| dc.description.abstract | Background: Hypertension is a common global health illness that often leads to severe and life time threatening diseases such as heart failure and other severe conditions if left untreated. Hypertension diagnosis and treatment is aimed at lowering high blood pressure (HBL) and protect important organs like the heart, kidneys from damage and brain thus stroke (reduced an average of 35% - 40%), heart attack (20% - 25%), and heart failure (more than 50%).Studies conducted in Uganda shows that prevalence of HTN is at 27.2% and awareness was 28.2%, higher among females 37.0% compared to males 12.4. The study aimed at developing a predictive model to predict HTN cases for improvement of early detection of hypertension (HTN) using risk factors.
Method: The use of a machine learning-based method for prediction of the hypertension was based on historical data in the database. A review of medical history and bio data of all patients screened at the Uganda Heart Institute in accordance with the results from UHI database of 2009 and 2010. Prediction model pipeline which included prediction target, cohort constructions, and feature construction, feature selection, predictions model and performance evaluation. As the input to the machine learning algorithms, the HTN parameters (features) obtained from database was used for the classification algorithm. As the classification algorithms, C4.5 decision tree classifier was used in hypertension prediction. The variables used included: demographic factors (age, sex), marital status, religion, tribe/race, family history of HTN, occupation, height, weight, body mass index BMI, smoking, physical inactivity (lack of exercise) and alcohol intake.
Results: A total of 1,371 participants of complete data were extracted from the dataset and 793 participants were included in the study after cleaning. Among them, there are 344 cases of hypertension 210 female and 134.The developed model obtained an accuracy at 72% at 95% CI (0.68-0.74), specificity 42% and specificity 96%.
Conclusion: In this study, decision tree algorithm was used for prediction and evaluated using confusion metric and ROC curve. The model performed well on small sample. HTN care should focus on reducing modifiable risk factors like life style changes in order to prevent, reduce and treat HTN. | en_US |
