Classification-based data mining for identification of risk patterns associated with hypertension in Middle Eastern...

Area of Science:

• Cardiovascular Health
• Public Health Research
• Data Mining in Medicine

Background:

• Hypertension is a major global health issue requiring effective risk identification.
• Traditional multivariable models have limitations in identifying complex risk patterns.
• Data mining offers novel approaches to analyze large datasets for health insights.

Purpose of the Study:

• To identify risk patterns associated with hypertension incidence in an Iranian adult population.
• To apply data mining techniques, specifically decision tree algorithms, for hypertension risk prediction.
• To compare the performance of different decision tree models in predicting hypertension.

Main Methods:

• Utilized data from 6205 adult participants (44% male) free of hypertension at baseline.
• Developed prediction models using three types of decision tree (DT) algorithms: Classification and Regression Tree (CART) and Quick Unbiased Efficient Statistical Tree (QUEST).
• Evaluated classifier performance using C-statistic and sensitivity on a testing dataset.

Main Results:

• QUEST algorithm showed best performance in men and women; CART performed best for the total population.
• Prediction models achieved C-statistics of 0.70 (men), 0.79 (women), and 0.78 (total population).
• Significant predictors included systolic blood pressure (SBP), age, diastolic blood pressure, and waist circumference. Specific predictors like wrist circumference and plasma glucose levels varied by gender.

Conclusions:

• Decision tree models are effective for predicting hypertension and uncovering interactions between risk factors.
• DT models identified distinct, high-risk subgroups based on easily accessible variables like SBP and age.
• The study highlights the utility of DTs in understanding hypertension risk patterns within specific populations.