Phase 1
Data Collection and Baseline
1Aggregate data and confirm integrity
Data pulled from the source system, whether LIMS, EM software, or paper records transcribed accurately, with completeness confirmed before any statistical work begins.
ALCOA+ principles
↓
2Establish the baseline dataset
A defined historical period, typically at least one year or a statistically meaningful number of data points, used to characterize normal process variation.
USP <1116> · PDA TR 13
↓
3Exclude assignable-cause events from the baseline
Data points tied to a documented, investigated, and resolved special cause excluded from baseline calculations so the baseline reflects genuine routine process performance.
PDA TR 13
↓
4Select the appropriate statistical method
Method selected based on data type. Continuous data such as conductivity may suit control charts; count data such as microbial recoveries typically require a non-normal, percentile-based approach.
USP <1116>
Phase 2
Statistical Analysis and Limit Setting
5Calculate baseline statistics
Mean and standard deviation calculated for normally distributed data; percentile-based statistics calculated for count or non-normal data using the established baseline dataset.
USP <1116>
↓
6Derive alert and action limits
Facility-specific alert and action limits derived from the baseline statistics, distinct from any compendial or regulatory ceiling limit that may also apply.
PDA TR 13
↓
7Document rationale and obtain QA approval
Limit-setting methodology and resulting values documented and formally approved before implementation, providing a defensible basis if the limits are later questioned.
21 CFR 211.22
↓
Do the newly calculated limits differ meaningfully from current limits?
YES → Change control to update
NO → Maintain, document review
Phase 3
Ongoing Trend Review
8Plot new results routinely
Each new result plotted against the established chart or limits at a defined review frequency, such as monthly, keeping the trend picture current.
ICH Q10
↓
9Apply run rules to detect signals
Predefined pattern rules applied to the plotted data, flagging a trend signal even when no individual result exceeds its alert or action limit.
PDA TR 13
↓
10Compare trends across related data streams
Trends reviewed alongside related programmes, such as EM trends alongside water system trends for the same area, to identify correlated signals a single dataset would miss alone.
EU GMP Annex 1 2022
Phase 4
Signal Response and Adjustment
11Investigate trend signals proactively
A detected trend signal triggers investigation even without any individual limit exceedance, since the value of trending lies in acting before a limit is actually breached.
ICH Q10
↓
12Determine root cause and corrective action
Root cause investigation and corrective action follow the same rigor as an individual exceedance investigation, addressing the underlying driver of the trend.
21 CFR 211.192
↓
13Periodically reassess the baseline and limits
Baseline and derived limits reassessed on a defined schedule, or following a significant process change, to confirm they still reflect current, genuinely in-control performance.
USP <1116>
Statistical Method Selection
Normally Distributed Continuous Data
Mean plus standard deviation approach, typically with alert at 2 standard deviations and action at 3 standard deviations, suits data such as conductivity or temperature readings.
Classical SPC methodology
Non-Normal Microbial Count Data
Percentile-based limits, often set at the 95th or 99th percentile of the baseline dataset, better reflect the naturally skewed distribution of microbial recovery data.
USP <1116> · PDA TR 13
Low-Frequency Event Data
For data streams with infrequent events, such as rare organism recoveries, rate-based trending over a longer rolling window may be more meaningful than point-by-point limits.
PDA TR 13
Trend Signal Response Steps
Step 1 - Confirm the signal is genuine
Verify the underlying data is accurate and complete, ruling out a data entry or transcription error before treating the pattern as a real trend.
Step 2 - Characterize the pattern
Describe the specific nature of the signal, such as a gradual drift, a step change, or a cluster of near-limit results, to focus the investigation appropriately.
Step 3 - Correlate with related data and events
Cross-reference maintenance records, process changes, seasonal factors, and related monitoring programmes for a plausible contributing cause.
Step 4 - Determine whether action is warranted now
Assess whether the trend justifies immediate corrective action or continued closer observation, documenting the rationale either way.
Step 5 - Implement and verify effectiveness
Implement any corrective action and continue trending afterward to confirm the signal resolves and does not recur.
Data Requirements for Meaningful Trending
Sufficient baseline data points
Baseline dataset large enough to characterize genuine variation, generally spanning at least a full year to capture any seasonal effects.
Consistent data collection method
Sampling method, location, and technique held consistent across the trended dataset, since a method change can introduce an artificial trend.
Documented exclusion criteria
Clear, pre-defined criteria for excluding assignable-cause data points from baseline calculations, applied consistently rather than case by case.
Defined review frequency
A set cadence for trend review, such as monthly, ensures signals are caught in a timely manner rather than only at an annual review.