Why Timing and Status Matter
A population PK analysis plan records decisions made before analysis, including the intended analysis set, data cutoff, and model-development approach. In practice, operational constraints or new information can make a planned step unsuitable. The important regulatory question is not simply whether the work changed, but what changed, when the decision was made, what information was available, and how the final analysis was affected.
An appropriately documented prospective amendment can preserve a clear account of an evolving analysis. A deviation, by contrast, records that an action departed from the plan in force at the time. Neither label by itself establishes that a choice was scientifically sound or unsound. The rationale, timing, evidence considered, and consequences for interpretation must be assessed together.
A Timeline-Based Classification
Classify changes by comparing the plan version in force with the sequence of decision, approval, and implementation. “Prospective” is a statement about timing, not a synonym for “reasonable.” To support the classification, the record should show that the amendment was settled before the changed action began and, where relevant, before access to information that could influence that decision.
For example, a team may discover before model fitting that a planned covariate is not available for a substantial portion of the analysis set. If it documents and obtains the required approval for a revised handling rule before fitting models, that may be a prospective amendment. If the team first tries several model specifications, sees which results are favorable, and then writes a revised rule, the document date alone does not make the decision prospective with respect to those analyses.
A departure occurs when implementation does not follow the plan then in force. The departure may be accidental, such as an incorrect data filter, or deliberate, such as an analyst proceeding with a different rule before an amendment is authorized. In either case, record what happened rather than retrospectively describing the action as if it had been planned.
Quote or summarize the relevant plan provision and identify the version effective on the date of the action.
Record when the issue was noticed, when the decision was made, when approval occurred, and when the changed procedure was implemented.
If a change was documented and authorized before implementation, classify it as a prospective amendment under the applicable governance process. If an action departed from the plan in force, record a deviation, even if a later amendment describes the future approach.
State which data, analyses, or conclusions may be affected, and whether the final report needs to qualify what the analysis can support.
What a Useful Record Contains
A concise record should allow a reviewer to understand the issue and reconstruct the analysis without guessing. Record the date and, when material, the time of the observation, decision, approval, and implementation. Identify the person or role responsible, the plan section affected, the old and new rules, and the reason for the change. Note what information was available to the decision-makers at the time, including whether they had seen concentration data, model results, or relevant summaries.
Also state the scope: which subjects, records, analyses, or decisions were affected; whether any work had already been performed under the earlier rule; and what was done with that work. If a departure occurred, describe its discovery, any correction or containment, and whether affected results were retained, repeated, or excluded from the final analysis. Document approval status accurately; do not imply approval that was not obtained.
The explanation should distinguish a scientific rationale from a convenience. “The revised rule was selected because it better addressed the prespecified regulatory question” is more informative when accompanied by the actual reason and supporting facts. “The original approach did not work” is not enough: specify what failed, when it became apparent, and how the replacement changes the evidence being interpreted.
Worked Examples
Worked Example: Changing a Covariate Rule Before Model Fitting
Hypothetical scenario. The signed plan states that body weight will be evaluated as a continuous covariate when recorded at baseline. On 6 May, before any model fitting, the team confirms that 18 of 120 subjects have no usable baseline weight. On 7 May, the responsible reviewers approve a revised rule: use the closest eligible predose weight within a specified window, and otherwise treat weight as missing. The first model using that rule is fitted on 9 May.
Classification. This is a prospective amendment for the model fitting that began on 9 May, provided the approval process was appropriate and the stated rule was fixed before implementation. The event sequence is issue on 6 May, approval on 7 May, implementation on 9 May. The amendment does not change the fact that the original plan specified baseline weight.
Interpretive assessment. The revised rule changes which weight records can enter the covariate analysis. The report should say that the analysis followed the amended rule and explain the reason and scope. It should not imply that the amended rule was in the original plan. The team should also make clear whether any exploratory work occurred before 7 May; if it did, its timing and role in the decision should be recorded and considered when interpreting the analysis.
Worked Example: A Data Filter Applied Contrary to Plan
Hypothetical scenario. The plan says to exclude records with an invalid dose date, but to retain records with a valid dose date when the nominal sampling time is missing, using the recorded time relative to dose if available. On 12 June, a programming filter removes all records with missing nominal sampling time. This removes 14 of 500 otherwise eligible concentration records. The error is found on 14 June, before the final model is fitted.
Classification and arithmetic. The filter departed from the plan, so this is a deviation. The fraction of the 500 records removed by the incorrect filter is
Action and interpretation. The count and percentage describe the operational impact; they do not establish that the effect on parameter estimates is negligible. Record the filter, discovery date, affected records, and corrective action. If the data are restored and the planned rule is applied before final fitting, report the corrected analysis as following the plan while still documenting the deviation and correction. If results from the incorrectly filtered data were reviewed, state that fact and assess whether it could have influenced later decisions. Do not erase the event because the filter was corrected.
Worked Example: An Amendment Proposed After Results Are Seen
Hypothetical scenario. The plan calls for evaluating renal function using a prespecified continuous relationship. After seeing initial model results, the team notices that the relationship is poorly estimated and proposes a categorical version using a cutoff chosen after inspecting the data. The team documents the proposal on 21 July and begins the categorical analysis that afternoon; approval is obtained on 23 July.
Classification. The categorical analysis began before approval, so its implementation departed from the plan in force. It should be recorded as a deviation. The document dated 21 July does not make the analysis a prospective amendment because approval came later and, more importantly, the proposal followed review of initial results. An amendment may authorize the approach for subsequent work under the applicable process, but it cannot change the historical sequence.
Interpretive assessment. The cutoff was informed by observed data, so the resulting analysis is exploratory with respect to that choice and should not be presented as an untouched prespecified test. Report the original planned analysis and clearly identify the additional analysis and its data-informed rationale. The reader should be able to distinguish evidence generated under the plan from evidence arising through a post-result change.
Why the Distinction Affects Interpretation
Plan changes and deviations matter because analysis choices can affect which observations contribute, how covariates are represented, and which model results are emphasized. A documented rationale helps reviewers judge whether a choice addressed a real data or scientific issue. Timing helps them judge whether the decision could have been influenced by observed results. Neither documentation nor approval removes uncertainty; they make the basis for interpretation visible.
A useful comparison is between a prospective amendment and an undocumented post hoc change. The first can provide a transparent, authorized response to information encountered during conduct, while preserving the original plan and the reason for revision. The second leaves uncertainty about when the choice was made and what information shaped it. A deviation is a separate record of what did not follow the plan; it is not automatically misconduct, but unexplained or selective departures can undermine confidence in the analysis.
| Situation | How to record it | Interpretive implication |
|---|---|---|
| Approved rule changed before implementation | Amendment, with dates, rationale, scope, and approval | Identify the analysis as following the amended rule |
| Action differed from the plan then in force | Deviation, with event sequence and corrective action | Assess which results or conclusions may be affected |
| Change proposed after reviewing relevant results | Record the timing and information available; distinguish exploratory work | Do not present the data-informed choice as wholly prespecified |
A common mistake is to say, “We amended the plan after the analysis, so the analysis was compliant with the amended plan.” That reasoning confuses a later description with the plan that governed the action. The correction is to identify the plan in force at implementation, record the departure, and explain how the final analysis handled it.
Diagnostics, Alternatives, and Limits
Several warning signs call for closer review: an amendment date that follows implementation; a rationale with no supporting event or evidence; missing dates for key steps; a change that appears only after favorable results; or inconsistent accounts across the analysis plan and report. These signs do not prove that a decision was biased, but they make the decision history harder to evaluate. Reconstruct the chronology from contemporaneous records where available, and identify uncertainty rather than inventing precision.
The main alternative to a formal amendment is to retain the original plan and document a deviation. That is appropriate when the action has already departed from the plan or when the team cannot establish that the change was authorized before implementation. A later amendment can still clarify the approach for future work, but it should not replace the deviation record. Conversely, not every operational clarification changes the analysis plan: if the action is already consistent with the plan, explain the clarification without overstating it as a substantive amendment.
These records cannot establish that an analysis choice was unbiased, that a model is correct, or that a result is robust. They establish a transparent account of decisions and departures, which supports—but does not substitute for—scientific review. Where the timing or impact is uncertain, state what is known, what remains uncertain, and how that uncertainty limits the conclusion.
Check Your Understanding
For each situation, classify the event and state the key interpretive point. Suggested solutions follow the questions.
- A revised analysis rule is approved on 3 September and first used on 5 September. No relevant model results were reviewed before approval. What should be recorded?
- An analyst applies a different exclusion rule on 10 October, then receives approval for that rule on 12 October. How should the earlier action be described?
- A team changes a covariate cutoff after inspecting the concentration data. Why should the final report distinguish this analysis from the planned analysis?
- In a hypothetical analysis, 9 of 300 eligible records are affected by a filter error. Calculate the affected percentage, and explain what the percentage does and does not show.
Solutions. (1) Record a prospective amendment, including the old and new rules, rationale, approval, dates, scope, and the fact that no relevant model results had been reviewed before approval. (2) Record a deviation because implementation preceded approval; the later approval does not retroactively change the status of the 10 October action. (3) The cutoff was data-informed, so the result is not equivalent to an analysis using a fully prespecified cutoff; report the planned and additional analyses distinctly. (4) The calculation is \(9/300\times100\%=3.0\%\). This is the proportion of records affected, not evidence that parameter estimates or conclusions changed by only 3.0% or were unaffected.