The Primacy of Observation Duration Over Event Accuracy in Cough Monitoring

Cough frequency is a clinically relevant biomarker, but current “gold standard” human annotation is operationally impractical beyond 24 hours. Hyfe’s analysis shows that sampling error from short observation windows is a greater source of inaccuracy than device misclassification.

A simulation of 1,000 chronic cough patients demonstrated:

• 1-day perfect monitoring (100% sensitivity, 0% false positives) produced a 13% median error in estimating true 30-day cough rate.
• Continuous imperfect monitoring (85% sensitivity, ~2 false positives/hour) reduced error to ~6% over 20 days, halving inaccuracy versus short-term perfect measurement.
• Most of the accuracy gains were captured within the first week of continuous monitoring.

Key takeaway: In highly variable physiological signals like cough, duration of observation outweighs per-event precision. Automated, wearable systems—even if imperfect—deliver more accurate, scalable, and clinically relevant estimates than short-term perfect systems.

Implications:

• Clinical trials: Continuous monitoring improves endpoint reliability, reduces cost, and enhances feasibility.
• Disease management: Long-term monitoring captures true patient burden, enabling earlier interventions.
• Digital biomarkers: Robust validation requires prioritizing longitudinal data capture over momentary precision.

Bottom line: We do not need perfect accuracy to measure cough optimally—only sustained accuracy, long enough to reflect reality.