An interpretable clustering classification approach for assessing and adjusting hospital service lines

Abstract

Hospital beds are often assigned among several major groups called service lines, each aimed to provide care for patients with similar medical needs such as cancer, musculoskeletal disorders, vascular, surgical, medical, and women and children. Besides the benefits of service lines in streamlining operations, they can lead to unintended issues if they are not assessed and adjusted regularly. These issues include fragmentation and inconsistencies in care, offering overlapping services, uneven resources, and growth, disparities in access to care, imbalanced capacity utilization, and system-wide flow issues. We propose an interpretable clustering classification approach for assessing and adjusting existing service lines regarding size and patient distribution. Our approach is useable in practice as it uses data available during patient admission and generates interpretable rules for patient assignment among service lines. Our results from two academic hospitals suggest the need for further splitting service lines such as internal medicine, general surgery, and neurological disorders. Further, our findings support the idea of providing specialized services such as orthopedic surgery, cardiology, physical medicine, and pregnancy, childbirth and the puerperium. These findings have several practical implications related to patient mix, capacity planning, bed assignment, and hospital administration overall.

Keywords

• Machine learning
• Clustering
• Latent class analysis
• Bayesian analysis
• Decision tree
• Service line