Abstract
Objectives:
Patient reported outcomes (PROs) serve as a means of measuring improvement and quality of care. Legacy PROs rely on a list of questions that have had to demonstrate accuracy, responsiveness, and validity in testing for intended measurements. While certain legacy PROs such as the International Knee Documentation Committee (IKDC) survey have demonstrated these properties well, a lengthy PRO creates a time burden on patients, making patient adherence and completion a challenge. In recent years, PROs such as the Patient Reported Outcomes Measurement Information System (PROMIS) Physical Function (PF) and Pain Interference (PI) surveys have been developed which leverage computer adaptive testing that produce equivalent accuracy, responsiveness, and validity of legacy PROs, but use only 4-12 questions per survey. This results in significant reduction in time to complete. As these new PROs are now being adopted, the ability to compare outcomes to prior studies that relied on legacy PROs is necessary. While prior studies have examined correlation between legacy PROs and PROMIS computer adaptive tests, no studies to date have developed effective prediction models utilizing PROMIS surveys to create an IKDC index score.
The objective of this study was to develop a predictive model utilizing PROMIS PF and PI to effectively recreate IKDC survey scores.
Methods:
The Military Orthopaedics Tracking Injuries and Outcomes Network (MOTION) database is a prospectively collected repository of patient reported outcomes and intraoperative variables. As part of inclusion in MOTION, research patients who underwent knee surgery were asked to complete the IKDC as well as the PROMIS PF and PROMIS PI at varying time points. This cohort of patients that completed both IKDC and PROMIS scores were included in the present analysis. Nonlinear multivariable predictive models using both Gaussian and beta distributions were created to establish an IKDC index score, which was then validated using leave-one-out techniques and minimal clinically important difference (MCID) analysis.
Results:
A total of 1,011 knee patients (Age: 37.7±14.4 years; 30% Female) completed the IKDC, PROMIS PF, and PROMIS PI providing 1,618 complete observations. The algorithms for both the Gaussian and beta distribution were strongly validated to predict the IKDC (Table 1). The MCID for IKDC was 27.0 (95% confidence intervals: 15.0-39.7) whereas the IKDC-index MCIDs for the Gaussian and beta distribution models were both 13.3 (95% confidence intervals 2.7-27.9), suggesting that the derived IKDC-index is effective and can reliably re-create IKDC scores. The surface plots of this nonlinear multivariable model also confirm the necessity that nonlinear prediction is necessary for effective modeling of legacy PRO scores (Figures 1 & 2).
Conclusions:
PROMIS PF and PI predictive models are able to approximate the IKDC score within 9.3-10.0 points. Given the 27.0 point minimally clinically important difference for the IKDC survey in this cohort, the results of this study can be used to compare PROMIS PF and PI scores to prior IKDC data by creating an IKDC index score. Moreover, repeated use of the IKDC-index from PROMIS PF and PI allows for a lower MCID than using the conventional IKDC survey. PROMIS PF and PI scores can be substituted in both clinical and research settings to reduce patient time burden, increase completion rates, and still create data that can effectively be compared with studies utilizing legacy IKDC scores.