Year
2020
Location
Scripps Research Translational Institute
Category
Research
Duration
3 Months
This study investigates how changes in body temperature align with subsequent shifts in heart rate (HR), respiration rate (RR), and aortic blood pressure (AOPA) during the onset of viral and bacterial diseases.
Rather than focusing on specific pathogens, the goal is to identify lead-lag relationships between these vital signs using signal dynamics. The research is inspired by advances in machine learning for early disease detection, aiming to develop non-pathogen-specific, signal-based prediction methods suitable for wearables and early warning systems.
Subjects: Non-human primates (NHPs) exposed to viral (e.g., Ebola, Marburg, Lassa, Nipah) or bacterial (Y. pestis) pathogens.
Signals: Time-series physiological data (HR, RR interval, AOPA, Temp), sampled every 30 minutes.
Preprocessing:
Mean imputation for missing values.
Standardization and smoothing with 1-day/48-sample rolling averages.
Hierarchical data transformation using pandas MultiIndex to handle multiple time series per subject.
Analysis:
Time-lagged cross-correlation (TLCC) to quantify temporal offsets between temperature and each of the other signals.
Offsets calculated to determine how many frames (0.5h intervals) one signal lags or leads another.
General Pattern: Most signals lag behind temperature, with the typical sequence being: Temperature → HR/RR → Respiration → Aortic Blood Pressure.
Quantitative Results (mean offsets in frames, i.e., 0.5h units):
AOPA: −56.67 (most delayed, also most variable)
Respiration: −27.28
Heart Rate: −32.28
RR Interval: −38.07
Exceptions: In some Nipah virus cases, AOPA preceded temperature, highlighting variability based on disease type.
RR-HR Relationship: As expected, RR intervals are inversely correlated with HR and show similar lag distributions.
This study shows that vital sign synchrony analysis via cross-correlation can detect early physiological changes post-pathogen exposure.
High body temperature is typically the first sign, followed by heart and respiratory changes, then blood pressure.
These lag patterns suggest potential for predictive algorithms in wearables without relying on biomarker-specific models.
This signal-based method provides a foundation for future machine learning models for early disease detection with low etiological specificity.
