Hospital-level patterns of palliative care delivery among patients with advanced cancer: A clustering analysis.

Abstract

89 Background: Palliative care (PC) delivery for patients with advanced cancer varies and may affect end-of-life (EOL) outcomes. Yet, existing structural measures such as number or type of clinician do not correlate with variation in EOL outcomes, with program maturity being a notable exception. Our objective was to use a measure of hospital-level PC delivery patterns (“clusters”) to characterize variation in PC delivery and examine associations with EOL outcomes. Methods: We used 2018-2019 Medicare and PC registry data to identify PC encounters in the last six months of life among decedents with advanced cancer. We characterized each encounter by clinician role (physician or advanced practice provider), clinical function (interdisciplinary team member, non-team specialist, or non-specialist clinician), and care setting (inpatient or outpatient). To identify hospitals with similar approaches to PC delivery, we constructed hospital-level patterns from encounter data and applied a clustering algorithm. We treated the resulting clusters as PC delivery process measures. We focused on clusters sharing the most prevalent pattern. EOL outcomes included advance care planning documentation, hospice enrollment, hospice referral within three days of death, more than one ED visit within 30 days of death, and receipt of life-sustaining therapy within 30 days of death. Structural program characteristics, including program maturity and interdisciplinary team size, were also examined. Group differences were assessed using Kruskal-Wallis tests with Dunn’s post-hoc comparisons and Benjamini-Hochberg correction. Results: Across 272 hospitals, the algorithm identified 21 clusters, each representing a distinct PC delivery pattern. Six shared the most prevalent pattern: consecutive inpatient encounters in which a specialist PC physician served as an interdisciplinary team member (“the consecutive inpatient specialist pattern”), representing 39% of hospitals (105/272). Among these six clusters, membership was not associated with any EOL outcome. Membership was associated with program maturity (𝜒2(5) = 12.943, p = 0.02) but not interdisciplinary team size. Post-hoc comparisons revealed differences: hospitals that expanded to both an inpatient and outpatient specialist pattern had greater program maturity than those whose inpatient care transitioned to a non-specialist (Z = -3.1, p = 0.01) or those with only the consecutive inpatient specialist pattern (Z = 2.3, p = 0.04). Conclusions: These findings suggest that hospital clusters, defined by PC delivery patterns (process measures), are associated with program maturity (structural measure) but not EOL outcomes. Since most structural measures do not explain EOL outcome variation, delivery patterns offer an alternative basis for grouping hospitals, one grounded in actual PC delivery rather than program structure.