Development and evaluation of patient reported outcomes-based performance measures during chemotherapy.

Abstract

e19175 Background: As value-based oncology care models become increasingly common, there is need for reliable and valid patient-reported outcome performance measures (PRO-PMs) to assess quality of symptom control at practices. We propose a method for combining PRO items across multiple symptoms into a single reliable and valid summary PRO-PM metric. Methods: Symptom surveys were administered to 607 patients at 6 cancer centers, between 5-15 days after starting a chemotherapy cycle. Twelve symptoms were elicited, each scored from 0-4, with higher scores indicating greater severity. Summary PRO-PM metrics were generated by summing scores for symptoms. A 3-item short form (score range 0-12) included nausea, diarrhea, and pain. A 12-item long form (score range 0-48) included short form items plus neuropathy, dyspnea, vomiting, fatigue, constipation, anxiety, sadness, insomnia, and appetite. PRO-PM summary item scores were dichotomized into high and low symptom burden categories. Cut points for dichotomization were compared based on reliability, validity, and prevalence. Reliability was defined as the average of practice-specific signal-to-noise ratios, with an optimal score ≥0.7. Practice-level validity was defined as the Pearson correlation between the proportion of patients with high symptom burden and low self-reported physical functioning. Patient-level validity was estimated as the relative risk (RR) of low physical functioning for patients with high vs. low symptom burden. For prevalence, measures were required to have > 5% or < 95% of patients with high symptom burden. Optimal cut-points were identified via a scatter plot comparing measure reliability and validity. Results: Optimal cut-points were identified for both summary PRO-PM metrics. For the short form, a cut-point at ≥5 had a reliability of 0.56, correlation of 0.14, RR of 2.56, and prevalence of 24.5%. For the long form, a cut point at ≥22 had reliability of 0.54, correlation of 0.56, RR of 2.61, and prevalence of 8.5%. Conclusions: Summary PRO-PM metrics can be generated from multiple PRO symptom items, with optimal cut-point scores determined by comparing reliability and validity. Reliability scores approached but did not meet the recommended 0.7 threshold due to sample size, therefore additional data are currently being collected to further support these cut points.