Title :
Temporal Evaluation of Risk Factors for Acute Myocardial Infarction Readmissions
Author :
Stiglic, Gregor ; Davey, Adam ; Obradovic, Z.
Author_Institution :
Fac. of Health Sci., Univ. of Maribor, Maribor, Slovenia
Abstract :
Risk-adjusted 30-day readmission rates for specific diagnoses including myocardial infarction are now used to index hospital reimbursement rates, making prediction of this outcome particularly salient. In order to consider how the importance of various predictors may be changing over time, we apply a modified prequential evaluation technique with an extended training set to this classification problem. Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression using cyclical coordinate descent was used for classification. This paper proposes a bootstrapping based approach to evaluation of sparse coefficients in large sparse datasets with binary and numerical features. It was evaluated on an eight-year dataset of hospital discharge records of myocardial infarction patients consisting of 312,309 discharge records. Results indicate diagnoses (clustered around related disease systems) and length of stay are the most important positive predictors, whereas procedures and diagnoses correcting for small groups of patients and total charges are more important among negative predictors. Temporal comparisons tend to suggest that the importance of features themselves is changing, rather than their prevalence.
Keywords :
cardiology; hospitals; medical administrative data processing; pattern classification; regression analysis; risk analysis; LASSO; acute myocardial infarction readmissions; bootstrapping based approach; classification problem; cyclical coordinate descent; hospital discharge records; hospital reimbursement rate index; least absolute shrinkage and selection operator logistic regression; patient diagnosis; prequential evaluation technique; risk factors; sparse coefficients; temporal evaluation; training set; Discharges (electric); Diseases; Hospitals; Logistics; Medical diagnostic imaging; Predictive models; Training; acute myocardial infarction; hospital readmission classification; sparse logistic regression;
Conference_Titel :
Healthcare Informatics (ICHI), 2013 IEEE International Conference on
Conference_Location :
Philadelphia, PA
DOI :
10.1109/ICHI.2013.87