EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

3251. Cluster analysis of treatment plans of rehabilitation patients

Invited abstract in session WC-10: Capacity and treatment planning in healthcare, stream OR in Health Services (ORAHS).

Wednesday, 12:30-14:00
Room: 11 (building: 116)

Authors (first author is the speaker)

1. Maik Overmars
Mathematics of Operations Research, University of Twente

Abstract

We perform a cluster analysis on treatment plans of rehabilitation patients. Rehabilitation patients receive multidisciplinary care during treatment, which allows treatment plans to be customised to each patient. We cluster patients with similar treatment plans with the goal of standardising these plans and to make predictions for new patients. We aim to use the clustering model as an input for capacity and planning models.
We partition each treatment plan into patterns based on decision moments where the doctor decides how to continue the patient’s care. The patterns are multivariate count data, as we count the number of appointments per discipline over the duration of the pattern.
We use finite mixture methods to cluster the patterns. Each cluster is modelled as a multivariate Poisson distribution. Given a number of clusters, we use Expectation-Maximization to find the optimal parameters for the model. We use the Bayesian Information Criterion to optimise the number of clusters.
We analyse the treatment plans, now sequences of patterns, using a non-stationary Markov chain. The transition probabilities are estimated from the clustered data. Subsequently, we calculate the probability of a treatment plan to identify common treatment groups.
Finally, we show the validity of the clustering model by simulating new patients and comparing to the actual data. Our method is used by Reade rehabilitation centre in Amsterdam to support the development of standard treatment plans.

Keywords

Status: accepted

Back to the list of papers