Why the Sampling Schedule Is Part of the Design
A longitudinal study does not merely collect samples “over time.” Its schedule determines which changes in cellular response can be observed, how those changes are anchored, and whether the same donors contribute information at the planned times. In single-cell RNA sequencing, the schedule also has to be feasible for donors and for sample handling. A carefully planned sequence of visits is therefore part of the scientific design, not an administrative detail.
The core planning task is to define a baseline and a set of follow-up times in relation to a meaningful event or exposure. For example, follow-up could be measured from treatment initiation, a controlled challenge, or a prespecified clinical event. Without a clear anchor, “week 2” can refer to different biological durations for different donors. The schedule should also state what counts as an acceptable visit time, how missed visits will be recorded, and how many donors are expected to remain available through each timepoint.
The donor is the biological unit that is followed. Cells measured from one donor at one visit are subsamples of that donor’s specimen; they do not turn one donor into many independent longitudinal participants. Earlier material on experimental units and paired designs establishes why this distinction matters. Repeated-timepoint planning extends the same within-donor idea beyond two observations, but this tutorial concerns the schedule and its operational consequences, not a particular trajectory model or test.
Specify the Time Axis Before Selecting Visits
First name the anchor and the meaning of baseline. Baseline may be a sample collected before an intervention, or a prespecified initial observation in a natural-history study. State whether baseline is taken on the same day as the anchor, immediately before it, or within a defined interval. If a donor’s baseline can occur after the exposure has begun, it may no longer represent the intended pre-exposure state.
Next, express each follow-up time relative to the anchor. A schedule might say “baseline before dosing, then 6 hours, 24 hours, day 7, and day 28 after the first dose.” For each planned time, give a visit window, such as 24 hours after dosing with an allowed deviation of plus or minus 2 hours. The window should be narrow enough that samples grouped under one nominal time remain scientifically comparable, while allowing for realistic collection logistics.
Choose times to match the changes the study is intended to observe. Closely spaced early samples may be useful when a response is expected to rise and fall quickly; more widely spaced later samples may be more appropriate for sustained changes. These are design judgments that require prior biological knowledge or pilot information. A schedule cannot recover a transient response that occurs entirely between visits, and adding many visits without a clear purpose can increase burden and missingness.
For each visit, specify the specimen, collection procedure, and processing plan sufficiently to make the measurement comparable across time. In single-cell work, record collection time, time from collection to processing or preservation, and any protocol deviations. If one timepoint is systematically processed under different conditions from the others, an apparent time pattern may reflect the measurement process rather than biology. The schedule should make such risks visible, without assuming that later analysis can automatically remove them.
Worked Example: Anchoring a Baseline and Follow-Up Windows
A hypothetical study follows donors after a standardized immune challenge. The scientific aim is to observe an early response and a later response. Draft a schedule with an interpretable baseline, explicit times, and feasible windows.
Step 1: Define the anchor. Let time zero be the documented start of the challenge. Baseline is collected within 30 minutes before time zero. This ensures that the baseline sample is pre-challenge under the stated protocol.
Step 2: Select follow-up times for the intended response phases. Suppose prior biological knowledge suggests an early response over hours and a later response over days. A hypothetical plan is 6 hours (window 5–7 hours), 24 hours (window 22–26 hours), and day 7 (window day 6 to day 8), all measured from time zero.
Step 3: Check what the schedule can support. The 6-hour and 24-hour visits can distinguish measurements at those two planned times, but they do not reveal the exact peak between them. The day-7 visit describes a later state; it cannot establish when a change began between day 1 and day 7.
Step 4: State the operational record. For every visit, record the actual collection time and the difference from its nominal time. A sample collected at 9 hours is outside the 6-hour window and should be flagged as a timing deviation, not silently relabelled as a 6-hour sample. The plan is interpretable because the anchor, nominal times, and acceptance windows are explicit.
Plan Donor Retention Separately From Cell Yield
A donor may miss a visit, withdraw, or provide a specimen that fails quality control. These are distinct events and should be recorded separately. A donor who attends but yields too few usable cells has not necessarily been lost to follow-up; conversely, a large cell yield from the remaining donors does not replace donors missing at a timepoint. The schedule should define what constitutes an available donor sample and track availability at every planned visit.
Retention planning requires an explicit assumption about the probability that an enrolled donor will provide the samples needed for the intended comparison. Let \(N\) be the number enrolled and \(q\) the probability that one donor is retained with usable samples through the final required timepoint. If donor outcomes are independent and each donor has the same retention probability \(q\), the number \(C\) of complete donors follows a binomial distribution:
The expected number \(Nq\) is useful for a first calculation, but it is not a guarantee. If the goal is to have at least \(m\) complete donors, an enrollment choice should ideally satisfy a probability target \(P(C \geq m)\), not only \(Nq \geq m\). These calculations concern retention, not statistical power; the appropriate target number of donors must come from the scientific design and analysis plan.
When retention is expected to decline across visits, define the probabilities carefully. If a donor has probability \(r_j\) of remaining available from visit \(j-1\) through visit \(j\), conditional on being available at visit \(j-1\), and these conditional probabilities are assumed equal across donors, then the probability of remaining through \(J\) follow-ups is
This multiplication is appropriate for conditional survival through successive intervals. It is not justified by treating visits as independent if the process is actually dropout: once a donor withdraws, later visits are unavailable. Use retention information from a comparable setting where possible, and state uncertainty when that information is weak.
Worked Example: Enrollment Under a Retention Assumption
A hypothetical pilot suggests that 80% of enrolled donors are expected to provide usable samples at all required visits. The study design calls for at least 5 complete donors, and the team considers enrolling 7. Under the assumptions of independent donor outcomes and a common retention probability \(q=0.80\), calculate the expected number and the probability of meeting the target.
The expected number complete is
The expectation exceeds 5, but this alone does not show how likely the target is. Since \(C\sim\operatorname{Binomial}(7,0.80)\), the probability of at least 5 complete donors is
Substituting each term gives
Thus, under the stated assumptions, enrolling 7 gives about an 85.2% probability of retaining at least 5 complete donors. This is not a guarantee, and the result depends on the 80% retention assumption being appropriate. If the study requires a higher assurance, the team should evaluate a larger enrollment using the same binomial calculation rather than treating 5.60 as a certain count.
Compare Schedules for Information and Feasibility
A schedule should be judged by the information its timepoints can provide and the likelihood that donors can complete them. A sparse schedule reduces visit burden and may improve retention, but it can miss short-lived changes. A dense schedule can locate changes more precisely in time, but it increases cost, participant burden, and opportunities for missed visits or inconsistent handling. The right choice is not automatically the schedule with the most visits.
| Schedule feature | Potential advantage | Risk to assess |
|---|---|---|
| Baseline plus a few prespecified follow-ups | Lower burden and simpler logistics | Changes between visits may be unobserved |
| Closely spaced early follow-ups | Better temporal resolution for rapid responses | Early visits may be difficult to coordinate consistently |
| Many visits across the full study | More observations of each donor over time | Higher burden, missing visits, and technical variation |
| Flexible, donor-specific visit timing | May accommodate practical constraints | Actual timing may become entangled with donor characteristics or response |
Before finalizing the schedule, ask whether each timepoint is needed to address the target question, whether the window is feasible across donors, and whether the same collection and processing procedures can be applied throughout. If an exact time is biologically important, define how a visit outside its window will be treated in the records and analysis plan. If timing flexibility is unavoidable, retain actual timestamps rather than only nominal visit labels.
Worked Example: Choosing Between Sparse and Dense Follow-Up
A hypothetical study expects a cellular response to change substantially during the first day after exposure and to be more stable during the following month. The team can support either three follow-ups or six. Compare two candidate schedules and select a defensible plan.
Candidate A is baseline, 24 hours, day 7, and day 28. Candidate B is baseline, 2, 6, 12, and 24 hours, then days 7 and 28. Candidate B adds three early visits, which could better describe the early time course if the response changes rapidly during those hours. But it requires more collections and may reduce completion if donors cannot reliably attend.
Suppose a pilot indicates that the 6-hour and 12-hour samples are both operationally difficult, while the response is expected to be mostly stable after 24 hours. A defensible revised schedule could use baseline, 6 hours, 24 hours, day 7, and day 28, with explicit windows for each. This preserves an early observation, a next-day observation, and later observations while removing one difficult visit. The choice is justified only if those times answer the study question; it cannot establish the exact response peak or exclude a brief change between visits.
The comparison shows why feasibility is a scientific concern: a theoretically informative visit that many donors miss may contribute less usable longitudinal information than a smaller, reliably completed schedule. The team should use pilot completion and sample-quality records to revisit the plan before recruitment is complete.
Diagnostics, Failure Modes, and Limitations
Track enrollment, attendance, usable specimens, and donor withdrawal separately by planned timepoint. A simple visit-completion table can reveal whether later visits have progressively fewer donors or whether a particular timepoint is unusually difficult. Compare actual collection times with the prespecified windows, and summarize deviations. Review sample-quality failures by timepoint as well: a pattern of lower quality at one visit can compromise interpretation even when attendance is high.
These diagnostics do not prove why observations are missing. If donors with particular characteristics or response patterns are more likely to miss visits, the observed set of later samples may no longer represent the original cohort. Do not assume that missing observations are harmless simply because the schedule was balanced on paper. Document reasons for missingness when feasible and report the number of donors contributing at each timepoint.
- Unanchored timing: “one week later” is ambiguous unless the reference event and timing convention are specified.
- Overly broad windows: samples labelled with the same nominal time may represent meaningfully different biological durations.
- Cell counts mistaken for donor retention: thousands of cells from a few retained donors do not restore missing donor-level follow-up.
- Silent relabelling: assigning an out-of-window sample to its planned time without recording the actual time conceals a design deviation.
- Retention treated as certain: an expected complete count is an average, not a minimum guaranteed outcome.
A schedule also has limits. No set of discrete samples captures the continuous response between visits, and retention estimates may be uncertain before the study begins. A plan can reduce avoidable missingness but cannot ensure that every donor remains available or that every specimen passes quality control. Report the schedule as planned and describe departures from it transparently.
Check Your Understanding
Answer each question, then compare with the solutions below.
- A study collects baseline samples after an intervention has already begun. What interpretation problem can result?
- With \(N=10\) enrolled donors and a common probability \(q=0.70\) of completing all required visits, what is the expected number of complete donors?
- Why should actual collection times be recorded when samples fall outside a planned visit window?
- A donor contributes 8,000 usable cells at baseline, but misses every follow-up. Is this donor complete for a baseline-to-follow-up comparison? Explain.
- Name one advantage and one limitation of adding several early follow-up visits.
Solutions. (1) The baseline may reflect the intervention rather than a pre-intervention state, so it may not serve as the intended reference. (2) \(E(C)=Nq=10(0.70)=7.00\) complete donors on average; this is not a guarantee of at least 7. (3) Actual times show the size and direction of timing deviations and help distinguish samples collected at the intended biological time from those collected outside the window. (4) No. Completeness for that comparison requires usable samples at baseline and the required follow-up timepoints; cell yield at baseline does not replace missing visits. (5) Several early visits can improve temporal resolution for a rapid response, but they increase burden and may lower visit completion or consistency.