UCLA

Background

Amidst high and rising health care spending in the US, there is a growing focus on patients with complex needs and high costs. Some argue high cost patients are emblematic of failures in the health care system, and efforts to intervene are proliferating. However, information about the characteristics and needs of patients with high costs is superficial. This dissertation aims to establish a highly granular and data-driven characterization of high cost patients.

Setting and Population

We used data from Kaiser Permanente Southern California to study the top one percent of adult members with respect to total health care spending in 2010. The high-cost cohort included 21,183 individuals. We had comprehensive data for all utilization across the continuum of care, from 2010 and four follow-up years through 2014.

Methods

We use applications of finite mixture models - latent class analysis and latent growth curve analysis - to identify unobservable segments or clusters of high cost patients with regard to their clinical conditions (Chapter Two) and their longitudinal trajectories of spending (Chapter Three). Finally, we used multilevel logistic regression to assess the association of neighborhood deprivation with sustained high spending over an extended duration (Chapter Four).

Results

We identified seven distinct clinical sub-types of high cost adults, with each class having distinctive patterns of utilization, survival, and individual characteristics. Following their individual trajectories of spending over four years from 2011 to 2014, we found five distinct patterns of spending after the base year of top one percent spending. Nearly half of the original cohort maintained consistent and relatively high spending for all four years of follow-up data available. Focusing on one sub-type of high cost adults, we found no significant association between neighborhood deprivation and sustained high spending.

Conclusions

We generated detailed and novel information about the segmentation of patients with high costs without making a priori assumptions about the population. Our studies were not designed to address the appropriateness of utilization, and there are persistent questions around the potential to impact spending for improved value and satisfaction. Future research should use the relatively homogenous subgroups we developed to investigate opportunities for improvement.