Quick Answer
KRI thresholds should be set using a combination of historical trial data, statistical methods (typically percentile-based or standard deviation-based), and clinical judgment. The most effective approach uses a two-tier system: an amber threshold (early warning, typically 1.5–2 standard deviations from the mean) and a red threshold (action required, typically 2–3 standard deviations). Thresholds must be calibrated during the first 3–6 months of enrollment and adjusted as trial data accumulates.
Why KRI Thresholds Are the Hardest Part of RBQM
Key Risk Indicators are only as useful as their thresholds. Set them too tight, and your team drowns in false positives — every site triggers an alert, nobody investigates, and the system loses credibility. Set them too loose, and genuine risk signals pass undetected until they become audit findings.
According to the ACRO 2025 RBQM Summary Report, 96% of clinical trials now include some form of RBQM, but organizations consistently report that KRI threshold calibration is one of the top three implementation challenges. The problem is not technical — most RBQM platforms support configurable thresholds. The problem is deciding what numbers to use and how to justify them.
Three Methods for Setting KRI Thresholds
Statistical Method: Standard Deviation from the Mean
Calculate the mean and standard deviation for each KRI across all sites. Set the amber threshold at 1.5–2 standard deviations and the red threshold at 2–3 standard deviations from the mean. This method works well for KRIs with normally distributed data, such as query response time or enrollment rate.
Best for: KRIs with sufficient data volume and approximately normal distributions.
Percentile-Based Method
Rank all sites by KRI value and set thresholds at specific percentiles. The amber threshold is typically set at the 75th–85th percentile, and the red threshold at the 90th–95th percentile. This method is more robust for skewed distributions and does not assume normality.
Best for: KRIs with skewed distributions (e.g., protocol deviations, adverse event rates).
Clinical Judgment Method
For KRIs where statistical methods are impractical (small sample sizes, binary outcomes, or regulatory requirements), use expert consensus. Assemble a cross-functional team (medical monitor, biostatistician, clinical operations lead) and define thresholds based on what constitutes a clinically meaningful deviation.
Best for: Safety-critical KRIs, compliance metrics, and early-phase trials with limited data.
Five Common Threshold-Setting Mistakes
Setting thresholds before you have data
Use provisional thresholds from historical trials or industry benchmarks for the first 3 months, then recalibrate with actual trial data.
Using the same thresholds for all therapeutic areas
Oncology trials have fundamentally different AE profiles than dermatology trials. Thresholds must be calibrated to the specific indication and patient population.
Never adjusting thresholds after initial setup
Schedule quarterly threshold reviews. As enrollment increases and data accumulates, your statistical estimates become more precise.
Too many KRIs with too many alerts
Limit your active KRI set to 8-12 indicators. If everything is flagged, nothing gets investigated. TransCelerate recommends focusing on KRIs that are actionable.
No documented rationale for threshold values
Regulators will ask why you chose specific numbers. Document the method (statistical, percentile, or clinical judgment) and the data source for every threshold.
Step-by-Step Framework
Define your KRI library
Select 8-12 KRIs aligned with your RACT-identified risks. Each KRI should map to a specific Critical to Quality Factor.
Choose the threshold method per KRI
Statistical for data-rich KRIs, percentile for skewed distributions, clinical judgment for safety-critical metrics.
Set provisional thresholds
Use historical data from similar trials, industry benchmarks (TransCelerate, ACRO), or expert consensus for initial values.
Define the two-tier alert system
Amber = investigate within 5 business days. Red = escalate to RBQM committee within 48 hours. Document response timelines.
Calibrate at 25% enrollment
Once 25% of target enrollment is reached, recalculate thresholds using actual trial data. Adjust and document changes.
Review quarterly
Schedule formal threshold reviews every 3 months. Track false positive rates and adjust to maintain signal-to-noise ratio.