Sub-Threshold Event Response Consistency (STERC)

GMI-PH-023Pharmaceutical OperationsIntegration-Required

Definition

Consistency of disposition across clustered sub-threshold events, computed as the mean share of each similarity cluster receiving the most common disposition. Similar anomalies below formal thresholds should be handled the same way.

The Measurement Gap

AI environmental monitoring detects events that fall below traditional thresholds, a temperature variation of 0.3°C in a controlled area with ±1.0°C limits, or a particulate count spike well below the classification limit. These events don't trigger formal excursions, but the AI flags them as anomalous relative to the zone's baseline. The organization has no decision framework for these sub-threshold events. In the aseptic suite, a QA reviewer quarantines the batch manufactured during the event. In the oral solid dosage area, a different QA reviewer looks at the same magnitude of sub-threshold event and dismisses it. Both are defensible given the absence of a decision framework: there is no SOP for sub-threshold environmental events because these events were invisible before the AI. The inconsistency means identical environmental conditions produce different product dispositions depending on who reviews them, which shift is working, and which area is involved.

Decision Supported:

When the AI detects a sub-threshold environmental event, does the outcome depend on the science or on who happens to be on shift, and how inconsistent has the organization's response pattern become?

Formula

STERC = (1/M) × Σ_m [N_majority_m / N_events_m]

Where M is the number of similarity clusters, N_majority_m is the count of events in cluster m that received the most common disposition, and N_events_m is the total events in that cluster.

Worked Example:

Over 6 months, 87 sub-threshold environmental events are clustered into 12 similarity groups based on zone classification, parameter type, magnitude relative to limit, and duration.

Cluster 3 (particulate spikes 2,000–3,500 in Class 7 areas, duration under 30 minutes): 8 events total. Five received "no action," two received "investigate," one received "quarantine batch." N_majority = 5, consistency = 5/8 = 62.5%.

Cluster 7 (temperature deviations 0.2–0.5°C in controlled areas, duration under 15 minutes): 11 events. Nine received "no action," two received "investigate." Consistency = 9/11 = 81.8%.

Averaging across all 12 clusters: STERC = 67.3%.

Sign Convention:

Higher is better. 1.0 means perfect consistency within clusters. Below 0.80 indicates the organization needs a formal decision framework for sub-threshold events.

Proposed formula, pending validation pilots.

At a Glance

Unit
Percentage
Trend Direction
Higher
Reporting Frequency
Monthly; per sub-threshold event
Category
Disposition Decision Quality
← Excursion-to-Disposition Decision Accuracy (EDDA)Chemical Space Exploration Breadth (CSEB) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment