Reducing Unwarranted Oncology Care Variation Across a Clinically Integrated Network: A Collaborative Physician Engagement Strategy

Abstract

PURPOSE: Addressing unwarranted clinical variation in oncology is a high priority for health systems that aspire to ensure consistent levels of high-quality and cost-effective care. Efforts to improve clinical practice and standardize care have proven challenging. Advocate Physician Partners undertook a patient simulation-based practice measurement and feedback project that was focused on breast and lung cancer to engage oncologists in the care standardization process. METHODS: One hundred three medical oncologists cared for online simulated patients using the Clinical Performance and Value platform, receiving feedback on how their care decisions compared with evidence-based guidelines and their peers. We repeated this process every 4 months over six rounds, measuring changes in quality-of-care scores. We then compared simulated patient results with available patient-level claims data. RESULTS: Over the course of the project, overall quality-of-care scores improved 11.9% ( P < .001). Diagnostic accuracy increased 6.7% ( P < .001) and correlated with improved treatment scores, including a nearly 10-percentage point increase in evidence-based chemotherapy regimens ( P = .009) and a 56% increase in addressing palliative needs for patients with late-stage disease ( P < .001). Unnecessary test ordering declined 25% ( P < .001). We compared these results with available patient data and observed concordance with the metastatic imaging workup order rate for early-stage breast cancer. As unnecessary workups declined in the simulations and became more closely aligned with evidence-based guidelines, we saw similar rates of decline in the patient-level data. CONCLUSION: This study demonstrates that an oncology care standardization system that combines simulated patients with serial feedback increases evidence-based and cost-effective clinical decisions in patient simulations and patient-level data.