Multilevel Latent Class Models for Evaluation of Long-term Care Facilities

Abstract

The Region Umbria has conducted a survey on elderly patients living in long-term care facilities since the year 2000. Repeated measurements of items on health conditions are taken for classifying care facilities to allocate resources (RUG system). We wish to evaluate the performance of the nursing homes in terms of some aspects of health condition and quality of life of patients. To this end, a Multilevel Latent Class Model with covariates is employed. It allows for modeling a latent trait hidden behind a set of items, also in the presence of a nested error data structure coming from repeated measurements. Eleven items, surveying cognitive status, activities of daily living, behavior and decubitus ulcers, are used to measure the latent variable related to health condition and quality of life. The probability of belonging to ordinal latent classes is modeled in terms of available covariates.