Increasingly, hospitals are building large clinical databases for the purpose of facilitating the delivery of health care and conducting clinical research. Although clinical databases provide a rich and convenient source of clinical data, most of them are stored in unstructured free text, which is difficult to use for decision support or data analysis. Medical language making clinical data readily and conveniently available for use. At Columbia Presbyterian Medical Center (CPMC), an MLP system called MedLEE has been developed. However, to apply MedLEE to a new clinical domain is not easy because the structured data of its output is very complex and it is difficult and time consuming for authoring the inference rules. This thesis proposes and documents defining variables, expanding the lexicon, constructing the inference rules, using MedLEE to overcome these difficulties.; which consisted of MedLEE and a set of inference rules processing the structured algorithm for using administrative ICD-9 diagnosis codes to detect comorbidities were both applied to the pneumonia patients admitted to CPMC. The evaluation showed that the MLP system performed better and detected more comorbidities than ICD-9 coding did.; I applied the refined MLP system and the ICD-9 algorithm to acute myocardial infarction patients admitted to CPMC. Using record linkage to ascertain patient survival status after discharge, I implemented survival analysis to construct the MLP-based and ICD-9-based comorbidity indices. Risk-adjusted logistic regression models showed that the MLP-based comorbidity index seemed to be a better predictor of both in-hospital mortality and one-year mortality after controlling for age and laboratory data. Finally, because the chart review of 36 deaths could not detect any cases in which death was preventable, I failed to demonstrate that the developed risk-adjusted mortality model could be a tool for screening quality-of-care problems.
展开▼
