Correlation analysis of laboratory blood tests and complications in Diabetes Mellitus using data mining technique

Abstract

The laboratory blood tests in this study were collected from patients diagnosed with diabetes mellitus (DM) who were admitted to a public hospital in Chiang Mai Province. The aim of this study is to examine the correlation between laboratory blood tests and the incidence of complications associated with DM using a data mining technique. The J48 classifier was applied to construct a decision tree model and verify the precision of the simulation model. The laboratory blood tests were collected from 1,736 patients diagnosed with DM who were admitted to a public hospital in Chiang Mai Province, Thailand in 2020. The results showed that among the total DM cases, 54.55% were diagnosed with chronic kidney disease stage 2 (CKD 2) or higher, consistent with the microalbuminuria category where early and advanced kidney damage was observed in 66.82% of total DM cases. Interestingly, clinical factors such as BMI, eGFR and microalbuminuria are related to DM complications, particularly with respect to kidney dysfunctions. The decision tree model, simulated with the J48 classifier, achieves a high predictive accuracy with a correct classified instance rate of 87.44%. Verification parameters were used to validate the quality of the model in each class of chronic kidney dysfunction which classifier exhibited the high %True positive rates of more than 70% in all targeted CKD classes (CKD 1-5) and with Precision of more than 80% indicating low %False positive rate. These findings highlight the advantages of clinical data analysis using data mining techniques.