Why a Symptom Checklist Can’t Diagnose Mold Illness

If you have researched mold illness, you have met the checklist: a long list of symptoms, sorted into groups, with a rule — have enough of them and you probably have CIRS. It is the front door to the diagnosis, and it is usually presented as if it were built from patient data and confirmed by statistics.

This article tests that claim against the instrument’s own source papers. The conclusion is narrow but important: the checklist was never derived or validated the way a diagnostic test has to be — and knowing why should change how you read your own results, whether you are a patient or a clinician.

The front door: a count, not a diagnosis

The screen is a roster of roughly thirty-seven symptoms sorted into thirteen “clusters” — fatigue, cognitive trouble, aches, sinus and breathing complaints, and so on. Eight or more clusters is treated as the threshold that flags a probable case and sends you on to laboratory and visual-contrast testing. Everything downstream depends on passing this gate.

“Cluster” is a mathematical claim — and it was never made good on

The word cluster is not decorative. It asserts that these symptoms were shown to travel together — that the thirteen groups fall out of the data. They do not. The groupings were assigned by clinical judgment, and no factor analysis, latent-class analysis, or cluster analysis of the symptom data has ever been published to support them.

A claimed structure that has never been subjected to a structure-finding analysis is not a finding; it is a layout. — Andrew Heyman, MD

You can watch the layout change. Across the two foundational papers the roster grew from twenty-six symptoms in eight organ systems to thirty-seven in ten, and the passing threshold moved with it — from four-of-eight to five-of-ten. The thirteen-cluster, eight-threshold version in use today appears in neither paper. A structure that shifts between its own founding studies, with no method to account for the shift, is curated, not derived.

The decisive gap: it was never tested against the look-alikes

This is the failure that matters most at the bedside. The symptoms on the roster — fatigue, brain fog, pain, headache, mood change — are the shared vocabulary of chronic fatigue syndrome, fibromyalgia, depression, anxiety, sleep apnea, and an underactive thyroid. To mean “this disease” rather than simply “some illness,” a screen has to be shown to separate its target from those alternatives. That test — discriminant validity — has never been performed. Without it, a high cluster count is consistent with any of a dozen other conditions, or several at once.

A test measured against itself

There is one study widely cited as validation. It does not validate the instrument; it applies it. It measures the checklist against the construct’s own case definition — the same symptoms on both sides of the comparison. In measurement science that is a named error, incorporation bias, and the high agreement it produces is an artifact of the circularity, not evidence of accuracy.

The threshold has the same problem. “Eight of thirteen” has no published derivation — no analysis placing it on a sensitivity–specificity curve, no calibration. And the statistics offered are significance values from public web calculators run on nineteen and twenty-six patients in a single specialty clinic. A real change in symptom count under treatment is a genuine finding; it is not, and cannot be turned into, the diagnostic accuracy of a thirteen-cluster classifier.

Why this matters — and what it does not mean

These are not bookkeeping objections. A count-based screen that has never been tested against look-alikes and is validated against itself has a predictable, harmful behavior: applied in a population already full of unwell people, it labels a large fraction of them as cases. The cost is twofold. Patients who in fact have a different, often treatable condition can be routed into a single explanation and away from the correct diagnosis. And the credibility of genuine environmental-illness research is weighed down by a prevalence claim the instrument cannot support.

None of this implies that patients are not ill, or that water-damaged buildings cause no disease; it implies only that this instrument cannot establish who has this disease, because it was never built to. — Andrew Heyman, MD

What would actually make it a diagnosis

The remedy is not rhetorical; it is methodological, and the field already holds most of the pieces. Let the structure be discovered rather than assumed — analysis that finds how many symptom dimensions really exist, a number that may or may not be thirteen. Establish reliability. Test discriminant validity against prospectively enrolled comparison groups: chronic fatigue syndrome, fibromyalgia, depression, sleep disorder, thyroid disease. Derive the threshold from a proper sensitivity–specificity analysis. And, decisively, anchor the symptoms to objective biology rather than to themselves.

That last step is where the Cell Danger Response program points: a properly derived symptom structure read alongside measurable markers — mitochondrial signals such as GDF15 and FGF21, and validated respiratory and allergic measures where the outcome is respiratory. The way out of the circle is to derive the structure, validate it against something other than itself, and let independent hands try to break it.

What it means for you

Do not let a checklist score be the last word — in either direction. If your symptoms fit, the right next step is objective testing that can separate mold illness from the conditions it imitates; and a low score does not clear you when your exposure history and your biology say otherwise. Symptoms are the reason to look. The proof is in the testing.

Until that is done, the cluster count should be read for what it is — a clinically curated tally of common symptoms — not for what it is presented to be: a validated diagnostic classifier. — Andrew Heyman, MD