OBJECTIVES: A common concern in using RWD is the heterogeneity of practices treating patients, and corresponding impact on patient outcomes. Here we describe the association of practice-level characteristics and real-world overall survival (rwOS) in multiple myeloma (MM) using nationwide community site data.

METHODS: This retrospective study used the US-based Flatiron Health electronic health record-derived de-identified database. We included patients diagnosed with MM between 1/1/2017-6/1/2022, with evidence of treatment at a community practice that was established prior to 2017. Descriptive statistics were summarized at the practice-level, including case-load (patient-physician ratio; numbers of visits, physicians, patients with cancer, and with MM), diversity (% Black patients, with Medicaid, low socioeconomic status, and rurality), location, and clinical factors (% received clinical study drugs, genetic testing, transplant and R-ISS workup). Mixed-effect Cox regression was performed to assess the association between practice-level characteristics and rwOS from first-line treatment, adjusting for patient-level demographic and clinical factors (conditional hazard ratio [cHR] and 95% confidence interval [CI]).

RESULTS: Among 4,552 patients with MM from 136 practices, wide distribution was observed for all 15 practice-level characteristics. Practices ranged from 1-163 physicians, 0-58% patients with Medicaid, 0-50% clinical study drug recipients, and 0-85% transplant recipients. No practice-level factors were associated with rwOS in univariate analyses, and 13 remained unassociated after adjustment. Lower hazard of death was seen among patients from practices with the highest quintile % transplant recipients compared to the lowest quintile (cHR=0.74; 95% CI: 0.55-1.00), and unexpectedly, among patients from practices with a higher % patients with Medicaid (cHR=0.76; 95% CI: 0.61-0.95).

CONCLUSIONS: In this hypothesis-generating study of patients with MM, we observed a wide variability in practice-level variables across community sites, yet rwOS was not strongly associated with most practice-level characteristics. Future studies in other care and oncological settings are needed, accounting for potential unmeasured confounding and interactions.