Full text loading...
薬理と治療
Abstract
The present study analyzed the effect of increasing doses of DPP-4 inhibitor to discover the factors affecting the dosage increase of oral hypoglycemic agents to diabetic patients. The analysis objects were 737 cases stored in a DPC receipt database, who were separated into two groups that one group showed the effect of increasing doses while the other group did not. We analyzed the two groups separately and observed no significant differences due to the usage of concomitant medications. By the machine learning model constructed in this study, the training data showed a high precision with AUC higher than 0.8, whereas the test data showed a standard precision with AUC between 0.6 and 0.75. We calculated the top 10 feature values that affected this model greatly. The maximum value, the minimum value and the trend of HbAlc were included in the top 10 feature values collectively. Besides, the minimum value of glucose was also recognized as one of the top feature values tested by various cross validations. Some of the parameters of liver function, such as γ-GTP, AST, ALT and cholinesterase, were recognized as a feature value but they were inconsistent by different cross validations with the contribution left unclear. In the future, it is better to consider the necessities of securing a larger blood test result data number with fine granularity, and utilizing databases storing daily outpatient clinical data covering the states of chronic diseases. In addition, adoption of Dimensionality Reduction Techniques, such as input of compressed data to extract feature values more concisely, may also help to increase the property of feature value extraction. The new utilization of RWD is expected by solving the above issues.
Data & Media loading...