User Guide

Comprehensive instructions for using the Motion Health Predictor platform effectively

This guide provides detailed instructions for navigating the Motion Health Predictor platform and understanding how to use all its features for comprehensive health risk assessment based on movement patterns.

Introduction

Welcome to the Motion Health Predictor! This advanced web-based tool is designed for researchers, healthcare professionals, and individuals to assess health risks based on physical activity patterns. Our platform utilizes state-of-the-art LSTM and Self-Attention deep learning algorithms to provide accurate, interpretable health risk predictions across 327 ICD-10 coded diseases.

The system analyzes 192-dimensional movement data representing activity proportions across weekdays and weekends, including four activity categories: sedentary behavior, light activity, moderate-vigorous activity, and sleep patterns.

Frequently Asked Questions

1 How to use Motion Health Predictor?

A: The platform offers two main pathways for health assessment:

Personal Assessment Pathway:

  1. Navigate to the "Personal Assessment" page
  2. Input basic information (age, gender, BMI)
  3. Choose between PPDM (basic) or PPDM+ (enhanced with blood tests) model
  4. Upload movement data CSV file or use example data
  5. Click "Assess Risk" for comprehensive health evaluation

Professional Analysis Pathway:

  • Access model performance metrics and feature importance
  • View AUC, sensitivity, specificity across all 327 diseases
  • Analyze model interpretability with SHAP values
EXAMPLE Load Sample Movement Data
CLEAR Clear All Inputs
ASSESS RISK Run Health Risk Assessment
SELECT CSV FILE Upload Movement Data
SAMPLE DATA Download Example Format
Model Selection Tip: Choose PPDM+ for enhanced accuracy with blood test parameters, or PPDM for basic movement pattern analysis only.
2 What data format is required?

A: The system requires specific CSV formats depending on the chosen model:

PPDM Framework (Basic Model):
• 196 columns total
• Columns 1-4: Basic Information (Participant ID, Age, Gender, BMI)
• Columns 5-100: Weekday activity data (96 values)
• Columns 101-196: Weekend activity data (96 values)

Activity Data Structure (per hour):
• Column 1: Sedentary proportion (0-1)
• Column 2: Light activity proportion (0-1)
• Column 3: Moderate-vigorous proportion (0-1)
• Column 4: Sleep proportion (0-1)
PPDM+ Framework (Enhanced Model):
• Additional blood test parameters:
  - WBC, RBC, Hemoglobin, Platelets
  - ALT, AST, Glucose, Cholesterol
• Can be uploaded separately or entered manually

All activity proportion values must be between 0 and 1, representing the fraction of each hour spent in that activity category.

3 How to interpret assessment results?

A: The health risk assessment provides comprehensive, multi-dimensional results:

Risk Classification System:

Low Risk : Healthy patterns
Medium Risk : Needs improvement
High Risk : Significant concern

Population Percentile Display: See how your risk compares to others in your demographic group with clear percentage rankings and risk interval displays.

Comprehensive Results Dashboard Includes:

Overall Risk Score & Level
Disease-Specific Risk Probabilities
Movement Pattern Analysis
Key Risk Factor Identification
Personalized Recommendations
Results Export: Download complete assessment reports in PDF, CSV, or text formats for sharing with healthcare providers or personal records.
4 What algorithms power the predictions?

A: Our platform utilizes cutting-edge machine learning architectures:

Core Technologies:

  • LSTM Networks: Capture temporal patterns in 24-hour activity data across multiple days
  • Self-Attention Mechanisms: Identify critical time periods and activity types
  • SHAP Analysis: Provide transparent, interpretable feature importance
  • LightGBM Ensemble: For disease-specific risk classification

Model Validation: All models undergo rigorous nested leave-one-site-out cross-validation with bootstrap resampling (1,000 iterations) to ensure reliability and generalizability.

Performance Metrics Reported: AUC, accuracy, sensitivity, specificity, precision, Youden index, and F1 score with 95% confidence intervals.

5 How accurate are the predictions?

A: Our models demonstrate exceptional performance across multiple validation studies:

Mean AUC: 0.892 (Metabolic Risks)
Mean AUC: 0.865 (Cardiovascular)
Mean AUC: 0.878 (Musculoskeletal)
High-Risk Sensitivity: 0.845
High-Risk Specificity: 0.872

Validation Framework: Nested cross-validation across 22 recruitment centers ensures robust performance estimates and prevents overfitting.

Medical Disclaimer: This tool provides health risk assessment and lifestyle guidance based on movement patterns. It is not a substitute for professional medical advice, diagnosis, or treatment. Always consult healthcare professionals for medical concerns.

Technical Support

For optimal experience and troubleshooting:

Need Help? Our support team is available to assist with technical issues, data format questions, and interpretation of results.