Today, Sebastiaan van Zelst defended successfully his M.Sc. thesis titled “Defining a Financial Forecasting Model for Healthcare Insurance Companies: a collaborative Markov chain approach incorporating institutional care pathway traversal”. Sebastiaan studied Business Information Systems and his project was a collaboration with KPMG.
This document concerns an exploratory research towards the application of collaborative Markov chains as a forecasting model within the field of healthcare insurance. The model proposed is based on both predicting care demand and associated institutional pathway traversal. A system of collaborative Markov chains allows the user to jointly model several probabilistic elements that share dependencies. It describes a collection of Markov chains in which the state of a certain chain within the collection influences transition probabilities in other chains within the collection. It allows the use of different types of techniques to estimate forecasting parameters as an input within one model. It entails a modular structure which allows the user to perform case-based analyses. Simulation of simplified proof-of-concept cases has shown accurate predictive behavior. Due to the fact that Markov chains and consequently systems of collaborative Markov chains are probabilistic in nature, simulation of a sufficient number of sampling replication yields results that tend to follow a Normal distribution. The statistical nature of the simulation results lends itself perfectly for consecutive statistical post-processing. Systems of collaborative Markov chains provide in modeling complex probabilistic systems in which several dependencies might exist. Current challenges within the application of systems of collaborative Markov chains involve the complexity in terms of the number of parameters to estimate and associated running times. Additionally the existence of potential “inactive elements” with respect to the field of healthcare insurance introduces additional challenges in parameter estimation of the model.
Today, Parvathy Meenakshy defended successfully her M.Sc. thesis titled “A Performance Measurement Framework for Clinical Pathways Monitoring”. Parvathy studied Business Information Systems and her project was a collaboration with Catharina Hospital Eindhoven.
Healthcare domain around the world is facing challenges due to the skyrocketing cost, chronic illness, ever increasing aging population, sedentary life style and changing demographics. There is an intense pressure to balance cost and quality of care. Several researches are conducted to investigate the benefits of adopting process orientation and process oriented information systems in healthcare, which ascertain the fact that process orientation can improve the quality of care, efficiency and cost effectiveness. Clinical pathway is an example of process management tool in healthcare to streamline process and standardize care. The actual execution and continuous monitoring of the pathway is of utmost importance for improvement. The monitoring of the pathway is essentially the measurement of indicators and its reporting to analyze the performance of the pathway. In this thesis, we develop a framework for performance measurement system for clinical pathway monitoring. The system uses an ontology of pathway indicators with a formal method to define them. A proof of concpet system has also been implemented.
Luis Jorge De Luna Orozco has defended successfully his M.Sc. thesis titled “Performance of Machine Learning Algorithms to Predict Anastomotic Failure in Bariatric Surgery”. Luis studied Business Information Systems and his project was a collaboration with Catharina Hospital Eindhoven.
The main purpose of this report is to perform a machine learning study to find out to which degree these algorithms can predict occurrence of anastomotic failure in bariatric surgery. The information used in this study was obtained from two databases provided by the Catharina Hospital in Eindhoven which contain preoperative information about patients (Sleeves) and perioperative blood pressure (4KP) recorded for 1116 surgeries since 2006. First part of this study covers an analysis to find out which features or variables can be obtained from this set of data and to which degree they have predictive power to identify complications after bariatric surgery, more specifically intestinal leakage when using them in classification predictive models.
Second, investigation and performance analysis of three different machine learning classification algorithms is carried using techniques to compensate imbalanced sets of data due to the fact that only 3% of the patients present anastomotic failure after surgery. Finally relation between hypotension episodes an anastomotic failure is studied and explained.
Main results of this study show that Random Forest classifying technique can offer prediction accuracy up to 90% when taking into consideration perioperative and preoperative bariatric surgery information like smoking assessment, surgery duration, medication assessment, surgery technique and occurrence of hypotension episodes under sedation. At the end all these results are discussed and a Random Forest model in WEKA is provided, moreover possible future work is suggested to continue with the creation of a full clinical decision support system.