Reading a Biomarker Timeline
A biomarker that declines before hospitalization can be clinically suggestive, but a timeline alone does not establish why it declined or whether the decline caused the event. The change could reflect worsening underlying health, a short-term process preceding hospitalization, treatment, measurement variability, or some combination. The timing and availability of measurements matter too: a value measured at a routine visit is not the same as a value measured at admission, and no measurement is available for a time when no visit occurred.
This tutorial uses a hypothetical example in which lower serum albumin is treated as a possible sign of worsening health. The values are illustrative, not clinical guidance. The goal is to read the observed pattern carefully: describe the measurements, place them relative to hospitalization, and separate plausible explanations from conclusions the data support.
A hospitalization is an event time, not a biomarker measurement. Follow-up may end at hospitalization, and a patient may have no routine visit immediately beforehand. If a patient leaves routine follow-up for another reason, that is a different observation process and should not automatically be treated as a hospitalization. The first task is therefore to establish which times are observed, which are missing, and what event—if any—occurred.
Worked Example: Quantifying a Decline Before Hospitalization
Worked Example: Quantifying a Decline Before Hospitalization
Suppose a hypothetical patient has serum albumin measurements of 3.8 g/dL on day 0, 3.4 g/dL on day 30, and 3.0 g/dL on day 50. The patient is hospitalized on day 54. Assume for this descriptive calculation that the dates and values are recorded correctly and that the same measurement units and comparable assay procedures were used at each visit.
From day 0 to day 50, the observed change is \(3.0-3.8=-0.8\) g/dL over 50 days. The average rate of change over that observed interval is
The negative sign indicates a decline. In words, the measured values fell by an average of 0.016 g/dL per day between the first and last observed visits. This is an interval average, not a claim that the value fell by exactly that amount on every day.
The last measurement preceded hospitalization by \(54-50=4\) days. Thus the observed decline occurred before the recorded event, but the data do not give the albumin value on the hospitalization date. Without additional measurements, we cannot tell whether the decline continued, leveled off, or changed in the four-day interval. Nor does the sequence show that low albumin caused the hospitalization.
What a Rate of Change Does—and Does Not—Describe
For two observed measurements at times \(t_a\) and \(t_b\), where \(t_b>t_a\), the finite-difference rate is
This rate summarizes the net change between the two observations per unit of time. Its units combine the biomarker’s units with the time unit; for example, g/dL per day. The calculation requires comparable measurements and correctly recorded dates. Interpreting the rate as a constant daily decline would require an additional assumption of linear change over that interval. The calculation by itself makes no such assumption.
The distinction follows directly from the formula: it uses only the two endpoint values and the elapsed time. Many different paths between those endpoints have the same rate. A gradual decline, a sudden drop followed by stability, and a temporary dip followed by recovery could all produce the same endpoint difference. With sparse visits, the detailed shape of the trajectory is not identified by the observed values.
One may draw a straight line between visits to help visualize the average change, but that line is an interpolation, not an additional observation. Extending it past the last visit is an extrapolation and requires further assumptions. In particular, projecting a pre-event decline to the hospitalization date is not justified merely because the last two measurements were decreasing.
Worked Example: A Missed Visit Near the Event
Consider another hypothetical patient with albumin of 3.7 g/dL on day 0 and 3.3 g/dL on day 30. A routine visit is planned for day 45, but the patient is hospitalized on day 43. No day-45 biomarker value is recorded.
The observed rate from day 0 to day 30 is
This calculation supports a measured decline over the first 30 days. It does not provide a day-43 value: the planned day-45 measurement never occurred, and hospitalization happened before that visit. Carrying the day-30 value forward would assert, without evidence, that albumin stayed at 3.3 g/dL through day 43. Extending the calculated rate to day 43 would instead impose a constant-rate assumption. Neither is an observed result.
Here hospitalization explains why the planned visit was not completed, but it does not reveal the unmeasured biomarker value. The timing suggests that the missing measurement may be related to the patient’s health and event, so it should not casually be treated as an ordinary absent value. It remains unknown whether the patient’s albumin declined further before hospitalization.
Several Explanations Can Fit the Same Timeline
A decline shortly before an event is compatible with several scientific stories. One possibility is that worsening disease changes the biomarker and also raises the chance of hospitalization. Another is that a process developing shortly before the event changes the biomarker; the marker then acts as an indicator of that process rather than its cause. This is one reason a pre-event decline may be informative without being causal.
Other explanations also deserve attention. A treatment started during follow-up could change the marker. A change in hydration, acute inflammation, assay procedures, or specimen handling could affect a measurement. Random measurement error can make one value unusually low. These possibilities cannot be resolved just by noticing that the final recorded value was lower than the first.
Worked Example: The Same Decline, Different Event Timing
Suppose two hypothetical patients each have albumin measurements of 3.8 g/dL on day 0, 3.5 g/dL on day 30, and 3.2 g/dL on day 60. Patient A is hospitalized on day 63. Patient B remains under observation without hospitalization through day 120.
For each patient, the average rate from day 0 to day 60 is
Both patients therefore have the same observed average rate over those 60 days. For Patient A, the final measurement was three days before hospitalization; for Patient B, the same measurements were followed by at least 60 days without a recorded hospitalization. The shared decline alone does not determine when an event will occur or show that the decline is sufficient to produce one.
This comparison is descriptive, not a risk estimate. It contains only two patients, and their observation periods differ. It illustrates why the event time must be read alongside the biomarker history, while also showing why an individual timeline cannot establish the strength of a marker–event relationship in a population.
How to Examine a Timeline
A disciplined reading separates what is directly recorded from what is hypothesized. The following sequence is useful when reviewing a short clinical history:
Identify the follow-up origin, the measurement dates, and the hospitalization date. Check that intervals are expressed in consistent units.
State which values rose, fell, or stayed similar, and over what interval. Calculate a rate only for times with observed endpoints.
Mark planned or expected visits with no recorded value. Ask whether the absence followed hospitalization, withdrawal, administrative loss, or another known reason.
Consider underlying progression, a process close to the event, treatment, and measurement variability. Treat them as hypotheses unless other evidence distinguishes them.
Say what the measurements support and what remains unobserved. Do not turn temporal ordering into a causal conclusion.
Comparison, Diagnostics, and Limitations
A baseline-only description uses the starting biomarker but ignores subsequent change. A last-observation description uses a later value but can hide when it was measured relative to the event. A timeline puts repeated measurements and event timing together, making their ordering visible. That is a useful descriptive advantage, but it is not by itself a statistical model of the longitudinal and event processes. Joint modeling is a framework for analyzing those processes together; a single plotted timeline does not estimate its parameters or establish its assumptions.
Check whether measurement dates are accurate, whether the assay or clinical measurement process changed, and whether values were taken before or after treatment at a visit. Note whether the final measurement is close to the event and whether follow-up ended because of the event or another reason. With few measurements, irregular intervals, or substantial measurement noise, an apparent slope may be unstable. If the marker has a clinically meaningful threshold, a trend and a threshold crossing are distinct facts: a decline does not necessarily mean the threshold was crossed.
The main limitation is that observed values are snapshots. Sparse measurements do not reveal the full path between visits, and a value missing near hospitalization cannot be reconstructed from the event time alone. A timeline can motivate questions about disease progression, event risk, or observation patterns; answering those questions requires appropriate data and analysis beyond the descriptive calculation.
Common Mistake
Check Your Understanding
Answer each question from the information given. Solutions follow the questions.
- A hypothetical patient’s marker is 5.0 units on day 10 and 4.4 units on day 40. Calculate the average rate of change and state what it describes.
- A patient’s last measured value is recorded 12 days before hospitalization. Is the marker value on the hospitalization date known? Explain.
- A planned visit was missed because the patient was hospitalized the day before. Why should the missing measurement not automatically be treated as an ordinary absent value?
- Two patients have identical observed declines, but only one is hospitalized soon afterward. What can and cannot be concluded from that comparison?
Solutions. (1) The elapsed time is \(40-10=30\) days, so the rate is \((4.4-5.0)/30=-0.0200\) units per day. It describes the average net change between the two observed visits, not a known daily path. (2) No. The measured value is 12 days before hospitalization; the value on the event date is unobserved unless it was separately measured. (3) The absence is temporally tied to the event and may reflect health-related follow-up rather than an unrelated recording gap; the unmeasured value remains unknown. (4) The timelines show that the same observed decline can precede different event histories. They do not, by themselves, establish an individual event prediction, a population-level risk relationship, or a causal effect of the marker.