Trend & change over time
When the finding is the change rather than the value, two things decide whether the clinician reads it correctly: how the change is scaled, and whether the measurements were ever comparable in the first place. Neither is visible in the number that gets reported.
Overview
A 7 mm nodule is not a finding. A 7 mm nodule that was 6 mm eighteen months ago might be. The clinical object is the difference, and a difference has properties a value does not: it depends on a baseline someone chose, an interval someone measured, and an assumption that the two ends are commensurable.
This page applies Data visualisation to the case where the change carries the meaning. It does not restate the charting rules — fixed axes, gaps, projections and comparability all live there. What it adds is the thing charts almost universally omit, which is how much change is more than noise.
Change is the finding
The delta, the interval and the rate are stated as clinical facts, not left for the reader to compute from two numbers.
Noise is stated
A change smaller than the measurement's own variability is not a change, and the interface says so rather than drawing it.
Comparable or not compared
Where the ends of a trend were produced differently, the trend is qualified — or it is not offered.
Anatomy
| Part | Rule |
|---|---|
| Current value | With unit and provenance, exactly as Key–value pair requires. |
| Change | Signed, with its unit. Never a bare percentage — a 40% increase on a small absolute value is a different clinical object from the same percentage on a large one. |
| Baseline | Named and dated. "Change from baseline" is meaningless without which measurement the baseline is, and baselines get silently reset by data migrations. |
| Interval | The elapsed time, in the units the clinical guidance uses. A delta without an interval cannot be assessed at all. |
| Variability | Required where known. The measurement's own reproducibility, and whether the change exceeds it. |
| Comparability | Any method, protocol, assay or unit change between the compared points, stated adjacent to the change rather than in the chart footnotes. |
How much change is a change
Every measurement has a reproducibility, and most clinical software draws changes far smaller than it as though they were real. A 1.2 mm difference on a measurement with ±1.5 mm inter-reader variability is not growth — it is the same nodule measured twice. Rendering it as a rising line manufactures progression, and progression manufactures scans, biopsies and appointments.
Where the variability is known, the interface states it and says whether the change exceeds it. Where it is not known, that is also stated — an unknown reproducibility is a real limitation and not a reason to imply precision.
The numbers are still there and still checkable. What has changed is the headline: the interface declines to call this growth, because on this measurement it is not distinguishable from none.
- Never suppress the underlying numbers. Declining to call it a change is a statement about interpretation, not a reason to hide data.
- Never round the change into significance. Reporting "+2 mm" for a 1.6 mm difference on a ±1.5 mm measurement invents precision twice over.
- Rate needs enough points. A doubling time computed from two measurements is arithmetic, not a trend; state how many points it rests on.
- Percentages carry their absolute values. Always both, never the percentage alone.
- Regression to the mean is not improvement. A value that was extreme and is now less so may be neither better nor worse, and a two-point trend cannot tell the difference.
Which baseline
"Change from baseline" quietly assumes everyone means the same measurement by baseline. Across three years, four clinicians and a data migration, they do not.
| Baseline | When it is the right one |
|---|---|
| First recorded | Surveillance of a known lesion — the question is total change since discovery. |
| Immediately previous | Interval change — the question is what happened since the last look. Most guidance is written against this one. |
| Nadir or peak | Treatment response, where the comparison is against the best or worst achieved. |
| Pre-treatment | Where an intervention resets what "normal for this patient" means. |
- The baseline is named and dated in the interface, never implied by the phrase "from baseline".
- Offer at most two comparisons at once — total change and interval change. Four simultaneous deltas is a puzzle, not a summary.
- The baseline never changes silently. If a migration, a re-read or a deletion moves it, that is announced on the trend, because every historical delta just changed.
- A superseded measurement stays visible. Where a value was corrected, both remain — see Model version change.
Aggregation and what it hides
Long series get aggregated because they have to be. Aggregation is a lossy operation performed on clinical data, and the loss has to be declared.
Time axis to scale, so the 14-month gap is visibly a gap. Four points, drawn individually — nothing is averaged, and the dashed segment spans both the gap and the scanner change.
- State the window and the operation — "daily mean", "hourly maximum". A mean and a maximum tell opposite stories about the same day.
- Aggregate toward the clinical concern. Where lows are the hazard, show the minimum; an average that never dips is the wrong summary of a night with three excursions.
- Excursions survive aggregation. If a threshold was crossed inside a window, the aggregate says so even when the summary statistic did not cross it.
- The underlying resolution is reachable in one action from the aggregate.
- Never aggregate across a method change. Averaging assay A and assay B produces a number that measures neither.
States
| State | Rendering |
|---|---|
| Single measurement | No trend is offered. One point is a value, and drawing a slope from it is the most common way software invents a finding. |
| Change within variability | Stated as such, with both values and the variability figure visible. |
| Variability unknown | Said plainly. The change is shown without a significance claim. |
| Not comparable | The delta is withheld or heavily qualified; the individual values are always shown. |
| Overdue | The trend states when the next measurement was due — a surveillance chart whose latest point is nine months late is mostly telling you that. See Pending work. |
| Re-analysed | Points produced by a different model version are marked, and the original values remain — see Model version change. |
Do's and don'ts
The interface declines to call noise a trend, and shows the numbers that justify declining.
A difference well inside the measurement's own error, labelled "growth" and given a direction arrow. This is how surveillance escalates for no reason.
The baseline is a named, dated measurement. Anyone can check which comparison is being made.
A percentage with no absolute values, no interval, and no indication of which measurement counts as baseline.
Overnight glucose · hourly minimum · 3 excursions below 3.9 mmol/L
Aggregated toward the hazard, with excursions surfaced. The summary answers the question that was actually being asked.
Overnight glucose · daily mean 7.8 mmol/L
A perfectly reasonable average concealing three hypoglycaemic excursions. The mean of a bad night looks like a good one.
One measurement on file. No trend available. 6.0 mm, 12 Jul 2023.
Refusing to draw a trend from a single point. The value is still shown.
Stable — no significant change detected.
A reassuring claim derived from one measurement. "Stable" asserts a comparison that has not been made.
Accessibility
- The change is text before it is geometry. Delta, interval and variability are readable without the chart, which is both an accessibility requirement and the only way to get exact figures.
- Chart labels state the trend in the
aria-label— direction, magnitude, interval and any discontinuity — see Data visualisation. - Significance is never an arrow alone. "↑" carries no magnitude, no interval and no indication of whether it exceeds noise.
- Out of range and out of noise are different statements, and neither is rendered in an alarm hue — see Colour.
- Discontinuities are announced as part of the series, not as a visual footnote positioned after it in reading order.
- Numerals are tabular and units are bound to values — see Typography.
- Reflow to 320 px: below the width at which a trend chart is legible, the table replaces it rather than the chart shrinking (SC 1.4.10).
Outcomes of use
What this contributes to, in the terms of Usability & context of use. These are attributes believed to contribute to an outcome; the outcome itself is settled by observing real use in a specified context, not by this page.
- Effectiveness — real change is acted on and measurement noise is not. Both directions are failures, and the second one is the invisible one: nobody counts the biopsies caused by a chart.
- Efficiency — no arithmetic, and no cross-referencing to establish what the baseline was. The delta, interval and rate are computed once, visibly, and can be checked.
- Satisfaction — confidence in a trend that survives being questioned. A clinician who discovers a scanner change hidden inside a growth figure will re-derive every trend by hand from then on.
Clinical safety notes
Trace these in your risk file (ISO 14971) and usability engineering file (IEC 62366-1).
- Measurement variability stated; sub-threshold change not called change. Mitigates: over-investigation driven by measurement noise rendered as progression.
- Variability stated as unknown where it is. Mitigates: implied precision the measurement does not have.
- Baseline named and dated. Mitigates: deltas computed against a baseline the reader did not expect, or one silently moved by a migration.
- Interval always shown with the change. Mitigates: a delta assessed against the wrong guideline window.
- Percentages carry absolute values. Mitigates: a large proportional change on a clinically trivial absolute one.
- Method and protocol changes stated adjacent to the change. Mitigates: instrument change read as disease progression.
- No trend from a single measurement. Mitigates: "stable" asserted from a comparison that does not exist.
- Aggregation window and operation stated; excursions preserved. Mitigates: an averaged series concealing the events that were the finding.
- Never aggregate across a method change. Mitigates: a summary statistic that measures neither method.
- Overdue surveillance stated on the trend. Mitigates: an old latest-value read as a current one.
Related
- Data visualisation — axes, gaps, projections and comparability, which this page applies rather than repeats.
- Pending work & closing the loop — what happens when the next measurement is due and nobody has taken it.
- Model version change — what happens to a trend when the thing producing the points changes.
- Key–value pair — the value, unit and reference rules a trend is built from.
- ECG review — serial comparison over minutes rather than years.
- CarryForward — the reference application.