Outcome Prediction in Clinical Treatment Processes

摘要: Clinical outcome prediction, as strong implications for health service delivery of clinical treatment processes (CTPs), is important both patients and healthcare providers. Prior studies typically use a priori knowledge, such demographics or patient physical factors, to estimate outcomes at early stages CTPs (e.g., admission). They lack the ability deal with temporal evolution CTPs. In addition, most existing employ data mining machine learning methods generate prediction model specific type outcome, however, mathematical that predicts multiple simultaneously, has not yet been established. this study, hybrid approach proposed provide continuous predictive monitoring on outcomes. More specifically, probabilistic topic applied discover underlying patterns from electronic medical records. Then, learned patterns, low-dimensional features CTPs, are exploited across various based multi-label classification. The proposal evaluated predict three typical classes outcomes, i.e., length stay, readmission time, discharge, using 3492 pieces patients' records unstable angina CTP, extracted Chinese hospital. stable was characterized by 84.9% accuracy 6.4% hamming-loss 3 latent discovered data, which outperforms benchmark classification algorithms prediction. Our study indicates can potentially improve quality assist physicians understand conditions, inventions, in an integrated view.