labor · health · family: the missing yardstick
the armband says you're fine
Heat-stress wearables tell workers their core temperature using often-secret algorithms validated mostly on young, healthy volunteers, with device agreement reported as close below 38 °C
Problem statement
Employers in construction, agriculture, mining, energy and manufacturing are turning to wearable armbands and chest straps that display a worker's core body temperature and issue heat-stress warnings. These devices do not measure core temperature directly; they estimate it, commonly from heart rate and skin temperature, using algorithms that are often proprietary. The systematic review of this field found 25 distinct prediction algorithms validated across a total of 592 subjects, with an average validation sample of 32 subjects whose average age was 26 — and the industrial hygienists who have to decide whether to trust these devices note that in the two device comparisons they cite, the agreement with a rectal or ingestible-pill thermometer is reported as close below 38 °C; the systematic review itself describes a core temperature of around 38.5 °C as a normal response to exercise and generally not cause for concern, with clinical diagnosis and treatment of heat illness beginning at 40 °C. The unsolved problem is that there is no standard for how accurate an estimated core temperature has to be before a safety decision can rest on it, and a buyer may be unable to evaluate the estimate, because the algorithm is often proprietary and not disclosed.
Why this matters
Heat illness kills quickly and the decision the device informs — keep working or stop — is irreversible in one direction. The AIHA authors describe the specific failure mode with unusual bluntness: too much error and "the device loses all value for warning or monitoring, instead functioning more as an in-flight recorder — that is, valuable only as an element to be investigated following a critical illness." They add a second: because most devices cannot summon an emergency response, "most will continue monitoring even during a medical emergency warranting intervention, such as heat stroke." The authors also warn that clear, unambiguous outputs can make professionals and workers overconfident in the numbers, and that it is tempting to conflate using such a device with some measure of safety. The problem also stacks on a base layer the authors themselves call uncomfortable: the ACGIH heat stress Threshold Limit Value is built on a 70-kilogram reference body while U.S. adult averages are 171 pounds for women and 200 pounds for men, and the "normal" core temperature the limit protects is itself modulated by age, sex, chronic disease, medication, fitness and acclimatization — and the review found the validation samples varied little on several of these: young subjects, only a third of the algorithms validated with any female subjects, and most studies excluding people with medical conditions or on medication.
What’s been tried and why it hasn’t worked
Direct measurement is often ruled out for good reasons: rectal thermometry and ingestible pills are accurate but invasive and ethically difficult in a workplace, and in-ear thermometry devices are, per the AIHA authors, "no longer available due to practical challenges with their usage and concerns about their validity." Indirect estimation replaced them, and comparison studies exist — a manufacturer-coordinated firefighter trial against rectal thermometry and a Bland–Altman comparison against an ingestible pill in nurses working in heat — but the AIHA authors report a close match in both for core temperatures below 38 °C, say nothing about agreement above it, and call such studies "limited in many ways." The systematic review does report that 17 of the 18 algorithms with a published RMSE met a clinical-validity benchmark of RMSE at or below 0.5 °C — so the gap is not the absence of any accuracy number, but the absence of an occupational standard tied to the populations, temperatures and field conditions where the safety decision is made. The review identifies the structural weakness rather than a bug: few of the 25 algorithms incorporate individual and environmental data despite the known influence of those factors, validation subjects are neither diverse nor numerous, and the reviewers conclude that validity in dynamic real-world environments remains unestablished. Evaluation by the buyer is blocked from the other side: proprietary models mean an occupational health professional "may be forced to assume that the intrinsic error is limited enough to rely on for health and safety," and firmware updates, radio interference, direct sunlight, strap tightness, caffeine, smoking, medication and sensor placement are all named as possible influences on the measurements, with how much each one matters left as an open question. The AIHA authors note there is simply "no standard to determine when an estimated value is considered adequately accurate," so even a manufacturer who wanted to demonstrate fitness for purpose has no target to hit. Per-worker calibration might help but adds time, complexity and cost.
What would unlock progress
What would unlock this is an evaluation protocol that does not require access to the algorithm: a standardized, published challenge protocol — defined work rates, defined heat and humidity conditions, a diverse subject panel, an agreed reference measure, and a required accuracy report in the range that matters (above 38 °C, during exertion, on a recovering body) — so devices can be compared as black boxes on the decisions they produce rather than on the temperatures they claim. A possible model, which neither source raises, is the way other sealed safety and medical products are evaluated — respirator fit, gas-detector challenge tests, pulse-oximeter accuracy — against a reference under stated conditions. A second, complementary unlock is decision-level rather than measurement-level: define the alarm-to-action mapping first (AIHA's own alarm guidance recommends associating specific alarms with specific actions) and then ask what measurement accuracy that action actually requires, which may be far looser for "take a shaded break" than for "call EMS."
Entry points for student teams
A team could write and publish the challenge protocol itself — the specification the field is missing: defined work rates and climate conditions, the reference measure, a subject panel composed to span body mass, sex, age and acclimatization, and the accuracy statistics a manufacturer must report in the range that matters (above 38 °C, during exertion, on a recovering body), down to language a purchaser could paste into a bid document. That is a semester of standards drafting and evidence synthesis needing no chamber and no volunteers, and it is the deliverable the AIHA authors are actually short of; running such a trial is a different project, requiring a climate chamber, an invasive reference thermometer and medical monitoring above 38 °C — hospital-grade thermal-physiology infrastructure that no student protocol clears. A modeling team could take the published algorithm families (heart-rate Kalman filters, skin-temperature models) and quantify how much the estimate moves under realistic perturbations — strap tightness, sun load, a caffeine dose — scoring them against simulated core-temperature trajectories from the free ISO 7933:2018 predicted-heat-strain implementation published by Ioannou et al. (Industrial Health, 2019), and producing the error budget the field currently lacks. A design team could tackle the acceptance half: the authors warn devices "might be left on the breakroom table by workers who feel they represent an unacceptable level of intrusion," so a co-design study with workers on data ownership and alarm handling is a real contribution. Relevant skills: human physiology, sensor evaluation, standards drafting, statistics (Bland–Altman and agreement methods), participatory design, research ethics.
Genes
Structural cousins: stuck the same way, in other fields
Sources
Spencer Pizzani, Emanuele Cauda, Maggie Morrissey, William Mills, "Wearable Wisdom: The Promise and Challenge of Wearable Sensors for Heat Stress Management," *The Synergist* (AIHA), April 2023, accessed 2026-08-17; Dolson CM et al., "Wearable Sensor Technology to Predict Core Body Temperature: A Systematic Review," *Sensors* 22(19):7639, 2022, accessed 2026-08-17 go to source 1 ↗ go to source 2 ↗
Working notes
Notes made while writing and checking this brief, shown as written. They use working shorthand.
The Synergist is AIHA's professional journal for industrial hygienists — practitioner-to-practitioner discourse rather than popular media — and this article is co-authored by the director of the NIOSH Center for Direct Reading and Sensor Technologies, the secretary of the AIHA Thermal Stress Working Group, and the research officer of AIHA's Real-Time Detection Systems Committee, which is why it is tiered 2 rather than 3. All quoted sentences are verbatim from that article; the 25-algorithm / 592-subject / 32.1 ± 43.2 average sample / 26.4 ± 4.7 mean age figures are from the open-access Sensors systematic review (PMC9572283), read in full text. `failure:unrepresentative-data` is the primary tag: the validation populations (young, healthy, small, homogeneous) and the validated range (close agreement reported below 38 °C in the two device comparisons; largely controlled conditions) do not represent the workers or the temperatures where the device's output is consequential. `temporal:worsening` was considered — occupational heat exposure is rising — and rejected under false-positive pattern (c): the measurement barrier itself is unchanged; the exposed population and the deployed device count are what grow. `constraint:regulatory` was considered and left off because neither source names a regulator with jurisdiction over these devices, which makes the gap, as far as the sources show, an absence rather than a mismatch. Distinct from `labor-heat-stress-informal-agricultural-workers` in the existing corpus, which concerns affordability and applicability of heat monitoring for informal workers with no formal workplace; this brief concerns whether the instruments used in formal workplaces are accurate enough to act on, and the two would need different solutions.
Source type: Self-articulated (professional society and federal sensor-research center telling their own members the tools they are buying cannot be evaluated).
Verified at intake 2026-08-17: gate (net) + adversarial source check + contested-tag second coding.
Reconciliation 2026-08-21: Entry-point repair (C37 realism triage, score 3 — flag confirmed). The lead suggestion asked a student team to pilot the challenge protocol on a diverse subject panel, which means driving volunteers above 38 °C in a controlled climate against an invasive reference measure with medical monitoring standing by — a heat-strain trial no student IRB authorizes in a semester. Applying the design-the-trial default, the deliverable is now the protocol specification itself (work rates, conditions, reference measure, panel composition, required accuracy reporting, purchaser-facing pass/fail language), which is what the AIHA authors say the field lacks, plus an explicit line naming the infrastructure that running it would take. The modeling and worker co-design doors were sound and are kept; the modeling door now names a verified free reference implementation — Ioannou LG, Tsoutsoubi L, Mantzios K, Flouris AD, "A free software to predict heat strain according to the ISO 7933:2018," Industrial Health 57(6):711–720, 2019, https://doi.org/10.2486/indhealth.2018-0216 (open access, verified 2026-08-21) — so the perturbation study has a published physiological baseline to score against. No tags touched.
Reconciliation 2026-10-01: The whole write-up was read against its three named sources, all fetched on 2026-10-01: the Synergist article in full text (https://publications.aiha.org/202304-wearable-sensors-heat, direct), the Sensors systematic review in full text (open-access copy from Europe PMC, https://www.ebi.ac.uk/europepmc/webservices/rest/PMC9572283/fullTextXML, with the citation record at https://api.crossref.org/works/10.3390/s22197639), and for Ioannou et al. the abstract and citation record only (Crossref and Europe PMC records for https://doi.org/10.2486/indhealth.2018-0216), which is all the write-up relies on. A second reader went over the repairs on the same day, re-fetched the sources, and adjusted the wording of several; this note describes the final state. Changes: (1) The long title said the algorithms are "Secret" and "Validated Only on Young, Fit Volunteers Below the Temperature Where Anyone Actually Gets Hurt." The Synergist article says the algorithms are "often proprietary." The review describes young subjects (average age 26.4), with only a third of the algorithms including female subjects and most studies excluding people with medical conditions or on medication; it does not describe them as fit, and it does not say the algorithms were validated only below a given temperature. The 38 °C figure comes from the Synergist article's account of two device comparisons. The title is now "Heat-Stress Wearables Tell Workers Their Core Temperature Using Often-Secret Algorithms Validated Mostly on Young, Healthy Volunteers, With Device Agreement Reported as Close Below 38 °C." The Display title is unchanged. (2) "None of these devices measure core temperature; they estimate it from heart rate and skin temperature using proprietary algorithms" was stronger than the article, which says heart rate and skin temperature are "among the most common" inputs and the algorithms are "often proprietary"; the sentence now says "commonly" and "often proprietary." "Are buying" became "are turning to," the article's wording. (3) The Problem Statement said "the published agreement" with a "gold-standard" thermometer is close below 38 °C, "which is to say in the range where nothing is wrong." The 38 °C finding is from the two device comparisons the article cites, and the sentence now says so. The point about that range is kept but put in the review's own terms: a core temperature of around 38.5 °C is a normal response to exercise and generally not cause for concern, and clinical diagnosis and treatment of heat illness begins at 40 °C. (4) "No way for a buyer to evaluate the estimate, because the algorithm is a trade secret" now reads "a buyer may be unable to evaluate the estimate, because the algorithm is often proprietary and not disclosed." (5) "A wrong device is worse than no device," with its picture of a crew delegating the hazard to a green light, is not in either source and was replaced by what the article does say: simple outputs can make users overconfident, and it is tempting to conflate using a device with some measure of safety. "Already mis-specified" became "the authors themselves call uncomfortable," their word. (6) "The exact variables the validation samples of young fit volunteers hold constant" claimed more than the review reports (it says nothing about fitness or acclimatization of the samples); the sentence now lists what the review found: young subjects, only a third of the algorithms validated with any female subjects, most studies excluding medical conditions and medication. (7) Direct measurement "was tried and abandoned" became "is often ruled out"; the article says invasiveness and ethics "often preclude" its use. (8) The text said manufacturers ran both comparison studies, that both "matched well only below 38 °C," and that "both were small." The article says a manufacturer coordinated the firefighter trial; it does not say who ran the nurse study, gives no sample size for either, and reports a close match below 38 °C without saying what happened above it. The sentence now says this and quotes the article's own caveat that such studies are "limited in many ways." (9) The list of interferences that "all perturb the inputs with no disclosed sensitivity" is posed in the article as open questions and potential confounders of the measurements; the sentence now says so. Per-worker calibration "would help" became "might help," as in the article. (10) The sentence presenting respirator fit factors, gas-detector challenge testing and pulse-oximeter accuracy standards as "the adjacent precedent" is not in either source; it is now offered as a possible model that neither source raises, and was not checked against a source. (11) In the notes above, "(young, fit, ...)" became "(young, healthy, ...)", the "validated range" wording now names the two device comparisons, and "no regulator currently claims jurisdiction over these devices" now says only that neither source names one. (12) The review's benchmark is an RMSE "less than or equal to 0.5 °C"; "below 0.5 °C" became "at or below 0.5 °C." Confirmed in the sources and left as written: both citations on the Source line; all seven quotations from the Synergist article, which appear there word for word; 25 algorithms, 592 subjects, average sample 32.1 ± 43.2, average age 26.4 ± 4.7; 17 of 18 algorithms meeting the benchmark; few algorithms using individual and environmental data; the 70-kilogram reference body and the 171- and 200-pound averages; the in-ear devices; the alarm guidance; the three authors' roles; and the Ioannou et al. citation and its free software. Not checked, because they are the write-up's own judgments and neither source addresses them: that running the trial needs hospital-grade infrastructure no student protocol clears, and that the number of deployed devices is growing. No tags touched.