Summary

Clusters based on preoperative PROMIS domains has prognostic value in predicting two-year postoperative outcomes in knee surgery patients with better cluster profile predicting better outcomes at two years and worse clusters predicting worse outcomes but greater improvement from baseline. 

Abstract

Introduction

Patient-reported outcomes are important measures of postoperative improvement in orthopaedic surgery. The Patient Reported Outcome Measurement Information System (PROMIS) tools are computer-adaptive testing metrics that are efficient and validated. However, meaningful clinical interpretation remains challenging. In addition, previous studies have strictly looked at the prognostic value of single PROMIS domain metrics even though grouping PROMIS scores may have superior clinical utility. The purpose of this study was to determine if clustering knee surgery patients into four preoperative PROMIS profiles would have prognostic value for two-year postoperative outcomes. We hypothesized that preoperative cluster would be predictive of two-year outcomes.

Methods

488 of 697 (70%) patients undergoing outpatient elective knee surgery at a single urban institution were enrolled in an orthopaedic registry and completed two-year follow up. All patients provided informed consent and were administered questionnaires for PROMIS, International Knee Documentation Committee (IKDC) Score, Marx Activity Rating Scale (MARS), Surgical Satisfaction (SSQ-8), and Completely Better improvement. Demographics were self-reported and electronic medical record review was conducted to record medical and operative history. A k-means cluster analysis was performed to identify the four preoperative clusters from 6 PROMIS domains (Physical Function, Pain Interference, Social Satisfaction, Fatigue, Anxiety, and Depression). MCID for each outcome was predetermined via literature review. Chi-square or Kruskal-Wallis tests were conducted for bivariate analyses. Least-squares multiple linear regression models were analyzed for relevant two-year outcome metrics to identify if cluster group was an independent predictor.

Results

Cluster analysis revealed four distinct groups based off preoperative PROMIS scores. Psychological distress was most significant in determining classification. More impaired subgroups were associated with higher rates of arthroplasty, African-American race, preoperative opioid use, lower income, higher comorbidity index, and other sociodemographic and operative factors (p < 0.05). Better preoperative clusters were associated with higher two-year postoperative scores in all PROMIS domains, IKDC, MARS Knee, and SSQ8 (p < 0.001 for all). Worse preoperative cluster status was associated with higher chance of achieving MCID on all metrics (p = 0.034) except PROMIS Pain Interference, IKDC, and MARS postoperatively. When confounding variables were controlled for via multivariable analysis, better preoperative cluster was predictive of better PROMIS Physical Function and Pain Interference, IKDC scores, and surgical satisfaction (p < 0.05), and worse preoperative cluster was predictive of greater improvement on PROMIS Physical Function and Pain Interference (p < 0.05) but not IKDC.

Conclusion

Empirically-derived preoperative PROMIS clusters have prognostic value in predicting outcomes for knee surgery patients even when associated sociodemographic factors are controlled for. Better preoperative cluster function predicts superior outcomes. While worse preoperative cluster predicts worse outcome, all clusters still significantly improve, so worse preoperative cluster is not a contraindication to surgery. PROMIS Physical Function and IKDC appear to have different utility in assessing postoperative improvement by subgroup. Ultimately, clustering knee patients indicated for knee surgery into preoperative profiles may have superior clinical relevance than single PROMIS-domain analysis. These results are pertinent for individualizing preoperative patient expectations.