3 Presentations

• 
  03:30 PM - 04:00 PM

  A heuristic algorithm for the Capacitated Vehicle Routing Problem with Synchronized Pick-ups and Drop-offs: a case study for medications delivery and supervision in DR Congo

  • Viviane Gascon, presenter, UQTR
  • John Clarke, Université de Montréal
  • Jacques Ferland, Université de Montreal

  In post-emergency contexts such as Western DR Congo, one of the crucial challenges that rural hospitals face is maintaining a pharmacy with essential medications and supplies. The cost of transporting medications and providing on-site supervision to remote hospitals is an extremely expensive endeavour and can cost as much as the medication itself. Using the province of Bandundu, DR Congo as a case study, our research attempts to determine the feasibility of a synchronized routing problem for medication delivery and on-site supervision visits. We propose a Capacitated Vehicle Routing Problem formulation that handles several requirements including activity-wise synchronization, precedence, and two activity frequencies. We implement a cluster-first, route-second heuristic with a geospatially-enabled database to solve the problem. We also present a web-based tool to visualize the solutions in a map. The preliminary results of our study suggest that a synchronized solution could offer significant savings to rural hospitals and increase the accessibility of medical services to rural populations.

• 
  04:00 PM - 04:30 PM

  The family of Discrete Conditional phase-type distributions for modelling patient outcome and survival

  • Adele Marshall, presenter, Queen's University Belfast

  The discrete conditional phase-type dustrbution is a hybrid approach which brings together data mining approaches with a special type of Markov model that represents patient survival. The first representation of the dc-ph utilised a Bayesian network to represent patient characteristics that influenced patient length of stay in hospital. Since then, naive Bayes, decision trees,random forests and most recently support vectors hsve been incorporated in the model alongside the survival data. This paper presents the dc-ph distribution and it's successful applications. In particular the dc-ph model utilising the support vector machines for predicting Retinopathy of Prematurity in neonatal babies will be described.

• 
  04:30 PM - 05:00 PM

  Modelling the Patient Recovery Process in an Intensive Care unit

  • Fermín Mallor, presenter, Public University of Navarre
  • Cristina Azcárate, Public University of Navarre
  • Julio Barado, Hospital of Navarre
  • Laida Esparza, García Orcoyen Hospital
  • Alba Agustín, Public University of Navarre

  Several methods have been used to analyze LOS data in intensive care unit (UCI). Previous studies have proved that LoS data have outliers and are usually heavily skewed to the right. Probability distributions with heavy tails and different regression models have been used for LoS modeling purposes. These models present a good performance when the primary goal is to estimate some characteristic of the LOS as its expected value or some percentile. Nevertheless, these models fall short of modeling the recovery process of the patient and can not be used as input of a dynamic simulation model of the ICU showing the health status of the patient.

  In this work we model the health status evolution of an ICU patient using historical data concerning the infections gotten in the UCI, the type of illness, age, APACHE II index, etc. The purpose is to describe and understand how these covariates influence LoS and dictate the patient´s recovery process. In particular, Coxian phase-type and general phase type distributions as well as semi-Markov process are considered. These distributions have been used to model LoS in hospitals, but no instances of their use in ICUs could be found in the existing literature.