Definition
PASDR measures the proportion of classified physiologic alerts where the system or clinician correctly identifies the true source: genuine patient deterioration, sensor artifact, equipment malfunction, or environmental interference. Higher PASDR indicates more reliable alert triage, reducing unnecessary interventions triggered by false alarms while preserving sensitivity to real clinical events.
The Measurement Gap
Measurement Gap and Decision
Decision Question: When a physiologic monitor fires an alert, can the system reliably determine whether the signal represents genuine patient deterioration or a non-clinical source, and does the classification accuracy justify clinical trust in the alert stream?
What Goes Unmeasured Today:
Physiologic monitoring systems generate enormous alert volumes. Industry estimates suggest that 80–95% of clinical alarms are non-actionable: triggered by patient movement, sensor displacement, lead disconnection, or equipment interference rather than genuine deterioration. Clinicians learn to ignore most alerts, creating alarm fatigue that endangers the patients whose deterioration signals are buried in noise. Current systems report alert counts and response times but do not measure whether anyone (human or algorithm) correctly identified the source of each alert. An AI-enhanced monitoring system may claim to reduce alert volume by filtering non-clinical signals, but without measuring discrimination accuracy, organizations cannot know whether the filter is correctly identifying artifact or silently suppressing real events. The gap is not in alert generation but in alert classification: knowing why the alarm fired and whether the classification was right.
Invisible Failure Mode:
Alert classification errors compound in both directions. False positives (artifact classified as deterioration) drive alarm fatigue and unnecessary interventions. False negatives (deterioration classified as artifact) create missed events with potential patient harm. Without measuring discrimination accuracy, organizations cannot quantify either risk or calibrate the appropriate trust level in alert classification systems.
Formula
Primary Formula:
PASDR = Correctly_Classified_Alerts / Total_Classified_Alerts
Variable Definitions:
| Variable | Definition | Source |
|---|---|---|
| Correctly_Classified_Alerts | Alerts where the assigned source category (patient deterioration, sensor artifact, equipment malfunction, environmental) matches expert adjudication | Alert management system + clinical review |
| Total_Classified_Alerts | All alerts that received a source classification during the measurement period | Alert management system |
| Missed_Deterioration_Rate | Subset metric: alerts classified as non-clinical that expert review determined were genuine deterioration events | Clinical review (secondary) |
Data Requirements:
Alert logs with timestamps, alert type, and source classification from monitoring or alert management system. Expert adjudication dataset from clinical review (sample-based is acceptable for ongoing measurement). Classification taxonomy agreed upon by clinical engineering and nursing leadership.
Edge Cases and Limitations:
Expert adjudication is itself imperfect. Establish inter-rater reliability before treating adjudication as ground truth. Sample-based measurement introduces sampling error; size samples to achieve statistical significance for the Missed_Deterioration_Rate specifically. Alerts that are silenced before classification cannot be measured: track unclassified alert rates separately.
Proposed formula, pending validation pilots.