Automatic Prediction of Health Status Using Smartphone-Derived Behavior Profiles

Area of Science:

• Biomedical Engineering
• Digital Health
• Wearable Technology

Background:

• Current methods for assessing patient health's impact on daily life are limited.
• Objective measurement of health status is crucial for effective treatment evaluation.

Purpose of the Study:

• To develop a sensor-based approach for objective health status measurement.
• To evaluate the use of human behavior profiles from smartphone sensors as health status predictors.

Main Methods:

• Generated human behavior profiles using smartphone accelerometer and gyroscope data.
• Employed Support Vector Machine (SVM) regression models for prediction.
• Trained and tested models on 171 participants' data, correlating with SF-36 self-ratings.

Main Results:

• Successfully predicted eight individual SF-36 scales with an average correlation of 0.683.
• Predicted two SF-36 component scores with an average correlation of 0.698.
• Achieved an average correlation of 0.752 for general health prediction.

Conclusions:

• The proposed sensor-based method offers a promising direction for objective health status prediction.
• This approach enables unobtrusive, inexpensive health monitoring using readily available hardware.
• Provides clinicians with objective data to assess treatment benefits on an individual patient basis.