Why a Health Data Category Name Does Not Make Two Metrics Equivalent

Labels are not definitions

Health records often place different measurements under a shared category such as heart rate, sleep, activity or recovery. The shared label helps people find related information. It does not prove that the values use the same unit, source, time window, sampling method or calculation.

Two metrics can therefore look compatible while answering different questions. One may be a direct sensor measurement, another may be an estimate, and a third may be a composite score derived from several inputs. Merging them because the category name matches can create a trend that never existed in the original records.

I check the definition before I compare the values. If the source cannot explain the window or calculation, I keep the records separate and state what remains unknown. That leaves the comparison open to revision when better documentation becomes available.

The introduction to a common health-data exchange standard explains why structured definitions are needed when information moves between systems. A category is only the beginning of that definition.

Start with the metric definition

Before combining values, write down what each one measures. For heart rate, ask whether it is a spot reading, a resting average, an overnight minimum, a workout sample or a model output. For sleep, ask whether the value is time in bed, estimated sleep, efficiency or a score.

Use the exact displayed name, unit and time window. If a definition is not available, mark it unknown. Unknown does not mean wrong, but it limits how strongly the values can be compared.

Source changes the meaning

Source means more than the name of the app that displays the number. It includes the sensor, the original recording system, an imported file, a manual entry and any transformation performed during transfer. A value can be copied accurately and still lose the source context that made it interpretable.

The health-data source guidance shows why provenance matters when information moves between systems. A source label helps answer where the value came from and what happened to it on the way.

Units can hide a mismatch

The same category can contain different units or scales. A heart rate may use beats per minute, while a recovery metric may use a percentage or score. A sleep value may use minutes, a ratio or a qualitative label. Converting the display to a common-looking number does not make the underlying measures equivalent.

Keep the original unit. If conversion is necessary, record the formula and do not erase the source value. A converted number should remain visibly derived.

Time windows matter as much as labels

A daily value can summarize a 24-hour period, a calendar day, a detected sleep window or only the minutes when the device was worn. Two “daily” metrics may include different hours. A weekly average can also use different inclusion rules.

Write the start and end rule beside the metric. Align windows before comparing them. If the windows cannot be aligned, compare them as related observations rather than as interchangeable values.

Calculations can create different outputs

One system may average raw samples. Another may filter artifacts, select specific epochs or use a baseline deviation. A composite health score may weight several inputs and change its result even when one component is stable. A category name does not expose those decisions.

The review of consumer wearable validation shows why a metric needs validation for its intended use. A label that looks familiar is not a substitute for knowing the calculation.

Preserve measured and estimated values

Measured values come from a captured signal. Estimated values come from a model or an inferred state. Both can be useful, but they should not be merged without an explicit rule. If a device estimates sleep stages and another records a heart-rate signal, the two values do not become equivalent because both appear under sleep.

The consumer wellness-device guidance keeps the boundary clear. A device can support general wellness tracking without establishing a diagnostic measurement.

Use provenance like a chain of custody

For each value, preserve source, timestamp, unit, method, transformation and confidence. If data are exported, imported or manually corrected, record the event. This is not unnecessary bureaucracy. It prevents a later reader from mistaking a transformed value for a direct observation.

The provenance overview describes how records can preserve the relationships between an entity, an activity and an agent. A personal health log does not need a full technical implementation, but the same idea applies: keep the path visible.

An existing guide on choosing a primary data source when apps record the same metric offers a practical rule: decide which source owns the trend before merging duplicates.

A simple equivalence test

Ask five questions before treating two metrics as the same: do they measure the same phenomenon, use the same unit, cover the same window, use comparable calculations and retain the same source context? If one answer is unknown, keep the values separate until it is resolved.

The test is deliberately strict. A related metric can still be useful beside another metric. The goal is to prevent a false merge, not to reject every comparison.

Category names can be useful without being exact

A shared category helps organize a dashboard. It can put overnight heart rate, resting heart rate and workout heart rate in one place so the reader can inspect the broader pattern. The category should be a navigation aid, not a claim that the values are identical.

Use sublabels for source, window and method. A small amount of visible detail often prevents a much larger interpretive error later.

Imported history needs a boundary

When old data enter a new system, mark the import date and the original source period. Do not assume that the new dashboard recalculated the original metric. A chart that looks continuous may contain a change in device, definition, time zone or missing-data handling.

The recent review of wearable-data limitations supports a metric-specific view of accuracy and interpretation. Data continuity is not the same as measurement equivalence.

Keep the narrowest valid comparison

If two metrics share only a category, compare their timing and direction qualitatively. If they share definition and window, a numerical comparison may be reasonable. If they share neither, preserve them as separate records.

Before joining records, write down the intended question. A display meant to browse related measures may reasonably group them, while a calculation that combines values needs stronger equivalence. Keeping those purposes separate prevents a navigation label from becoming an undocumented conversion rule.

This rule helps when an app offers one combined trend but hides the inputs. Reconstruct the inputs where possible. If they cannot be recovered, label the composite as a separate model output.

Test the relationship before merging

Start with the smallest defensible comparison. If two values share a definition, unit and window, compare their direction and magnitude. If they share only a category, compare their timing and general context without calculating a combined score. A careful qualitative comparison is better than a precise-looking invalid one.

Keep a data dictionary beside the record. It can be a short table with field name, definition, unit, source, window, method and status. When an app or export changes, update the dictionary instead of silently changing the meaning of an old field.

Composite scores deserve their own name

A composite score is a separate metric even when its inputs come from familiar categories. It may combine sleep, heart rate, activity, stress and baseline deviations under a particular weighting. The score should not be substituted for any one input, and one input should not be used to explain the entire score without seeing the other contributors.

If the weighting is hidden, describe the score as a source-specific model output. Preserve the inputs when available. This keeps the chart useful and makes it possible to revise the interpretation when the model or data source changes.

Do not average unlike uncertainty

Measured, estimated and manually entered values carry different uncertainty. Averaging them into one line can make the result look more stable than the evidence allows. Keep the status visible and decide in advance which records are eligible for a calculation.

The same rule applies to missing data. A blank, an estimated value and a zero are not interchangeable. A category name cannot resolve those differences.

Keep those states visible in the table and in the prose that explains a trend.

That visibility protects later comparisons.

A label organizes data; it does not validate it

The meaning of a health metric comes from its definition, source, unit, window, calculation and provenance. A common category name can help a reader explore related observations, but it cannot make unlike values interchangeable.

Keep the original record, visible transformations and uncertainty. The result is a trend that can be audited, revised and explained instead of a smooth chart built from hidden mismatches.

FAQ

Can two metrics with the same category name be compared?

Sometimes, but only after checking definition, unit, time window, source, calculation and provenance. A shared category alone is not enough.

Should estimated and measured values be stored separately?

Yes. They can appear together for context, but the record should show which values came from a direct signal and which came from an estimate or model.

What should I preserve when moving health data?

Keep the original value, unit, timestamp, source, method, transformation and any missing-data or confidence note. Mark imports and manual corrections.