What is the right amount of utilization in Home Health Care?
*Iordan Slavov, Visiting Nurse Service of New York 

Keywords: Service Utilization, Predictive Modeling, Dose-response, Generalized Propensity Score

Chronic budget issues in federally funded health care programs such as Medicare make the question of what is the optimal amount of home care someone needs especially important.Using 2010 data from a large Home Care agency in the New York Metro area, we are trying different approaches to answer this question. We focus on rehabilitation/physical therapy (RT) services and:

(a) Build a dose-response relation between the number of visits (per week) and an outcome characterizing the patients’ improvement. It is important to pick the outcome variable so that it represents objective change in clinical conditions and largely excludes financial considerations. Here we model the relationship between frequency of visits (i.e. number of visits per week) and the change in the “timed get-up-and-go” variable (a measure of patient mobility,) adjusting for clinical and other differences using a generalized propensity score. The resulting curve clearly identifies optimal limits (upper as well as lower) for the amount of rehabilitation therapy.

(b) Model separately length of stay (LOS) and intensity of visits. LOS can be seen as somewhat separated from financial restrictions and is modeled here by a quasi-likelihood generalized linear model (with a log link and constant variance.) It is more difficult to predict the intensity of care on a case-by-case basis. We select a group of patients who seemed to have received care with the proper frequency. Categorical frequency groupings are then described with a proportional odds model. The predicted number of visits for a new patient is the (rounded) product of the estimated LOS and visit frequency.

Both approaches illuminate important sides of the problem and suggest possible solutions.