Putting CATs and item banks to work: How to construct predictive and sensitive PROMIS screeners for use in ambulatory oncology

Abstract

Purpose: Incorporating PROMs into routine clinical care comes with numerous challenges, including pressure to move patients through office visits with brief, efficient assessments. The purpose of our study was to select and evaluate very short PROMIS screeners with full-bank simulations and predictive accuracy analysis.

Methods: We identified candidate pairs of items for PROMIS Anxiety, Depression, Pain Interference, Fatigue, and Physical Function, on the basis of frequently selected CAT items at both normal and clinical thresholds. We then simulated full-bank responses for 10,000 patients with a T-score mean of 50 in Fatigue, Depression, Pain Interference, and Anxiety, but 45 for Physical Function. We scored candidate 2-item short-form pairs, but also evaluated a 1-item screener and a 4-item fixed CAT for comparison purposes. Clinical thresholds were set at 1 SD in the direction of worse health; diagnostic values were computed based on agreement with full-bank T-scores. We aimed for sensitivity and positive predictive values (PPV) above 0.80, considering also diversity and redundancy of content.

Results: Average sensitivity of our chosen 2-item pairs across the five domains was 0.84 (range 0.80-0.89), while the average PPV was 0.82 (0.80-0.85). While the 1-item screeners for Anxiety and Depression performed poorly, they met or approached our criteria for Pain Interference, Fatigue and Physical Function.

Conclusion: Five 2-item customized PROMIS forms are available for use in routine clinical care, such as distress screening in oncology. Our study provides a useful complement to psychometric criteria, helping to minimize misclassification of patients in need during screening.

Keywords: Cancer; Clinical screening; Health-related quality of life; Item banks; PROMIS; Short-form development.