Population-specific optimization of PROMIS® short forms using constrained mathematical programming

Background Computerized adaptive testing (CAT) offers efficient and precise administration of Patient-Reported Outcomes Measurement Information System (PROMIS) instruments. However, its implementation is often hindered by costs and logistical barriers. PROMIS short forms are more feasible in clinical settings but typically lack the measurement precision of CAT and, like CAT, may not have optimal content validity for specific populations. This study evaluated whether mathematical programming can be used to construct optimized, fixed-length PROMIS short forms that approximate CAT precision while allowing explicit control over content validity. Methods We developed an optimization framework that selects items to minimize expected measurement error variance across a target score distribution, subject to constraints on test length and content balance. Using the PROMIS Physical Function item bank, we compared optimized 10-item forms to CAT, standard PROMIS short forms, and random forms across simulated and empirical datasets. A constrained, highly optimized scenario was also evaluated to explore the framework's ability to simultaneously optimize measurement precision and content validity by enforcing item coverage across ICF chapters. Results Optimized short forms consistently outperformed standard and random forms and closely approximated CAT performance, particularly in information-sparse regions. In the constrained optimization scenario, content-balanced forms achieved marginal reliabilities above 0.92, demonstrating that precision and content validity can be jointly optimized without substantial performance loss. Conclusions The mathematical programming framework we explored in this study makes it possible to create fixed-length PROMIS short forms that remain highly precise even when explicit content constraints are applied. These findings show that this approach may be useful to develop targeted short forms with high content validity that still yield highly reliable scores and can more easily be applied in practical research or daily care settings compared with CAT.