Why Visit Timing Needs a Source
A record may show that a patient was seen on day 37 of follow-up, but that date alone does not say why the visit occurred then. It could be a protocol visit set for day 37, an appointment moved from day 35 because the clinic was closed, or an appointment the patient requested after symptoms worsened. These mechanisms have different implications for interpreting longitudinal measurements.
The key task is to distinguish the planned schedule from the realized encounter process, and then determine what generated any difference between them. Timing that is fixed by protocol, conditional on the study design, is not itself a random health signal. A timing change may nevertheless be associated with health if health influenced the change, or if circumstances that affected timing also relate to health. Conversely, a visit that appears irregular is not automatically informative about health.
These are classifications of the source of timing, not conclusions about whether the data can be analyzed while ignoring the visit process. That question depends on what information determined the schedule or change, what health information was available, and which measurements were observed. The earlier material on observation intensities describes how visit timing can be represented as a stochastic process; here the emphasis is on identifying the mechanism behind particular recorded times.
Classifying the Three Sources
A useful classification starts with the appointment history rather than the spacing between visits. For each encounter, ask what was planned, who changed the plan, when the change occurred, and what reason was documented. A short gap does not prove a visit was patient-initiated, and a long gap does not prove it was administrative.
| Source | What determines the timing? | Potential health relevance |
|---|---|---|
| Protocol-scheduled | A prespecified study or care rule, perhaps conditional on treatment or known risk group | Usually no random timing information conditional on the protocol and its inputs; protocol assignment may still be related to health |
| Administrative rescheduling | Operational or logistical changes, such as clinic capacity, holidays, or staff availability | May be unrelated to health after relevant factors are accounted for, but can be health-associated if reasons or affected populations are health-related |
| Patient-initiated | A request by the patient, potentially in response to symptoms, concerns, or other health changes | Can carry health information, particularly when health concerns affect the request; the label alone does not establish this link |
“Conditional on the protocol and its inputs” matters. For example, a protocol may schedule more frequent monitoring for a high-risk group. Within each group, the planned dates might be fixed; across groups, however, the assigned schedule reflects information related to health. If the target analysis treats schedule assignment as part of the design, that distinction should be preserved rather than calling all scheduled timing uninformative.
Administrative changes also require care. A clinic closure may shift appointments without regard to any patient’s health. But an appointment changed because a patient was hospitalized, or because staff prioritized patients with urgent clinical needs, has a different interpretation. “Administrative” describes how the change was recorded or processed; it does not guarantee independence from health.
Worked Example: Separating Planned Dates From Actual Dates
Suppose a hypothetical protocol plans visits at follow-up days 0, 30, 60, and 90. A participant attends on days 0, 30, 64, and 90. The appointment record says the day-60 visit was moved to day 64 because the clinic was closed on day 60. Classify the encounters and calculate the timing change.
The visits on days 0, 30, and 90 are protocol-scheduled visits that occurred on their planned dates. The encounter on day 64 is also the realization of a protocol-scheduled visit, but it was administratively rescheduled. It should not be reclassified as patient-initiated merely because the actual date differs from the plan.
The planned-to-actual shift is \(64-60=4\) days. Thus the visit occurred 4 days later than planned. This calculation describes the deviation; it does not show that the deviation was related to health. The documented clinic closure supports an operational explanation, but an assessment of health relevance would still consider whether the closure or its consequences affected particular patients differently.
When Timing Can Carry Health Information
Let \(N_i^V(t)\) denote the cumulative number of realized visits for patient \(i\) by time \(t\). Let \(H_i(t)\) denote the health history available up to time \(t\), including relevant observed measurements, symptoms, and treatment history. An observation intensity describes the instantaneous rate of a visit given the history and other specified information. If the intensity depends on health history, then visit timing may be informative for longitudinal analysis: the process determining when measurements are collected is associated with the health process being measured.
Here \(\mathcal{S}_i\) represents the planned schedule and \(Z_i\) represents other recorded factors relevant to visits. The notation \(H_i(t-)\) indicates health information available just before time \(t\). A dependence of visit intensity on observed health history is one possible source of health-related timing. Timing can also depend on health that is not recorded in \(H_i(t-)\), such as a symptom that prompted an appointment but was not captured until the visit. In that case, the visit record alone may not reveal the trigger.
For a protocol schedule fixed at dates \(s_{i1},\ldots,s_{im}\), conditional on the schedule and its inputs, the planned times have no residual timing variation: the probability that the \(k\)th planned time equals \(s_{ik}\) is 1. Consequently, the planned dates do not behave like random visit times that fluctuate in response to a patient’s current condition. This statement concerns the plan, not whether the patient attends, whether the clinic can deliver the visit, or whether additional visits occur.
A deviation from that plan creates a separate question: what mechanism produced it? The short justification is that a fixed plan has no random timing variation left to explain after conditioning on the plan, whereas a variable actual time may reflect a change in the plan, a missed visit, or an additional encounter. Only the latter sources can introduce timing variation beyond the protocol dates, and whether that variation contains health information depends on its cause.
Worked Example: A Patient Request After New Symptoms
In a hypothetical study, participants are scheduled for measurements every 30 days. One participant attends on day 0 and day 30 as planned, then calls on day 43 to request an additional appointment because of worsening breathlessness. The clinician measures the biomarker at that appointment.
The day-43 encounter is patient-initiated, not a rescheduling of the day-60 protocol visit: it is an additional appointment requested between planned visits. The stated reason directly links the request to a health change. The measurement at day 43 is therefore collected at a time that may be related to the participant’s health, rather than at a time determined solely by the fixed 30-day schedule.
This classification does not establish that the biomarker value is higher, nor that breathlessness caused a particular biomarker value. It establishes that the timing of measurement was prompted by a health concern. If symptom severity also relates to the biomarker, analyzing the observed measurements as though their times were unrelated to health could distort a summary of the biomarker trajectory.
A Practical Classification Process
A single patient’s records may contain all three sources. Classification should therefore be done at the encounter level and should preserve the distinction between an appointment’s original purpose and what happened to it. A useful record has separate fields for planned date, actual date, encounter type, initiator, reason for change or request, and whether the visit was completed.
Identify whether a visit was due under a protocol or care schedule, and record the planned date and any rules that determine it.
Determine whether the planned visit occurred as scheduled, was moved, was missed, or was accompanied by an extra encounter. Keep the planned date distinct from the actual date.
Use appointment logs or clinical records to distinguish a clinic-initiated change from a patient request. Do not infer the source solely from the size of the time gap.
Ask whether the reason was operational, patient preference, symptoms, clinical concern, or another factor, and whether that factor may be associated with health or measurement availability.
If the record does not establish why timing changed, label the source as unknown rather than assigning a confident mechanism. Missing reason information can itself limit what can be concluded.
Worked Example: Distinguishing Rescheduling From an Extra Visit
Consider hypothetical records for one participant: a protocol appointment planned for day 45 is moved by the clinic to day 48 because the assigned clinician is unavailable. Separately, the participant contacts the clinic on day 52 about a new concern and is seen on day 54. The next protocol visit remains planned for day 75.
The day-48 encounter is an administratively rescheduled protocol visit. Its shift is \(48-45=3\) days. The day-54 encounter is patient-initiated and additional; it does not replace the day-75 protocol appointment. The day-75 visit remains a planned protocol visit unless later records show that its status changed.
The first timing change has a documented staffing explanation, while the second has a patient health concern as its documented trigger. These facts support different assessments of possible health relevance. They do not prove that the administrative change is independent of health in a broader population, or that the patient-initiated timing is associated with a particular biomarker outcome.
Assumptions, Comparisons, and Failure Modes
A simple assumption sometimes used for interpretation is that, conditional on recorded design variables and relevant history, administrative changes do not depend on the patient’s current or unobserved health. Under that assumption, administrative timing variation may add no health-related selection beyond those variables. It is an assumption to assess, not a consequence of calling a change “administrative.” By contrast, a patient-requested appointment plausibly depends on health when the request is prompted by symptoms, but the degree of dependence must be supported by clinical or process information.
The distinction can be summarized as follows:
| Approach to timing | Strength | Limitation |
|---|---|---|
| Use only protocol dates | Preserves the planned measurement calendar and clarifies intended timing | Does not explain missed, moved, or additional visits, or the actual times at which measurements were obtained |
| Classify each realized visit by source | Separates planned encounters, operational changes, and patient requests | Depends on accurate records; reasons may be incomplete or ambiguous |
| Treat all actual visits as one undifferentiated process | Can be a compact description of encounter frequency | Mixes mechanisms with different health relationships and may obscure why measurements were observed when they were |
Common warning signs include many unexplained deviations from scheduled dates, patient-request reasons recorded only after the visit, systematically different rescheduling patterns across health groups, and missed visits concentrated among patients with deteriorating health. Such patterns motivate questions about the mechanism, but they do not by themselves establish the cause of timing differences. Comparing documented reasons with appointment logs, protocol records, and available health history is important.
A common mistake is to reason, “The study protocol specified the visits, so the visit process is noninformative.” The correction is to distinguish planned dates from realized encounters. A fixed plan has no random timing variation conditional on its inputs, but attendance, rescheduling, and extra visits may still depend on health. A second mistake is to label every visit outside the protocol as patient-initiated. The reason might instead be clinic operations, clinician recommendation, an emergency pathway, or an unknown source.
These classifications also have limits. Reasons can be misreported, a patient’s unrecorded symptoms may prompt a request, and administrative constraints may affect some health groups more than others. A recorded association between timing and health does not establish that timing caused a health change. Classification clarifies plausible mechanisms and informs modeling decisions; it does not remove uncertainty about unobserved triggers.
Check Your Understanding
Answer each question, then compare with the solutions below.
- A study visit planned for day 80 occurs on day 84 because the clinic was closed. What is the classification, and how large is the shift?
- A participant requests an extra appointment after developing a new symptom. Why might the timing be health-related, and what does that fact not prove?
- Why is it incorrect to treat a fixed protocol schedule as proof that all realized visit times are noninformative?
- A visit is 12 days earlier than usual, but the record has no reason field. How should it be classified?
- Give one way an ostensibly administrative rescheduling process could still be related to health.
Solutions. (1) It is an administratively rescheduled protocol visit, shifted by \(84-80=4\) days. (2) The symptom may have prompted the request, so measurement timing may depend on health; this does not prove the symptom caused a biomarker value or establish the size of any association. (3) The protocol fixes planned dates conditional on its inputs, but attendance, deviations, and additional encounters can depend on health. (4) Record the deviation and classify its source as unknown until evidence identifies it; do not infer the initiator from the interval alone. (5) For example, staff may prioritize patients with urgent clinical needs, so operational rescheduling could vary with health status.