PRO Recall Periods: When Daily and Weekly Answers Differ

FDA's 2009 PRO guidance defines the recall period as the period of time patients are asked to consider in responding to a PRO item or question, and notes that recall can be momentary (real time) or retrospective of varying lengths. That definition matters most at the moment a study team turns a licensed questionnaire into an electronic build. A recurring operational question is whether a validated 7-day recall item can be replaced by a 24-hour daily diary whose seven entries are averaged into a weekly score.

The short answer is no, not without evidence. The two designs ask for different judgments. A 7-day recall item asks the respondent to integrate a week of experience into one answer given on one day. A diary average is computed by the sponsor from seven separate judgments about the same day. FDA's 2009 guidance observes that a response is likely to be influenced by the patient's state at the time of recall, that items asking patients to average their response over a period of time are likely to undermine content validity, and that short-recall or current-state items are usually preferable. Where detailed recall over a period is necessary, the guidance points to techniques such as worst- or best-experience framing or a diary.

Neither design is automatically better. Criteria papers by Stull and colleagues (2009) and Norquist and colleagues (2012) conclude that no single recall period suits every concept, and Norquist warns that a short recall window can underestimate burden when symptoms fluctuate across the day or from day to day. What follows for operations is that the recall period has to be written into the endpoint itself. Protocol wording under ICH M11 protocol endpoint definitions and derivation rules in statistical analysis plans for endpoints should name the instrument version, including its recall period, the completion schedule, the aggregation rule, and the rule for missing days.

The most direct summary is an open-access systematic review by Peasgood, Caruana, and Mukuria (2023) in The Patient, which identified 24 quantitative studies and 158 unique results comparing one-day with seven-day recall. For symptoms compared against the mean of daily reports over the same week, weekly recall reported more severe symptoms in 76% (58 of 76) of results; the difference was not statistically significant in 24% (18 of 76). For pain specifically, 79% (19 of 24) of mean-daily comparisons found higher weekly scores.

On size, 53% (35 of 66) of results comparing the mean or same-day one-day score with weekly recall found the one-day score at least 10% lower for symptoms, or 10% higher for quality of life, and 89% (59 of 66) found one-day recall showing fewer problems; only 4 of 66 went the other way. The authors could not pool results, so the magnitude remains uncertain. The pattern is also not universal: for psychosocial wellbeing, 14 of 22 mean-daily comparisons found weekly scores lower than the daily mean, an anomaly the review flags for further research.

Weekly recall also does not simply reproduce the worst day. Mendoza and colleagues (2017, Clinical Trials) had 127 patients receiving cancer treatment complete 27 PRO-CTCAE items daily and at 1-, 2-, 3-, and 4-week recall. The median difference between the maximum daily rating and 1-week recall was −0.20 on the 5-point response scale (effect size −0.20; item range 0.00 to −0.33), and the gap widened at 2, 3, and 4 weeks (−0.36, −0.45, −0.47). Recalled ratings fell below the maximum daily rating but above the average daily rating. Median intraclass correlations for 1-week recall were 0.70 against the daily maximum (range 0.54 to 0.82) and 0.74 against the daily average. The authors concluded that 1-week recall corresponds best to daily reporting and is the preferred PRO-CTCAE recall period, and noted that completing daily reports may itself alter longer-period recall.

Taken together, the evidence describes a tendency rather than a law: weekly recall tends to soften the worst day while scoring higher than the average day. That makes it a different summary of the same week. Choosing between them is an endpoint decision about which summary answers the trial question.

Respondents also do not always use the window they are given. In a qualitative study of 72 adults with cancer answering physical-function items, Coles and colleagues (2025, Quality of Life Research) found that with no stated recall period the most used period was more than 4 weeks. With a 7-day instruction the most used period was 7 days, yet almost 30% of responses used periods longer than one week. The authors note that recall inaccuracy may conceal or exaggerate actual treatment effects.

The divergence evidence, with denominators and limits, is summarized below.

The criteria literature then translates direction into design. Stull (2009) and Norquist (2012) tie the choice to what the measure is meant to capture, how variable and frequent the experience is, respondent burden and ability to recall, and the trial's duration and visit schedule. Peasgood's qualitative synthesis adds that some patients preferred 7-day recall for highly variable conditions or for concepts that require integrating infrequent experiences. Two broad situations illustrate the trade-off:

• Fluctuating or episodic concepts: a single 24-hour report on a good day can miss a bad week, which is the underestimation risk Norquist describes. Repeated daily reports capture the fluctuation, but only if completion stays high, so the endpoint then depends on how missing days are handled.

• Relatively stable concepts: when day-to-day variation is low, the gap between a diary average and a weekly recall is likely to matter less, and a weekly item carries less respondent burden. That is a reasoned expectation, not a measured guarantee; it still needs evidence for the specific concept and population.

The collection designs themselves differ in what the resulting score summarizes:

FDA's December 2009 PRO guidance lists changing an instrument's recall period alongside changes such as paper-to-electronic conversion and item wording as modifications that can alter how patients respond. When an instrument is modified, the guidance says sponsors generally should provide evidence to confirm the new instrument's adequacy, while noting that not every small change requires extensive studies and that additional qualitative work may be adequate depending on the modification. FDA also states that it intends to review the protocol for the steps taken to ensure patients understood the recall period.

The final PFDD Guidance 3 (October 2025, Selecting, Developing, or Modifying Fit-for-Purpose Clinical Outcome Assessments) makes the point concrete in its discussion of modifying an existing COA. It contrasts changes unlikely to alter scores, such as moving from several items per paper page to one item per tablet screen, with changes likely to affect scores and their interpretation, giving the recall-period change as its example:

Draft PFDD Guidance 4 (April 2023, Incorporating Clinical Outcome Assessments Into Endpoints for Regulatory Decision-Making) is still a draft distributed for comment and not for implementation, but its endpoint-construction text is directly relevant. It says justification of a fixed analysis time point should take the COA's recall period into consideration. It lists the mean score over a fixed period and the maximum score during a period (for example, worst pain recorded during a 7-day period) as common summaries, and asks sponsors to consider robustness, missing-score handling, power, and interpretability. For meaningful-change work, it recommends that anchors be assessed at comparable time points and that, where applicable, the anchor's recall period be consistent with the period covered by the COA-based endpoint. An anchor asking about today does not cleanly calibrate a score that summarizes seven days.

Review scrutiny of recall periods predates the PFDD series. A 2010 ISOQOL presentation by RTI Health Solutions reproduces FDA Study Endpoints and Labeling Development (SEALD) review comments on the Dermatology Life Quality Index (DLQI). One comment notes that “justification for the recall period for the DLQI was not included in the submission” and another that “the one week recall period used in the clinical trial may introduce bias.” The slides also reproduce the concern that when a symptom varies from day to day, it is unclear whether patients can recall one or two weeks in an unbiased way. These are secondary reproductions of review comments, not FDA guidance, but they show that an unjustified recall period is a real review issue.

The same logic runs through fit-for-purpose COA selection under PFDD Guidance 3 and COA endpoint construction under draft PFDD Guidance 4: validation evidence gathered under one recall period does not automatically transfer to another.

The following is a hypothetical worked example. It describes no real sponsor, product, instrument, or regulatory decision, and its details are illustrative.

A Phase III team is testing a treatment for a symptom that fluctuates from day to day and worsens at night. It licenses a validated single-item 0–10 severity questionnaire whose validated version asks about the past 7 days, and names change from baseline in that score as the primary endpoint. The protocol's eligibility cut-off, sample-size assumptions, and responder threshold all come from earlier studies that used the 7-day version.

To reduce recall error and allow central compliance monitoring, the eCOA lead configures a daily diary with the same item reworded to the past 24 hours, and the draft SAP defines the primary score as the mean of the seven daily entries before each visit. The team treats this as the same instrument, collected more often. The decision that should have been escalated is: is the averaged daily score the same endpoint as the 7-day recalled score, what evidence must exist before the protocol, SAP, and eCOA build are locked, and who must agree to the change?

If the build ships unchanged, three problems follow, each predictable from the evidence above:

1. The numbers move: for symptom items, diary means tend to run lower than weekly recall (about three-quarters of comparisons in the systematic review). An eligibility cut-off, baseline expectations, and power assumptions taken from 7-day data no longer describe the population being scored.

2. Thresholds stop transferring: the responder threshold and meaningful-change estimates were derived on the 7-day version. A seven-day mean of daily ratings has a different distribution and less day-to-day noise, so the same point change does not carry the same meaning. If the anchor still asks about a different period, draft PFDD Guidance 4's alignment recommendation is not met either.

3. The documents describe three different measures: the protocol names the licensed 7-day instrument, the SAP derives a diary average, and the eCOA collects a 24-hour item. Under PFDD Guidance 3 the recall change may in effect create a new measure, and the instrument owner may also need to approve the modification (see COA licensing before an eCOA build). Without correspondence evidence or a documented FDA discussion, the sponsor has to defend the endpoint after data collection instead of before it.

The fix begins with a choice: keep the 7-day item as the primary measure and add the diary as a supportive or exploratory measure, or make the diary-based score the primary measure and treat it explicitly as a modified instrument. If the team chooses the second path, two steps belong before protocol and build lock:

• Step 1: Build correspondence evidence: PFDD Guidance 3 names the kinds of evidence that can help. Qualitative interviews can show how patients arrive at answers under each window. A quantitative study, for example in Phase 2 where participants complete both the diary and the 7-day item, can describe correspondence between them. Thresholds for the diary-based score should then be derived with anchors whose recall period matches the endpoint period.

• Step 2: Discuss, document, then lock: discuss the planned recall period, aggregation rule, and missing-day rule with the FDA review division, as PFDD Guidance 3 recommends, and record the outcome. Then lock the protocol endpoint wording, SAP derivation, and eCOA specification so they describe the same measure. Route any later change through eCOA protocol-amendment change control rather than a configuration ticket.

The worked example's failure starts with a choice nobody escalated. Working through five questions before the protocol, SAP, and eCOA configuration are locked forces that choice into the open:

flowchart TD
  Q1["1. How variable is the concept day to day?"] --> Q2["2. What do the visit schedule and analysis window require?"]
  Q2 --> Q3["3. Which summary defines the endpoint: daily mean, worst day, or recalled week?"]
  Q3 --> Q4["4. What burden and missing-day risk does the design add?"]
  Q4 --> Q5["5. Do anchors and thresholds share the same period?"]
  Q5 --> D{"Is the recall period different from the validated version?"}
  D -->|No| Lock["Lock protocol, SAP, and eCOA specification"]
  D -->|Yes| Mod["Treat as a modified instrument: gather evidence and discuss with FDA"]
  Mod --> Lock

Each question maps to a document that will be hard to change once collection begins:

1. How variable is the concept day to day? Use qualitative and natural-history evidence. High variability strengthens the case for repeated daily reports, or for worst-experience framing, which FDA's 2009 guidance names as a way to improve retrospective reports. Low variability weakens the case for daily burden. Record the rationale, because FDA's 2009 guidance asks sponsors to justify the recall period against the variability, duration, frequency, and intensity of the concept.

2. What do the visit schedule and analysis window require? Norquist ties recall to trial duration and scheduled visits, and draft PFDD Guidance 4 says a fixed analysis time point should take the recall period into consideration. A 7-day item completed at a visit covers the preceding week; a diary needs a defined window relative to each visit.

3. Which summary defines the endpoint? If daily data are collected, the protocol and SAP must name the operator: mean, worst day, or another prespecified summary. Draft PFDD Guidance 4 lists mean and maximum scores over a period as common choices and asks sponsors to weigh each summary's robustness, missing-score handling, power, and interpretability; each needs its own COA measurement properties evidence.

4. What burden and missing-day risk does the design add? Daily collection adds respondent burden, which PFDD Guidance 3 asks sponsors to minimize, and creates missing days. Prespecify the minimum number of completed days that makes a weekly block evaluable and connect that rule to the estimand under the ICH E9(R1) estimands framework before unblinding.

5. Do anchors and thresholds share the same period? Following draft PFDD Guidance 4, anchors used to interpret meaningful change should cover the same period as the endpoint. A threshold derived on a 7-day item should not be applied to a diary mean, or the reverse, without new evidence.

Recall-period decisions interact with collection mode. A web backup used during device failure, for example, must present the same window and instructions as the primary device; see eCOA mode measurement comparability for when mixed modes need their own evidence.

Some trials collect both streams, such as a daily symptom diary plus a 7-day recall item at clinic visits. When the two disagree, the first question is whether the gap is the expected pattern (weekly higher than the daily mean, lower than the worst day) or a sign of a data problem. Expected divergence is a property of the two measures, not a defect to be corrected.

There is also no accepted conversion factor. Arizmendi and colleagues (2024, Value in Health) reviewed 30 studies that gave participants the same measure with different recall periods, ranging from momentary to 6 months. They found that analytic approaches varied across relative-agreement, absolute-agreement, and combined methods, and that there is no consensus on how to quantitatively evaluate consistency between recall periods. The authors offered starting-point recommendations rather than a standard. A post-hoc multiplier or regression that rescales one measure into the other therefore lacks an accepted methodological basis.

A defensible mid-study response has four parts:

• Do not re-derive the primary endpoint in reaction to the gap: changing the aggregation rule or primary score because of observed results undermines prespecification. If a change is genuinely necessary, document the rationale and implement it through a protocol amendment and SAP revision before unblinding, with regulator discussion where it affects the primary endpoint. See what must be locked before unblinding.

• Rule out data problems first: apparent divergence can come from late or backfilled diary entries, device clock errors, or visit-date mismatches rather than from recall. Check timestamps and synchronization records using the principles in offline eCOA data collection and eCOA and EDC timestamp reconciliation.

• Run the prespecified sensitivity analyses: vary the minimum-completed-days rule and compare mean and worst-day summaries, following estimand-aligned missing-data sensitivity analysis. These analyses test robustness; they do not turn one measure into the other.

• Report the two measures as distinct: in the clinical study report, state each measure's recall period, schedule, and aggregation rule, and interpret each against the question it answers.

The divergence figures come from heterogeneous studies, instruments, and comparison definitions (mean daily, maximum daily, same-day), so they describe a pattern, not a transferable effect size for any particular instrument. The PRO-CTCAE results come from one cancer population and a 27-item subset, and the authors acknowledge that daily completion may itself change weekly recall. The qualitative evidence on recall-window adherence covers physical-function items in adults with cancer only. The SEALD comments are known through a secondary conference presentation. Draft PFDD Guidance 4 may change when finalized. None of this weakens the core operational point: a recall-period change is a measurement decision that needs evidence, documentation, and, where FDA guidance recommends it, discussion with the agency before data collection begins.