Why Historical Health Data Can Lose Context During a Device Switch

A device switch is a measurement boundary

Moving from one wearable to another can make a chart look continuous. The dates line up, the units appear familiar and the app may place every value on one timeline. That visual continuity does not prove that the measurements are equivalent.

The new device may use a different sensor position, sampling schedule, filtering method, sleep window, baseline or missing-data rule. The old record may also lose its source labels during export. Treat the switch as a boundary until the two systems have been compared.

I preserve the original series before evaluating the new one. The overlap, when available, can show a consistent offset or a change in coverage, but it cannot by itself explain why either device reports a particular value.

Health-data source guidance explains why source information matters when records move between systems. A value without its source is harder to interpret even when the digits remain unchanged.

Preserve the old record before importing

Keep an untouched export from the original system. Preserve timestamps, units, source identifiers, coverage notes and any definitions that describe the metric. Do not rely on the new dashboard as the only copy of the history.

Record the last date under the old method and the first date under the new one. If the periods overlap, use the overlap for comparison rather than quietly replacing one series with the other.

Same label, different calculation

Terms such as resting heart rate, sleep duration, recovery or stress can hide different calculations. One device may use a fixed morning sample. Another may use an overnight average. One may estimate a sleep stage from several signals; another may use a different model.

Write the definition beside the imported field. If the definition cannot be recovered, label it unknown. An unknown definition can still support a visual history, but it should not be used as a precise baseline.

Sensor position changes the signal

Wrist, finger, chest and other positions do not capture identical signals under identical conditions. Fit, movement, skin contact and body position also matter. A new device can report a different value because it sees the signal differently, not because the body changed by the same amount.

The wearable-validation literature is a reminder that accuracy is specific to the device, metric and intended use. Avoid declaring one device correct and the other wrong from a short overlap.

Importing can lose missing-data meaning

The old system may have marked a night as partial, estimated or unavailable. The new system may display the imported value without that state. A complete-looking line can therefore contain records that were never complete in the original source.

Keep a missing-data field next to the value. Preserve whether the value was measured, estimated, manually corrected or copied from another system. The history remains more useful when its limitations travel with it.

Time zones can move a day

Overnight records are especially sensitive to local time. A sleep period that begins on one calendar day and ends on another can be assigned differently after export. A device switch may also change the time-zone rule used for daily summaries.

Keep the original timestamp and time zone where possible. Do not correct a visible date shift by hand without recording the correction. A shifted date can change daily totals, weekly averages and the apparent relationship between sleep and next-day values.

Provenance is the audit trail

Provenance means keeping the path from original observation to current display. Note the source, import event, conversion, calculation and manual correction. You do not need a complicated system to preserve the idea.

The provenance overview describes how entities, activities and agents relate through a record. For personal health data, that can be as simple as “recorded by source A, exported on this date, imported into source B, converted to this unit.”

The common exchange-standard overview shows why structured fields help systems exchange information. A personal spreadsheet or journal can apply the same discipline with plain labels.

Use an overlap period

If both devices are available, wear or use them under comparable conditions for a defined overlap. Compare direction, coverage and timing rather than expecting identical values. The overlap reveals whether the new series is systematically shifted, more variable or simply using a different window.

Do not recalibrate the old history to match the new one unless you have a documented reason. A transformation can be useful, but it should be visible and reversible.

An existing guide on choosing a primary data source when apps record the same metric provides a practical starting point. Decide which source owns the trend and keep the others as supporting records.

Keep measured and modeled history separate

Imported sleep stages, recovery scores and stress labels are often model outputs. A new system may recalculate them from raw data, import only the labels or ignore them entirely. The result may look like one history while containing several different models.

Separate raw or direct signals from derived labels. A model output can be compared with another model as an editorial or exploratory exercise, but it should not be described as the same measurement merely because the display uses the same word.

The baseline may need a reset

An app that learns a personal baseline may use a new start point after the device switch. If the old history is imported, the application may still treat the first days as fresh data, or it may blend them into an existing model. Check the documentation and record the boundary.

Do not assume that a long imported history gives the new system the same confidence it had in its original source. The imported record may omit the inputs needed for recalculation.

Watch for a false trend

A step change at the switch can look like a sudden change in sleep, HRV or resting heart rate. First ask whether the metric definition, sensor, window, time zone or coverage changed. Only after those checks should the biological interpretation begin.

The review of wearable-data limitations supports comparing accuracy and meaning by metric, not by brand reputation or a single attractive chart.

Information about a device source and the path of a record is useful at exactly this boundary. If the new system does not preserve that information, add a local source note so later comparisons do not mistake an imported value for a current direct observation.

Keep the scope of a conclusion small

You may be able to say that the new device reports higher overnight values under a comparable period. You may not be able to say that health changed by the same amount. The narrow statement preserves the useful observation while respecting the measurement boundary.

Keep the old and new series identifiable in any summary shared later. A short note naming the switch date, source and unresolved differences helps a later reader distinguish a continuous display from a continuous method.

Review the boundary before using the new baseline

Look at the last several records under the old method and the first several under the new method. Compare the collection time, coverage, source, units and definitions before comparing the values themselves. A discontinuity in the chart may be a method change rather than a change in physiology.

If the two devices overlap, keep the overlap as evidence about comparability, not as a demand that the numbers match. Differences may be systematic and still predictable. A short overlap cannot establish clinical accuracy, but it can show whether the new series has a different scale or missing-data pattern.

Review the baseline window used by any score. The destination system may learn from only new records, may accept imported history without recalculating it, or may blend old and new values under rules that are not visible. Preserve the application state when possible and record the date on which the score changed behavior.

When the source cannot be recovered, downgrade the conclusion. “The new record reports a different level after the switch” is supported. “The body changed at the switch” is not, unless other evidence supports it.

The same caution applies to a gradual drift that begins near the boundary. Check the method before calling it a long-term biological trend.

Mark the switch before making that claim.

Then compare the two methods on their own terms.

A clean switch keeps history honest

Before merging data, preserve the original export, mark the boundary, compare overlapping periods and keep source, units, windows, calculations and missing-data states visible. If those details are missing, treat the old and new series as related but not identical.

The purpose of a health timeline is not to look smooth. It is to remain interpretable when the method changes. A visible break is often more honest than a continuous line built from incompatible records.

FAQ

Can I merge data from two devices into one trend?

You can place them on one timeline, but merge them numerically only after checking definitions, units, windows, source, calculation and coverage. Mark the device boundary.

Why can the same metric change after a device switch?

Sensor position, fit, sampling, filtering, calculation, time zone and missing-data rules can all change. The displayed difference is not automatically a biological change.

What should I preserve before switching?

Keep the original export, timestamps, units, source, definitions, coverage notes, missing-data states and the date on which the new method begins.