State transition modeling of complex monitored health data

Area of Science:

• Biostatistics
• Health Data Science
• Medical Informatics

Background:

• Complex health data often involves multiple signals indicating distinct health states.
• Covariates can influence transitions between these health states over time.
• Analyzing time-dependent covariate effects requires flexible statistical methods.

Purpose of the Study:

• To introduce a non-parametric state intensity regression method for analyzing complex health data.
• To investigate the time-dependent effects of covariates on health state transition intensities.
• To apply the method to analyze newborn resuscitation data.

Main Methods:

• Application of a non-parametric state intensity regression.
• Use of weighted median and hysteresis filter for continuous health signals.
• Aggregation of covariates over time history windows.
• Analysis of cumulative regression parameters for short- and long-term effects.

Main Results:

• The non-parametric method effectively models time-dependent covariate effects on state transitions.
• Data pre-processing steps enhance robustness for continuous health signals.
• The framework allows for investigation of both immediate and delayed covariate impacts.
• Application to newborn resuscitation data yielded insights into covariate influences.

Conclusions:

• The proposed non-parametric state intensity regression offers a flexible approach for complex health data analysis.
• The method is suitable for various data types and minimal assumptions.
• Understanding time-dependent covariate effects is crucial for clinical insights, as demonstrated in the Tanzania newborn resuscitation study.