Predicting secondary complications for chronic disease patients using multi-dimensional wellness data. Powered by XGBoost + SHAP on GCP.
Dataset Size
500K
User assessments
Wellness Dimensions
8
Tracked per user
Risk Tiers
3
Low · Moderate · High
Model
XGBoost
5-fold stratified CV
Business Question
"How can machine learning models use Nytia Health's multi-dimensional wellness data to predict the risk of secondary complications in users with existing chronic conditions, so that interventions can happen before those complications develop?"
Risk Tier Distribution
500Kusers
Low — 21,083 (4.2%)
Moderate — 162,117 (32.4%)
High — 316,800 (63.4%)
Top Declining Dimensions
Nutrition
92.4%
Sleep
57.7%
Anti-Stress
50.4%
Anti-Stress
26.1%
Depression
23.9%
Obesity
17.9%
Wellbeing
0.0%
Sleep
0.0%
Risk Prediction
Enter wellness scores to get a real prediction from the live Cloud Run API.
Trajectory Scores (Direction of Change)
Current Health Levels
EDA Insights
Key findings from 500,000 user wellness assessments (real sponsor data).
Total Records
500,000
Zero missing values
Missing Rate
0.00%
Across all columns
Highest Risk Tier
63.4%
Users in High tier
Key Finding: Compound Decline Drives Risk
Individual dimension correlations with risk tier are weak. But the derived compound features show strong signal: total_declining_count correlates at 0.69 and critical_cval_count at 0.38 (Spearman). This confirms that complication risk is driven by multi-dimensional deterioration, not single-factor decline.
Nutrition Dominates Decline
In the 500K real data, Nutrition declines in 92.4% of users — far above every other dimension, which cluster near 50%. Nutrition is the single most widespread health challenge and the primary intervention target.
Nutrition
92.4%
Sleep
57.7%
Others
~50%
Note: Other dimensions sit near 50%, indicating a largely uniform distribution outside nutrition.
Feature Correlation with Risk Tier (Spearman)
declining_count
0.69
critical_cval
0.38
dif_move
0.07
dif_smoke
0.06
dif_nutri
0.05
System Architecture
Hybrid GCP: Vertex AI Workbench (EDA) + Cloud Run (production pipeline + API)
# Run full pipelinegcloud run jobs execute nytia-ingestiongcloud run jobs execute nytia-featuresgcloud run jobs execute nytia-training# Test APIcurl /healthcurl /predict -X POST -d '{...}'