PAT Model Concordance Rate (PMCR)

GMI-PH-001Pharmaceutical OperationsPilot-Ready

Definition

Percentage of batches where the PAT model's predicted critical quality attribute value fell within tolerance of the traditional lab reference result, tracked on a rolling 30-batch window. Detects PAT model degradation that batch outcome metrics cannot see.

The Measurement Gap

Batch Success Rate measures final outcomes (pass or fail) but cannot detect whether the PAT model producing those outcomes is degrading. Right First Time does not distinguish between a batch that passed because the PAT model was accurate and one that passed because the model's error happened to fall within specification. OEE improves when the lab testing bottleneck is removed, regardless of whether the model replacing that testing is still calibrated. None of these metrics can see a model that is confidently wrong but not yet wrong enough to cause a failure. By the time a failure occurs, the model may have been degrading for months, and every batch released during that period is at retrospective risk.

Decision Supported:

Is our PAT model still accurate enough to support real-time release decisions, or has it drifted to the point where traditional testing must be re-established as the primary release gate?

Formula

Plain language: Count the number of batches where the PAT model's predicted CQA value fell within ±X% of the traditional lab result, divide by the total number of batches where both measurements exist, and express as a percentage. Track on a rolling 30-batch window.

Notation:

PMCR = (Σ Concordant_Batches / Σ Dual-Tested_Batches) × 100

Where a batch is concordant when: |PAT_Predicted_CQA − Lab_Measured_CQA| ≤ Tolerance

Variables:

Variable Definition Unit
Concordant_Batches Batches where the absolute difference between PAT prediction and lab result is within the tolerance band Count
Dual-Tested_Batches Batches where both the PAT model prediction and a traditional lab result exist for the same CQA Count
Tolerance Organization-defined acceptable discordance threshold, typically ±2–5% of the specification range % of spec range

Proposed formula, pending validation pilots.

At a Glance

Unit
Percentage
Trend Direction
Higher
Reporting Frequency
Per batch; monthly model-concordance summary
Category
AI Model Fidelity & Drift
← Inspection Sensitivity Distribution Index (INSDI)Regulatory AI Currency Index (RACI) →

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