Introduction
Patient-reported outcomes measures are becoming increasingly utilized to help guide clinical decision for orthopaedic surgical patient care. Patient Reported Outcome Measurement Information System (PROMIS) domains are validated computer adaptive testing metrics that are employed in research and clinical practice to assess a wide array of outcome measures. However, meaningful clinical interpretation and utilization of patient-reported outcomes scores remains difficult. Previous research has focused on the clinical utility of single domains. Yet, grouping patients with similar preoperative scores on various measures may be more clinically useful than individual scores. The purpose of the study was to determine if grouping shoulder surgery patients into clusters profiles based on preoperative PROMIS scores has prognostic value for two-year postoperative outcomes.
Methods
533 of 761 patients (70%) undergoing elective shoulder surgery at a single urban academic medical center were prospectively enrolled in an orthopaedic registry and completed two-year follow up. Patients were administered questionnaires to assess demographic information and patient-reported outcomes preoperatively and at two years postoperatively. Questionnaires were administered for PROMIS, American Shoulder and Elbow Surgeons Shoulder Score (ASES), Marx Activity Rating Scale (MARS), Tegner Activity Scale (TAS), Musculoskeletal Outcomes Data Evaluation and Management Scale (MODEMS) Expectation domain, and surgical satisfaction (SSQ-8). A k-means cluster analysis was used to identify preoperative PROMIS clusters. Descriptive Chi-square or Wilcoxon ranked sum tests statistics were utilized for bivariate analyses, and least-squares multivariate analysis was used to identify if cluster groupings were independent predictors for two-year and change outcomes.
Results
Clusters based on four preoperative PROMIS domains (Physical Function, Pain Interference, Fatigue and Anxiety) were used to classify patients in 4 profiles: “Normal Function”, “Mild Impairment”, “Impaired without Distress” and “Impaired with Distress”. More impaired cluster status was associated with higher rates of arthroplasty, female gender, preoperative opioid use, lower income, greater ASA score, among other sociodemographic and operative factors. Worse cluster status was associated with worse scores, but better improvement from baseline for most two-year patient reported outcomes. More impaired clusters were also associated with greater probability of achieving MCID on PROMIS Physical Function, Fatigue, and Depression. Multivariate regression analysis revealed better preoperative PROMIS cluster was predictive of better two-year outcomes across nearly all patient reported outcomes while the most impaired cluster was independently predictive of worse two-year outcomes for the same patient reported outcomes. Worse preoperative cluster status was also independently predictive of greater improvement in most outcomes except ASES.
Conclusion
Clusters based on only four PROMIS domains have significant prognostic value for patient outcomes two years following shoulder surgery. Better preoperative cluster predicts better outcomes, while a worse preoperative cluster predicts a worse outcome. However, all preoperative clusters still saw significant improvement from baseline, with the worst preoperative cluster demonstrating the greatest improvement. Therefore, a worse preoperative cluster status should not be considered a contradiction to surgical treatment. Grouping shoulder surgery patients may be superior to utilizing single PROMIS domains in the clinical setting and may serve as a guide to inform individual expectations prior to shoulder surgery.