Some American insurers have begun integrating biometric data into their risk assessments. The global market for biometric health technologies will reach $11.57 billion in 2026, according to Towards Healthcare, with compound annual growth of 17.34%. The promise is real: objective physiological data in place of approximate administrative variables. But when the absence of data becomes itself a penalty, discrimination changes address without disappearing.

The essentials

  • The global health biometric market reaches $11.57 billion in 2026 and could exceed $48.77 billion by 2035 according to Towards Healthcare (CAGR of 17.34%).
  • Humana and UnitedHealthcare offer wellness and rewards programs using notably activity data.
  • The Commonwealth Fund 2026 confirms that no American state has eliminated disparities in access to care based on race or ethnicity.
  • Populations without reliable connectivity, without measurement devices, or without access to digital infrastructure risk having less complete biometric files.
  • The regulator has not yet set the threshold at which growing data granularity creates a class of invisible uninsurability.

The medical file had its limits

For decades, American insurers priced risk based on indirect variables: age, sex, zip code, declared medical history. These variables were crude. A forty-five-year-old man living in a disadvantaged neighborhood could be in better physical condition than his thirty-year-old sedentary neighbor. The administrative file could not make this distinction.

Biometrics promises to correct it. A heart rate monitor, an accelerometer, a pulse oximeter: these devices measure what the file declared by approximation. Humana launched incentive programs tied to smartwatch data. UnitedHealthcare Motion uses activity goals for rewards. The commercial logic is clear: measure better to discriminate better, in the statistical sense first, in the economic sense second.

Towards Healthcare estimates North America’s share at roughly 42% of the global market in 2025. North America represents neither 42% of the world population nor 42% of people with chronic illness: it concentrates the private insurers that have both the means to invest in these technologies and the most direct financial incentives to refine their segmentation.

The meaning of the word “objective” when the machine learns

The key word in this market is “objective.” Physiological data is presented as neutral because it measures a real physical signal: a heartbeat, a distance traveled, oxygen saturation. The argument has force. A pedometer doesn’t know if you’re Black or white, rich or poor.

But algorithms trained on this data are not separable from the populations that produced the training data. Studies, notably published in the New England Journal of Medicine, documented that pulse oximeters overestimated oxygen saturation in patients with darker skin, producing incorrect measurements that delayed treatment during the Covid-19 pandemic. The sensor was objective. The bias was structural.

This mechanism differs from intentional discrimination and, precisely for that reason, proves harder to contest. When an insurer refuses coverage or applies a surcharge invoking a “physiological risk score,” the complainant may have to demonstrate an identifiable practice, a disparate effect, and a causal link; access requirements for the model depend on procedure and applicable law. Some anti-discrimination protections, notably Title VI as judicially interpreted, require intent, while Title VII explicitly recognizes disparate impact in employment.

The Commonwealth Fund measures what algorithms won’t see

The Commonwealth Fund’s State Health Disparities Report 2026 arrives with a stable finding across several cycles: no American state has eliminated disparities in access to care based on race or ethnicity. The gaps concern infant mortality, undiagnosed chronic disease, delay before first cancer diagnosis.

These disparities include significant access gaps but stem from multifactorial causes, notably structural racism, social conditions, unequal coverage and treatment. A forty-year-old Black woman living in a rural medical desert does not have access to the same screenings as a forty-year-old white woman in a well-covered urban area. Her biometric file will be less complete because of documented inequalities in access to measurement and screening infrastructure.

The substitution of administrative files with biometrics produces a counterintuitive effect. The administrative file, crude as it was, recorded actual medical acts: a consultation, a hospitalization, a prescription. Continuous biometrics records the behaviors and physiologies of those who own the devices enabling them to be measured. Comparable dynamics are developing on the American market. The Commonwealth Fund measures the state of this terrain each year, already unequal before these technologies arrived.

The absence of data as implicit penalty

In this market, uses include identification, authentication, access control, and fraud prevention; measurement continuity does not by itself sum up the innovation. An annual health check-up is a snapshot; a smartwatch is a film. Risk models built on continuous data offer granularity superior to those built on periodic check-ups.

But this precision generally requires connectivity to transmit and synchronize data, although the measurement itself doesn’t necessarily require it. An Apple Watch costs between $250 and $800. A recent smartphone capable of reliably collecting health data represents comparable investment. In 2025, 34.5 million Americans lived below the official federal poverty threshold. Low-income people more often encounter obstacles to reliable connection, devices, and digital skills necessary for some continuous collection systems.

Insufficient network coverage can compromise continuous data transmission in some rural areas.

Insurers offering premium reductions in exchange for physiological data present the device as voluntary and incentive-based. But a “voluntary” device in a market where base premiums rise structurally produces real pressure on those who don’t participate. Non-sharing of data cannot, in the individual and small group markets subject to the ACA, directly serve to set a premium based on health status; collective wellness incentives remain possible under conditions. And this cost affects particularly those whose risk profiles don’t benefit from continuous biometric data.

This phenomenon joins a broader dynamic documented in other sectors. The delay between deployment of a technology and establishment of safeguards creates windows during which unintended effects take hold and become hard to untangle.

Data governance that sets the price, from 2026 to 2035

The global market should move from $11.57 billion in 2026 to $48.77 billion by 2035 according to Towards Healthcare. This projection rests notably on uses of identification, security, fraud prevention, and digital integration in care. It opens three plausible trajectories, distinct in the regulatory posture adopted.

In the first trajectory, regulation follows deployment with the usual lag of five to seven years. During this window, biometric pricing models take hold and create established facts: cohorts of insureds whose contract was built on continuous data, and unmeasured populations, for whom the absence of biometric data does not, in the individual and small group markets governed by the ACA, constitute an authorized pricing factor. When the regulator intervenes, it finds a market structured around practices it must disentangle without breaking existing contracts. The Affordable Care Act showed the difficulty of correcting after the fact: certain regulatory provisions implementing its anti-discrimination provisions faced suspensions or judicial injunctions.

In the second trajectory, the federal regulator or a bloc of states intervenes before practices uniform, by imposing two symmetric constraints: an obligation of algorithmic transparency for biometric pricing models, and a prohibition on implicitly penalizing the absence of data. This second constraint is hardest to operationalize, because it requires distinguishing a premium calculated on physiological data from a premium calculated on the absence of physiological data, two outcomes that can produce the same figure by different paths. The Federal Trade Commission has tools to investigate unfair business practices and has already stated a policy on biometric data and pursued actions regarding health data, but its framework specific to health algorithms remains partial.

The third trajectory is regulation through the data itself: public programs that expand biometric collection to underrepresented populations, so that learning models include physiological and socioeconomic diversity sufficient to reduce structural biases. The National Institutes of Health launched the All of Us initiative in this spirit, seeking to build a genomic and physiological database representative of American diversity. If insurers were required to integrate this data into their models, or simply incentivized to do so, the quality of representation in training data would change in nature.

Each of these trajectories produces distinct collective needs. The first requires capacity for rapid regulatory feedback once damage is documented. The second demands public expertise in algorithmic audit that American regulators don’t yet possess at necessary scale. The third assumes public investment in health data infrastructure whose funding is politically contested.

The concrete signals to follow are: the speed at which states, rather than the federal government, adopt laws on biometric health data privacy; the evolution of Humana and UnitedHealth practices under class action pressure; and the trajectory of the All of Us initiative, which will give in the next three years a first indication of political will to build representative databases as counterweight to private ones.

Precedents that guide without resolving

Auto insurance went through a comparable shift. The introduction of telematics boxes, which measure actual driving instead of age and zip code, was presented as an objectification of risk. Analyses conducted in the United Kingdom and United States indicate that drivers in disadvantaged neighborhoods, who circulate more at night and on denser roads, encountered difficulties with telematics models. Their risk profiles reflected more statistically riskier road configurations than less safe driving behavior.

Health insurance has added complexity: health is both an individual and collective good. An auto insurer can argue it merely reflects each driver’s statistical risk. A health insurer operates in a system where access to care directly determines mortality, and where the Commonwealth Fund establishes geographic disparities in care and outcomes at state scale. The concentration of benefits on those already starting in better position is a cross-cutting phenomenon, not specific to health insurance.

The difference between a technology that reduces inequality and one that crystallizes it is not in the sensor. It is in the rules governing what can be done with the data it produces.

Sources

  1. Coherent Market Insights – Healthcare Biometrics Market (2026)
  2. Commonwealth Fund – 2026 State Health Disparities Report (commonwealthfund.org)
  3. National Institutes of Health – Initiative All of Us Research Program (allofus.nih.gov)
  4. Toward Healthcare – health biometric market data North America (2026)