Identifying subgroups of adult high-cost health care users: a retrospective analysis

Few studies have categorized high-cost patients (defined by accumulated health care spending above a predetermined percentile) into distinctive groups for which potentially actionable interventions may improve outcomes and reduce costs. We sought to identify homogeneous groups within the persistently high-cost population to develop a taxonomy of subgroups that may be targetable with specific interventions.

We conducted a retrospective analysis in which we identified adults (≥ 18 yr) who lived in Alberta between April 2014 and March 2019. We defined "persistently high-cost users" as those in the top 1% of health care spending across 4 data sources (the Discharge Abstract Database for inpatient encounters; Practitioner Claims for outpatient primary care and specialist encounters; the Ambulatory Care Classification System for emergency department encounters; and the Pharmaceutical Information Network for medication use) in at least 2 consecutive fiscal years. We used latent class analysis and expert clinical opinion in tandem to separate the persistently high-cost population into subgroups that may be targeted by specific interventions based on their distinctive clinical profiles and the drivers of their health system use and costs.

Of the 3 919 388 adults who lived in Alberta for at least 2 consecutive fiscal years during the study period, 21 115 (0.5%) were persistently high-cost users. We identified 9 subgroups in this population: people with cardiovascular disease (n = 4537; 21.5%); people receiving rehabilitation after surgery or recovering from complications of surgery (n = 3380; 16.0%); people with severe mental health conditions (n = 3060; 14.5%); people with advanced chronic kidney disease (n = 2689; 12.7%); people receiving biologic therapies for autoimmune conditions (n = 2538; 12.0%); people with dementia and awaiting community placement (n = 2520; 11.9%); people with chronic obstructive pulmonary disease or other respiratory conditions (n = 984; 4.7%); people receiving treatment for cancer (n = 832; 3.9%); and people with unstable housing situations or substance use disorders (n = 575; 2.7%).

Using latent class analysis supplemented with expert clinical review, we identified 9 policy-relevant subgroups among persistently high-cost health care users. This taxonomy may be used to inform policy, including identifying interventions that are most likely to improve care and reduce cost for each subgroup.