Medical alarms usually prefer false positives over false negatives; consumer settings often reverse that

What it is

On a monitor, oxygen or heart-rate alarms are institutionally biased to sound extra rather than miss. On a watch, stress nudges or auto activity detection are biased to miss rather than sound extra. The same family of physiological signals reverses the loss matrix between medical alerting and consumer interruption. Hospital sensitivity on a wrist gets uninstalled; consumer quiet in a ward misses a crisis.

Why it happens

The institutional cost of a clinical alarm sits on the miss: regulation, litigation, and professional norms punish false negatives. The cost of false positives is alarm fatigue—nurses mute sound or raise thresholds—which then imports misses as a second-order effect, but the first principle remains “sound first.” Consumer exit is nearly frictionless: one annoying notification and the permission dies, which is equivalent to removing the sensor. False positives destroy the channel here; false negatives often go unnoticed. Base rates differ too: a life-threatening event in intensive care needs high recall more than an office “high stress” peak. The reversal is not taste. It is who holds exit rights and who holds harm.

Studying it

Compare disablement curves in the two places: how fast hospital alarms are muted or thresholds raised, how fast notification permission is revoked in a consumer app. Factors: setting, reversibility of the event, presence of a professional on watch. Outcomes: chosen sensitivity/specificity, fatigue scales, days the channel survives. Asking medical students to rate a consumer app, or consumers to rate a monitor, warps the orientation. Field base rates must be shown separately: ventricular tachycardia in a ward and an EDA peak at a desk are not the same rarity.

Where it stops holding

Fall detection or arrhythmia flags on a consumer device that claim a medical use should lean toward the alarm side and accept the matching validation burden; sold as “health interest,” they must not quietly borrow hospital-grade sensitivity. Home care sits between: no nurse on watch, and not as cheap as one-tap uninstall, so it needs its own loss table. When alarm fatigue is severe enough, hospitals also deliberately drop sensitivity—orientation is not dogma, it is who the channel is about to abandon.

Applying it

• Classify the feature first: a regulated alarm, or a dismissible consumer hint; do not share one default threshold.
• On the clinical side, protect recall and use tiers (advisory / warning / crisis) to ease fatigue, rather than leaving users to mute everything.
• On the consumer side, default quiet and make high sensitivity an explicit optional mode.
• Verify with 30-day channel survival in the target setting (permission still on, alarms still answered) and missed events, plotting both orientations.

Related

• Same group: C9.06.1 False-positive and false-negative costs are asymmetric · C9.06.2 Threshold choice is a product decision, not an algorithm decision · C9.06.3 High-consequence actions must not be decided by a single sensor · C9.06.5 Cost asymmetry should appear as a concrete classification threshold, not only as a stated principle · C9.06.6 Reusing one sensor across functions may require different false-positive tolerances · C9.06.7 Evaluating false-positive cost must include long-term trust loss, not only the harm of a single error
• Adjacent: C7.08 False Wakes · C9.11 Latency of Physiological Input
• Search: alarm fatigue · medical versus consumer threshold · recall-first alerting