Clustering Healthcare Expenditure Trajectories of Organ Failure Patients at End of life

Trajectories of functional decline in end-of-life (EoL) patients are split into four distinct groups, with organ failure patients typically exhibiting an erratic decline near death1. Given the tendency to incur higher healthcare expenditure as function declines towards death, there is potential to examine expenditure trajectories of these patients to look for similar trends. Data for the last year of life, consisting of patient demographics, medical history and bills, was extracted from an administrative database in a public hospital in Singapore. Patient bills before subsidy were used as a proxy for healthcare expenditure. Monthly expenditures per patient were modelled using a Latent Class Mixed Model (LCMM) with B-splines2, to cluster expenditure trajectories into distinct groups. Between 2005 to 2013, there were 1,281 inpatient decedents with a diagnosis of either heart, lung, kidney or liver failure and had at least 3 months' data in the system. A 3-cluster LCMM was identified as the final model. A cluster of 436 patients (34.0%) consistently incurred high expenditures (HE) from the start of the year. A cluster of 657 patients (51.3%) remained low in expenditure (LE) until a sharp increase in the last month of life. A cluster of 188 patients (14.7%) exhibited a sharp increase in expenditure (IE) from the 7th month onwards. The largest organ failure group in HE, LE and IE were liver (27.5%), heart (40.6%) and multiple organ failure (35.6%) respectively. Relative to LE, IE was more likely to have a history of fluid or electrolyte disorders prior to the last year of life. Healthcare expenditure trajectories of EoL patients can be clustered into different groups of low, increasing and high expenditures over the last year of life. Early segmentation of these patient groups could facilitate care planning to improve quality of care and reduce healthcare consumption.