Evidence
The mathematics is not ours, and it is not new.
The idea that a complex system slows in its recovery from small disturbances as it approaches a tipping point — and that this slowing is measurable before the tip — is an established body of science, developed over decades across ecology, climate and physiology. We did not discover it. We apply it, under a per-subject baseline. Here is the lineage, and an honest account of what it does and does not establish.
What this page is, and is not
This is the published, third-party science our method draws on. It is not evidence about any Intuitive Pathos product. We have not published clinical performance data, and the presence of these references is not a claim that our application of the mathematics has been validated in our settings. That validation is exactly the work a regulatory pathway exists to establish, and it is ahead of us, not behind us.
We would rather show you the shoulders we stand on — and where the ground runs out — than imply results we do not have.
The generic signal
Critical slowing down.
Many systems — a lake tipping into algal bloom, a climate approaching a threshold, a body approaching a physiological transition — share a mathematical signature as they near that boundary. They recover from small perturbations more and more slowly. In a time series, that slowing shows up as rising autocorrelation, rising variance, and a lengthening return time.
The remarkable, and repeatedly demonstrated, fact is that this signature is generic: it appears without a model of the specific mechanism driving the transition. That is what makes it a candidate early-warning signal in a body, where the mechanism is often exactly what you do not yet know.
Our contribution is not this signal. It is insisting that the signal be measured against the subject’s own dynamics rather than a population constant, and that departures be treated symmetrically. The framework explains why; this page is about where the underlying science comes from.
The measurable signatures
- Rising short-lag autocorrelation (the memory coefficient).
- Rising variance in the fluctuations.
- Lengthening recovery time after a perturbation.
- Proximity to a criticality boundary in the system’s own dynamics.
These map onto the primitives described on the framework page — memory coefficient, critical slowing down, λ* criticality, sliding-window regime detection.
Selected literature
The published lineage.
A representative, not exhaustive, trail — from the foundational statements of the theory to its application in human physiology and critical care. Follow them; the science stands on its own, independent of us.
-
Early-warning signals for critical transitions.
The foundational synthesis: how generic early-warning signatures precede critical transitions across very different systems.
-
Anticipating critical transitions.
Extends the framework and confronts its limits — when the signals appear, and when they can mislead.
-
Methods for detecting early warnings of critical transitions in time series illustrated using simulated ecological data.
The methods paper — how the signatures are actually computed from a time series, and how to test them.
-
Critical slowing down as early warning for the onset and termination of depression.
The signal in human data: momentary-state time series slowing before a transition in mental health.
-
Slowing down of recovery as generic risk marker for acute severity transitions in chronic diseases.
The argument carried explicitly into critical care — the setting the clinical flagship addresses.
Go to the source
Each reference links to its DOI. These are landmark works, listed to point you at the field, not to borrow their authority — read the originals, particularly the parts on where early-warning signals fail, which we take seriously on the framework page.
The disparity case
Population-reference and instrument bias are documented, not assumed.
The framework’s claims that a population reference range absorbs its cohort’s composition, that a sensor can carry its own measurement bias, and that quiet deterioration is under-attended relative to loud deterioration — are not analogies we made up. Each has a published, peer-reviewed literature behind it. None of it is a study of an Intuitive Pathos product; it is the documented problem our approach is a response to.
-
Racial bias in pulse oximetry measurement.
Pulse oximeters overestimate arterial oxygen saturation in Black patients nearly three times as often as in white patients — a sensor, not a comparison, carrying bias into a reading that looks the same for everyone.
-
Racial and ethnic discrepancy in pulse oximetry and delayed identification of treatment eligibility among patients with COVID‑19.
Traces the same sensor bias through to a consequence: occult hypoxemia that delayed or withheld treatment eligibility, disproportionately for Black and Hispanic patients.
-
Hidden in plain sight — reconsidering the use of race correction in clinical algorithms.
The reference-range argument in clinical form: algorithms and normal ranges built to adjust for race routinely direct less attention and fewer resources to the patients who need them, across kidney function, lung function and more.
-
Delirium and its motoric subtypes: a study of 614 critically ill patients.
The asymmetry case: hypoactive (withdrawn, quiet) delirium was far more common than hyperactive delirium in this ICU cohort, and is the subtype clinical staff are least likely to recognize — the loud direction gets noticed; the quiet one does not.
What this does, and does not, establish
These papers establish that population-reference and sensor bias are real, measured, and consequential in today’s instruments, and that asymmetric attention to quiet deterioration is a documented clinical pattern, not a rhetorical device. They do not establish that a per-subject baseline or symmetric scoring fixes any of this in a deployed product — that is a claim we have not yet made with data, and the framework page says so directly under “sensors carry their own bias.”
A sensor that measures some people less accurately still does, no matter whose baseline it is compared against. Per-subject comparison addresses the reference-range problem; it does not, on its own, repair the instrument underneath.
The no-self-report case
In animals, pain is common and recognition is rare.
The veterinary product’s premise — that chronic pain in animals is widespread and routinely goes unrecognized, because the patient cannot report it and the signs are quiet — is documented. It is the same asymmetry as the clinical case, in a setting where self-report is not merely unreliable but absent.
-
Retrospective radiographic study of degenerative joint disease in cats: prevalence based on orthogonal radiographs.
The gap, starkly: radiographic appendicular osteoarthritis in 74% of 101 cats, while lameness was noticed by owners in two cats and identified by a veterinarian in two, and palpation found no obvious pain in any cat with appendicular osteoarthritis. Note the cohort was presented for arthritis screening, so prevalence is not a general population estimate — the recognition gap is the point, not the rate.
-
Cross-sectional study of the prevalence and clinical features of osteoarthritis in 100 cats.
Radiographic osteoarthritis in 61% of cats aged six and over, in at least one joint. What owners reported was not pain but decreased mobility and grooming — a quiet change in baseline behaviour rather than a complaint. The authors note that association also tracked age, so it is not cleanly attributable to osteoarthritis alone.
What this does, and does not, establish
These establish that the condition is prevalent and that routine recognition of it is poor. They do not establish that continuous per-subject monitoring detects it, or that detecting it earlier improves welfare. Both remain ours to demonstrate.
The honest ledger
What the science establishes — and what it doesn’t.
What the literature supports
- That critical slowing down is a real, measurable, and generic precursor of state transitions in complex systems.
- That the signatures have been observed in human physiological and psychological data, not only in models and ecosystems.
- That the mathematics is sound, published, and open to inspection.
What it does not, on its own, establish
- That our specific implementation detects deterioration in our specific settings — that is our burden to demonstrate.
- Any clinical performance figure — sensitivity, specificity, lead time, or effect on outcomes — for any product.
- That the per-subject-baseline framing improves on existing tools. We believe it does; belief is not evidence.
If you are a researcher who can help close that gap — retrospective analysis, prospective study design, an adversarial look at where this would fail in your unit — that is the most valuable thing you can bring us. It is worth more than agreement.