When a connected diagnostic device study reports how well it works, some images or readings almost always come back unclear, blurry, or impossible to interpret. These are called indeterminate results. A trustworthy study does not quietly drop them from the count. It reports how many indeterminate results occurred and exactly how they were handled in the math — because leaving them out, or silently treating them as “negative,” can make a device look more accurate than it is.
This guide shows you how to find that information in a study and how to tell whether the denominator — the total number a percentage is calculated from — is honest.
Why this matters for a connected device study
Connected diagnostic tools, including digital otoscopes, connected examination cameras, and mobile microscopy devices, produce an image or reading that a person or an algorithm then has to interpret. In diagnostic accuracy research, the test being evaluated is called the index test. Its result is checked against a reference standard — a more reliable test, exam, or diagnosis used as the benchmark for what counts as a “true” positive or negative.
Some captures will always be too dark, too blurry, out of focus, or technically unreadable. That is a normal, expected part of using any imaging device — not a flaw unique to one product. The question worth asking is not “did this happen,” but “what did the study do with it once it happened.” That single design choice can shift a device’s reported accuracy numbers significantly.
Evidence ladder: what is established, what is guidance, and what is still open
Established requirement (reporting standard)
The STARD 2015 guideline (Standards for Reporting Diagnostic Accuracy Studies), the widely used checklist for publishing diagnostic accuracy research, requires studies to report:
- A participant flow diagram (STARD item 19) showing what happened to everyone enrolled, from initial testing through final results
- How indeterminate index test or reference standard results were handled (STARD item 15) — in plain language, what the researchers did with the unreadable or unclear results
- How missing data were handled (STARD item 16) — results that were never obtained at all, as distinct from results that were obtained but unreadable
STARD is a reporting checklist, not a law. A study can be published without following it. But a study that follows STARD gives you enough information to check its math yourself, which is exactly what the worksheet below helps you do.
FDA guidance (recommended statistical approach)
The FDA’s statistical guidance for reporting diagnostic test studies addresses this problem directly. Its core points:
- A test that can return something other than a clean positive or negative result is, in the FDA’s words, “not technically a qualitative test” in the way the standard formulas assume
- The guidance states plainly: “Discarding or ignoring these results and performing the calculations in this guidance will likely result in biased performance estimates”
- Instead, the FDA recommends reporting two separate sets of performance numbers: one where indeterminate results are grouped with the positives, and one where they are grouped with the negatives
- Where a study’s situation is more complicated, the FDA recommends the study team consult directly with FDA statisticians rather than pick an ad hoc method
In other words, there is no single official formula that converts an indeterminate result into a percentage. There is a documented expectation that the number gets shown to you, not absorbed into the background.
Still open (not established by either source)
Neither source sets a maximum acceptable indeterminate rate, and neither source dictates one universal method a study must use to fold indeterminate results into a single headline accuracy number. A study that reports a high indeterminate rate is not automatically failing a rule — but it does owe you the number and its handling method, per STARD, so you can judge it yourself.
Known vs. unknown: what a study reveals about itself
Known, if the study reports it well:
- The total number of people or samples enrolled
- The number of index test attempts that produced a readable, interpretable result
- The number of index test attempts that were indeterminate, unreadable, or technically failed
- Whether the indeterminate results were excluded, counted as positive, counted as negative, or reported both ways
Unknown or unclear, in many published or marketing summaries:
- What device setting, lighting condition, or operator step caused a given indeterminate result
- Whether the accuracy percentage quoted in a headline used the interpretable-only denominator or the full enrolled denominator
- Whether a marketing summary of a study matches the full study’s own reported numbers
When a summary is silent on all of this, that silence is itself information. It does not tell you the device performed poorly — it tells you that you cannot check the claim from what you were given.
The reporting worksheet
Use this worksheet on any connected-device diagnostic study you are trying to evaluate. Fill in each line from the study’s own text, not from a press release or product page summarizing it.
- Total enrolled or tested (A): How many people, samples, or images went into the study at the start?
- Interpretable results (B): How many attempts with the test being evaluated (the index test) produced a result the study could actually classify as positive or negative?
- Indeterminate or unreadable results (C): How many did not? Check that A = B + C (plus any separately reported missing data, per STARD item 16). If the numbers do not add up, the study has an unexplained gap.
- Handling method: Does the study state whether the indeterminate group (C) was excluded from the headline accuracy number, folded in as positive, folded in as negative, or reported under both scenarios (the FDA-recommended approach)?
- Which denominator is in the headline number: Is the accuracy percentage you were quoted calculated using B (interpretable results only) or A (everyone tested)? These can produce very different numbers from the same study.
- Flow diagram check: Does the study include a participant flow diagram (STARD item 19) you can trace from enrollment to final result? If yes, your worksheet numbers should match it exactly.
If a study or a summary of one cannot answer question 4 or 5, treat the headline accuracy number as incomplete rather than final.
Practical next steps
- Before trusting a device’s advertised accuracy rate, look for the word “indeterminate,” “uninterpretable,” “equivocal,” or “unreadable” in the source study, not just the summary
- If you can only find a marketing summary, look for a link to the full published study or the device’s regulatory submission, and check the worksheet numbers there instead
- If the indeterminate rate is not stated anywhere you can find, treat that as an open question rather than assuming it was zero
- Bring a specific question to your device’s documentation or your healthcare provider if you personally received an unreadable or unclear result from a connected device — ask how that outcome is meant to be handled by the device’s own instructions, not just what the study reported
What this article is not
This is general educational information about how to read diagnostic accuracy reporting. It is not medical advice, and it does not evaluate, endorse, or rank any specific connected diagnostic device or brand. If you have a health concern related to a specific test result, image, or symptom, contact a licensed healthcare provider. If you are experiencing a medical emergency, contact your local emergency services immediately.
Sources
- U.S. Food and Drug Administration, Statistical Guidance on Reporting Results from Studies Evaluating Diagnostic Tests — Guidance for Industry and FDA Staff
- EQUATOR Network, STARD 2015: Standards for Reporting Diagnostic Accuracy Studies
Related reading on this site
- Start Here — an orientation to how Connected Diagnostics Evidence approaches home and connected health imaging
- How We Research — our sourcing hierarchy and how we separate established evidence from open questions
Connected Diagnostics Evidence is an independent editorial publication and is not affiliated with, and does not continue, the former CellScope company that previously operated this domain. We do not sell, test, or endorse diagnostic devices. Updated September 14, 2026.
By Connected Diagnostics Evidence Editorial Team
Leave a Reply