Uso de Técnicas de Data Mining para Análise de Bases de Dados Hospitalares com Finalidades de Gestão

• Autor(es): Freitas, José Alberto da Silva
• Data: 2009
• Identificador Persistente: http://hdl.handle.net/10216/7549
• Origem: Repositório Aberto da Universidade do Porto
• Assunto(s): CIÊNCIAS EMPRESARIAIS; Porto
• Tipo de Documento: Tese de doutoramento
• Idioma: Português

Descrição

Today s medicine produces vast quantities of data that is not always properly analyzed and explored. The aim of this work is to contribute to the process of knowledge discovery in health, namely with the discovery of new, potentially useful, knowledge, and also with the definition and implementation of new methodologies relevant for hospital management. New knowledge was achieved using data mining methods applied to hospital databases, including techniques from classification, regression and clustering. For the definition of new methodologies for knowledge extraction, cost-sensitive classification strategies were analysed and implemented, considering various economical and non-economical costs, usually present in medical tests. These new strategies were implemented using decision trees modified to account for cost minimization. In our results, regards new knowledge, we note that exceptionally high lengths of stay (outliers) are decreasing over years, although readmission rates did not increase, low lengths of stay did not increase and the number of comorbidities (secondary diagnoses) increase. Large hospitals have significantly more outliers than other hospitals, even after adjustments for the patients characteristics. Concerning new methodologies, we defined a new strategy for learning and for utilization, sensible to various costs, including a risk factor for each test, group costs and other specific costs. We compared our new strategy with traditional, non cost-sensitive, approaches and obtained better results, with lower total costs. In this study, we applied data mining techniques to medical data and we discovered new knowledge. This new knowledge can represent important indicators for some analysis in hospital management and planning, as they can point to more hospital efficiency, specifically regarding length of stay. Our cost-sensitive learning strategy aim to be adjusted to the real world as it includes costs commonly considered by health professionals. This work combines different relevant costs in the area of health management and suggests how these should be processed.