Nonlinear Models And Computer Adaptive Testing Can Decrease Orthopedic Patient Survey Burden

Abstract

The National Institutes of Health Patient Reported Outcomes Measurement Information System (PROMIS) computer adaptive tests (CATs) assess patient symptom level in fewer questions than legacy patient reported outcomes (PRO). While there are benefits to CATs, clinicians commonly desire legacy PROs where they have an existing knowledge base. The American Shoulder & Elbow Surgeons score (ASES) is commonly used to assess shoulder function; it is unknown if PROMIS measures can predict ASES scores. PURPOSE: Design a nonlinear model using PROMIS CATs to predict ASES scores. METHODS: Military Health System beneficiaries who underwent a shoulder surgery and consented to allow their clinical data be used for research (n=897) completed the ASES, PROMIS Physical Function, and PROMIS Pain Interference at varying time points, providing 1,471 complete observations. PROMIS Physical Function and Pain Interference surveys were modeled as these are the theoretical constructs the ASES evaluates. For the prediction models, PROMIS CAT scores were re-weighted based on the standard error of the score, reflecting the confidence in the score, via inverse-variance reweighting. A beta distribution Generalized Additive Mixed Model (GAMM) accounted for multiple observations while incorporating nonlinear interactions into the model. The model’s predictive quality was assessed via four-fold cross-validations evaluating the following metrics between predicted vs true data: 1) Pearson correlation coefficients, 2) linear regression R2 values, and 3) root mean square error (RMSE). RESULTS: The GAMM predictive model (Figure 1) had the following characteristics: Pearson coefficient = 0.74-0.76, R2 = 0.55-0.58, and RMSE = 13.4-14.2. CONCLUSIONS: PROMIS CATs, in conjunction with nonlinear predictive modeling, can reliably predict legacy PRO scores. PROMIS CATs reduce patient question burden and can provide clinicians the information they are accustomed to receiving from legacy PROs.