets.

Purpose of the Study:

• To empirically and computationally test the assumption that PROPr can be estimated from different item subsets.
• To evaluate the impact of using varying numbers of items on PROPr estimation accuracy and agreement.
• To determine if a reduced item set (2 items) adequately represents the full PROPr score.

Main Methods:

• Empirical analysis: Estimated PROMIS Pain Inference (PI) scores from 3 item subsets (4, 2, and 10 items) in 199 cancer patients.
• Statistical comparison: Utilized intraclass correlation coefficients (ICC) and Bland-Altman (B-A) plots to assess differences and agreement between score estimates.
• Simulation: Modeled item responses using IRT to compare PROPr estimates from simulated item banks with empirical results.

Main Results:

• Empirical comparison showed smaller bias and high agreement (ICC > 0.98) for PROPr estimated from 4 vs. 10 items compared to 4 vs. 2 items.
• Simulation results indicated a slight increase in mean PROPr when using 2 items (PROPr_sim) compared to 4 items (PROPr_4), with good agreement (ICC = 0.95).
• Both empirical and simulation approaches demonstrated that different item subsets yield comparable PROPr estimates, with minimal bias when using fewer items.

Conclusions:

• The study confirms that diverse item subsets can effectively estimate the PROMIS PI for PROPr calculation.
• Reducing the item set to 2 items per domain does not substantially alter the average PROPr estimate.
• While average estimates are similar, agreement varies, highlighting the need to consider individual comparisons and the full spectrum of scores.