Enterprise AI Analysis
Prediction of Personalised Hypertension Using Machine Learning in Indonesian Population
This study leverages machine learning (ML) to improve hypertension risk prediction in Indonesia using the SATUSEHAT IndonesiaKu (ASIK) system. It compares models with and without personal hypertension history, evaluating their predictive accuracy across 9.58 million adult health records. Model A (with history) achieved an AUC of 0.85, while Model B (without history) achieved 0.78, indicating that robust ML models can effectively predict risk even with limited historical data. Key predictors identified by SHAP analysis include age, family history, body weight, and waist circumference. The research highlights the potential of ML for broad-scale risk screening and personalized hypertension management in low-resource settings.
Executive Impact: Key Findings
Implementing AI-driven hypertension prediction models in Indonesia can significantly reduce healthcare costs, improve early detection rates, and lead to more personalized patient management strategies. The study demonstrates that even with readily available data (Model B), substantial predictive power is achievable, making it highly applicable for national health systems.
0.85
Predictive Accuracy (AUC) - Model A
0.78
Predictive Accuracy (AUC) - Model B
9,580,646
Records Analyzed
2
Top Algorithms
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
| Metric | Model A (with Patient History) | Model B (without Patient History) |
|---|---|---|
| ROC AUC | 0.85 (superior predictive accuracy) | 0.78 (robust without history) |
| Sensitivity (Recall) | 0.69 | 0.71 (higher recall for screening) |
| Specificity | 0.85 | 0.69 |
| Algorithm | ROC AUC | Key Advantage |
|---|---|---|
| XGBoost | 0.78 |
|
| LightGBM | 0.78 |
|
| CatBoost | 0.76 |
|
| Logistic Regression | 0.75 |
|
| Random Forest | 0.74 |
|
Top Predictor: Age
0.62 SHAP Value for AgeAge was identified as the most influential non-modifiable predictor of hypertension risk, especially when personal history data is unavailable. This highlights its critical role in risk assessment.
Family History's Impact
0.30 SHAP Value for Family History of HypertensionFamily history emerged as the second most influential factor, reinforcing its importance as a non-modifiable risk factor in hypertension prediction.
Modifiable Lifestyle Factor: Salt Consumption
0.15 SHAP Value for Salt ConsumptionSalt intake was a notable modifiable risk factor, emphasizing the importance of dietary control in managing blood pressure and offering clear intervention pathways.
Enterprise Process Flow
Initial Dataset (21.4M records)
→
Age Filter (<18 or >90 years)
→
Outlier Exclusion (anthropometric data)
→
Missing Data Removal (complete case analysis)
→
Final Cleaned Dataset (9.58M records)
Addressing Data Limitations in Low-Resource Settings
The study acknowledges the challenges of data quality and availability in low-resource settings. While personal history significantly enhances predictive accuracy, Model B demonstrates that robust ML models can still effectively predict hypertension risk using other accessible demographic, clinical, and lifestyle features. This finding is crucial for countries like Indonesia where comprehensive historical medical records may be sparse or unreliable.
Key Learning: Model B's performance (AUC 0.78) without relying on personal hypertension history shows that practical and generalizable risk screening is achievable even with limited data, making AI solutions viable for broader public health applications in low-resource contexts.
Future Directions: Enhancing Specificity and Generalizability
Future research aims to integrate richer datasets, including medication status and other physiological systems, and expand data collection to rural and under-represented populations. The ultimate goal is to practically implement these tools within the national SATUSEHAT platform to create a scalable public health solution.
Key Learning: Continuous model enhancement through diverse data streams (wearables, genomics) and broader population coverage is essential for achieving higher specificity and ensuring equitable application of AI in public health.
Advanced ROI Calculator
Estimate the potential return on investment for implementing AI-driven solutions in your enterprise.
Annualized AI Savings
$0
Reclaimed Productive Hours
0
Implementation Roadmap
Our phased approach ensures a smooth and effective integration of AI into your existing workflows.
Phase 1: Data Integration & Model Adaptation
Integrate SATUSEHAT data with additional health records and adapt the LightGBM model for optimal performance in diverse regional contexts.
Phase 2: Pilot Deployment & Validation
Deploy the AI model in selected primary care facilities for a pilot phase, gathering real-world feedback and validating predictive accuracy in clinical settings.
Phase 3: Scaled Rollout & Continuous Improvement
Expand the deployment across Indonesia, coupled with ongoing monitoring, model retraining, and integration of new data sources (e.g., wearables) to enhance long-term efficacy and generalizability.
Ready to Transform Your Enterprise with AI?
Unlock the full potential of AI and drive significant growth and efficiency.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat