Diagnostic Cutoffs in Connected Devices: Why a Threshold Changes the Result

By the Connected Diagnostics Evidence Team

The Number Behind the Result

When a connected diagnostic device — a phone-linked otoscope, a home ECG patch, a digital rapid-test reader — tells you “positive” or “negative,” that word is standing in for a number crossing a line. The line is called a cutoff, or threshold. Move the line, and the same raw measurement can flip from negative to positive. Before you weigh a study’s claims about a connected device, it helps to know where that line came from and when it was drawn.

This guide shows you how to find that information in a study, tell a locked-in threshold from one chosen after the fact, and build a simple evidence map for the device you’re reading about. It does not tell you what any individual result means for you — that’s a conversation for the clinician who ordered or is reviewing your test.

Step 1: Find Out When the Threshold Was Set

A cutoff can be set two ways, and the difference matters:

• Prespecified: Researchers pick the threshold before they look at the results, usually based on prior studies or a standard method. This is the stronger design.
• Post hoc (exploratory): Researchers try out several thresholds after collecting the data and report the one that performed best. This tends to make the device look more accurate than it will in real-world use, because that “best” cutoff was chosen to fit this one dataset.

Look for the study’s methods section and search for words like “prespecified,” “predefined,” or “a priori” next to the cutoff value. If the threshold only shows up in the results section, with no mention of how or when it was chosen, treat it as unconfirmed until you can find out more.

Step 2: Identify What the Device Was Compared Against

A cutoff’s meaning depends entirely on what “correct” meant in that study. The FDA’s statistical guidance for diagnostic test studies draws a hard line between two comparison types, and it changes which words a study is allowed to use:

• Reference standard: The best available method for confirming whether the condition is truly present or absent. When a device is compared to a genuine reference standard, a study can properly report sensitivity (how often it catches a true case) and specificity (how often it correctly clears a true negative).
• Non-reference standard: A comparison method that isn’t considered the gold standard — another device, a less-established test, or clinician judgment. When this is the comparator, the guidance says the proper terms are positive percent agreement and negative percent agreement — agreement with the comparison method, not proof of being correct.

This distinction is not a technicality. A study can report 95% “agreement” and still be describing two methods that agree with each other while both being wrong on the same cases. If a study uses the words “sensitivity” or “specificity” but the comparison method described isn’t clearly a reference standard, that’s worth flagging as a red flag in your evidence map.

Step 3: Read the Cutoff Alongside the Prevalence

The same threshold can look very different depending on how common the condition was among the people tested. A device tested in a group where half the participants had the condition will report different-looking numbers than the same device tested in a general population where the condition is rare — even if nothing about the device changed. When you’re comparing claims across two studies or two connected devices, check whether they tested similar populations before assuming the numbers are comparable.

Your Threshold Evidence Map

Use this worksheet with any study or product page describing a connected diagnostic device’s accuracy. Answer each line in your own notes:

• The cutoff value: What specific number or measurement counts as “positive”?
• Timing: Was it stated as prespecified, or does the source only show it after the fact?
• Comparator: What was the device measured against — and is that comparator described as a reference standard or something else?
• Terminology check: Does the source use “sensitivity/specificity” or “percent agreement” — and does that match what its comparator actually was?
• Population: Who was tested, and how common was the condition in that group?
• Confidence interval: Is a range given alongside the headline number, or just a single percentage?
• Open questions: What’s missing that you’d want before trusting this claim — sample size, independent replication, real-world testing conditions?

A study that answers all seven lines clearly is giving you enough to evaluate its claim. A study that leaves several blank isn’t necessarily wrong, but it hasn’t given you enough to weigh it with confidence.

Why This Matters for Reporting Quality

The STARD reporting guidelines for diagnostic accuracy studies exist specifically because incomplete reporting on cutoffs, comparators, and populations was common enough in the research literature to need a fix. STARD was built to help researchers report the target condition, the reference standard used, and the study population clearly enough that readers — including you — can judge a study’s reliability without needing to be a statistician. When you see a study or device page that’s transparent about all of the items in the worksheet above, that transparency is itself a marker of a more trustworthy source.

What This Guide Doesn’t Tell You

Nothing here indicates whether any specific connected diagnostic device is accurate enough for a particular use, and nothing here should be used to interpret an individual test result. If you or someone you’re caring for has a result from a connected diagnostic device that raises concern, or if you’re facing a possible medical emergency, contact your clinician or local emergency services rather than trying to resolve it through study thresholds.

Related Reading on This Site

For the broader picture of how we evaluate connected and home diagnostic tools, start with our Start Here guide. For how we select and read the research behind these articles, see How We Research. [Editorial note — not for publication: these two paths are unverified this session; confirm they resolve under the site’s actual current branding before release.]

By Connected Diagnostics Evidence Editorial Team

Medical information disclaimer: This article is for general education only and is not a substitute for professional medical advice, diagnosis, or treatment. It does not evaluate or endorse any specific product. Always consult a qualified healthcare provider with questions about a diagnostic result or medical condition. Connected Diagnostics Evidence is an independent educational publication and is not affiliated with, and does not continue the operations, products, research, or clinical relationships of, any company that previously operated this domain.

Last updated: September 15, 2026.