A framework for hospital-specific readmission risk prediction and intervention recommendation with cost analysis approach | |
| Author | Landicho, Junar Arciete |
| Call Number | AIT Diss no.IM-22-01 |
| Subject(s) | Hospitals--Admission and discharge--Philippines--Mathematical models Hospitals--Admission and discharge--Philippines--Data processing Hospitals--Admission and discharge--Forecasting Patient Readmission--Philippines Risk assessment--Philippines Neural networks (Computer science) |
| Note | A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Information Management |
| Publisher | Asian Institute of Technology |
| Abstract | This study designed a hospital-specific framework that integrates readmission risk predictions, intervention recommendations, and cost analysis. The main purpose of the framework is to predict the readmission risk score of patients and provide the right intervention strategies that can lead to quality patient care, minimizing hospital care costs incurred by patients. A retrospective cohort was performed between June 2017 and December 2018 at the Northern Mindanao Medical Centre (NMMC), Cagayan de Oro City, Philippines, to investigate patient readmission within one year after discharge, including acute myocardial infarction (AMI), heart failure (HF), chronic obstructive pulmonary disease (COPD), and pneumonia (PN). Of the 2,234 patients during the study period, 663 patients were included in the analysis – 200, 127, 75, and 261 were eligible for the AMI, HF, COPD, and PN cohorts, respectively. In identifying the risk factors associated with readmission, the Regularized Logistic Regression (RegLR), Support Vector Machine (SVM), C5.0 Decision Tree, Artificial Neural Network (ANN), and Classification and Regression Tree (CART) algorithms were used and evaluated using performance metrics. The Bayesian Network model was then used to analyze intervention recommendations using three different types of learning algorithms: Grow-Shrink (GS), Hill Climbing (HC), and Max-Min Hill Climbing (Max-Min Hill Climbing) (MMHC). All learning algorithms were evaluated using standard metrics and a scoring function to assess the network's quality. Some of the risk factors from the predictive model and selected clinical attributes were incorporated into the system prototype, subjected to a preliminary evaluation by external experts comprised of domain experts and medical personnel. The study found that the Artificial Neural Network and Regularized Logistic Regression models outperformed other models to provide accurate prediction for 10-fold cross-validation and low misclassification cost. The experimental results showed that HC outperformed the other algorithms for the intervention recommendations model and was the best network in a small-scale dataset setting. The model generated intervention rules that could result in risk reduction and recommendations based on the similarities between the actual patient data and the existing patient profile. Furthermore, according to the user experience evaluation, the system prototype was a relevant and beneficial tool for physicians and health practitioners. The proposed framework established that predictive models provide insights into condition specific interventions and reduce readmissions of high-risk patients in developing countries. Additionally, based on the knowledge of the researcher, the interactive intervention recommender is the first of its kind in the Philippines. |
| Year | 2022 |
| Type | Dissertation |
| School | School of Engineering and Technology |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Information Management (IM) |
| Chairperson(s) | Vatcharaporn Esichaikul; |
| Examination Committee(s) | Chutiporn Anutariya;Huynh, Trung Luong; |
| Scholarship Donor(s) | University of Science and Technology of Southern Philippines; |
| Degree | Thesis (Ph.D.) - Asian Institute of Technology, 2022 |