What insurers don't ask about buildings

Key findings

• David Mudarri put the annual US cost of illness from damp and moldy housing at 22.4 billion dollars, the sum of four diagnosis groups (Mudarri, 2016, Journal of Environmental and Public Health). The better known figure of 3.5 billion dollars from his earlier work with William Fisk covers asthma alone (Mudarri and Fisk, 2007, Indoor Air).
• As of June 2026 there is no documented case of a health, life or property insurer using building health data in underwriting. The closest step on record is Chubb joining the Well Living Lab Alliance in March 2024, an advisory role without premium consequences.
• UnitedHealth Group has invested more than 1.2 billion dollars in affordable housing since 2011 and has no audited health effect to show for it. The 115 dollar per member figure that circulates with the program comes from a 2016 Providence study without a control group. The 60 percent drop in emergency admissions is the company’s own number from one unnamed state.
• US law bars building quality from individual and small group premiums. Rates may vary only by age, rating area, tobacco use and family size (42 U.S.C. § 300gg). Whatever advantage exists runs outside pricing: through care management, the risk models behind it, the large group and self funded business and the investment book.
• The cost evidence is consistent across four estimation rounds and two decades. Whether an insurer can capture that value is untested, which is why this report is exploratory.

A billion dollars for housing, no question about the building

UnitedHealth Group, the largest US health insurer, reported in June 2025 that its affordable housing investments had passed 1.2 billion dollars since 2011. The money helped create and preserve 32,000 homes across 33 states and the District of Columbia (UnitedHealth Group, June 2025). The pace is rising: 400 million dollars by 2019, a billion by May 2024, 1.2 billion a year later. The program’s metric is the number of homes. Whether those homes are damp or dry, high radon or low, appears nowhere in the reporting.

Two effect figures travel with this program, and neither survives a source check. The first is 115 dollars less in health care costs per member per month. That number is not UnitedHealth’s. Bill Wright and his team at CORE, the outcomes research center of Providence Health & Services in Portland, analyzed Oregon Medicaid claims of 1,625 residents in 145 affordable housing properties and found an average reduction of 115 dollars per member per month, but only in properties with integrated health services, in a before-after design without a control group (Wright et al., CORE, Health in Housing, February 2016). Enterprise Community Partners commissioned the study, and UnitedHealth invests in housing through Enterprise vehicles, which is the likely bridge for the mix-up: Yahoo Finance attributed the number to UnitedHealth in 2022 without naming a source (Khemlani, Yahoo Finance, 2022), and it has been quoted that way since, including in an earlier essay of mine.

The second figure is UnitedHealth’s own, and it is thinner than its circulation suggests. In a March 2019 press release the company wrote that “in one state, emergency room admissions dropped 60 percent” for homeless Medicaid members enrolled in a housing program (UnitedHealthcare, March 2019). The release names no state, no sample size and no methodology, and it has since disappeared from the company newsroom; it survives in the Internet Archive. The insurer that has spent the most on the housing and health connection measures availability, borrows its impact numbers and cannot say what building quality does to its own claims.

What insurers do ask is well defined. In the US individual and small group market, the Affordable Care Act allows premiums to vary by exactly four factors: whether the plan covers an individual or a family, rating area, age within a three to one band and tobacco use within one and a half to one (42 U.S.C. § 300gg). Medicaid managed care programs experiment with social risk factors such as income and neighborhood. In Switzerland and Germany, solidarity systems rule out individual risk pricing almost entirely. Across all of these systems the building appears nowhere, neither as a rating factor where rating is allowed, nor as a variable in the risk models behind care management.

The cost side has been priced four times

The research that insurers are not using has a forty year history. Povl Ole Fanger built the measurement basis for how indoor environments affect people, first for thermal comfort (Fanger, Thermal Comfort, 1970, Danish Technical Press), later with units for perceived air quality (Fanger, 1988, Energy and Buildings 12(1):1-6).

William Fisk and Arthur Rosenfeld made the first national cost estimate in 1997 (Fisk and Rosenfeld, 1997, Indoor Air 7(3):158-172). Fisk refined it three years later: better indoor environments in the US could avoid 6 to 14 billion dollars a year in respiratory disease, 1 to 4 billion in allergies and asthma and 10 to 30 billion in sick building syndrome symptoms, plus 20 to 160 billion in productivity gains that are not illness costs (Fisk, 2000, Annual Review of Energy and the Environment 25:537-566). Fisk reported these as separate bands and never added them up. Quoting a single total from his work misstates it.

David Mudarri and Fisk then narrowed the lens to damp and mold in homes. They estimated that 21 percent of current US asthma is attributable to dampness and mold, about 4.6 million cases, at an annual cost of 3.5 billion dollars (Mudarri and Fisk, 2007, Indoor Air 17(3):226-235). In 2016 Mudarri widened the calculation to four diagnosis groups: 3.7 billion dollars for allergic rhinitis, 1.9 billion for acute bronchitis, 15.1 billion for asthma morbidity and 1.7 billion for asthma mortality, together 22.4 billion dollars a year (Mudarri, 2016, Journal of Environmental and Public Health). The jump from 3.5 to 22.4 is not a contradiction. It is the same author covering more diseases with fuller cost accounting.

These are model estimates, attributable fractions laid over epidemiology, not measured claims. A different method points the same way. The Swiss Federal Office of Public Health counts 200 to 300 lung cancer deaths per year from radon in buildings, close to ten percent of all Swiss lung cancer deaths, based on an epidemiological estimate by Menzler and colleagues (BAG, Radon Action Plan 2021-2030; Menzler et al., 2005). More than ten percent of the over 150,000 measured Swiss buildings exceed the reference value of 300 becquerel per cubic meter. In France, Boulanger and colleagues put the social cost of indoor air pollution at about 19 billion euros a year (Boulanger et al., 2017, Environment International 104:14-24). Roger Ulrich delivered the clinical end of this evidence chain in 1984: patients with a window view of trees recovered faster after surgery (Ulrich, 1984, Science 224(4647):420-421), covered in detail in TBR-01.

The dispute is not about these numbers, it is about whether an insurer can capture any of this value.

Blind spot or design decision

The healthy building reading of this gap, and an earlier essay of mine took that reading, is that the data are simply unconnected: building science knows the effects, actuarial science never looked. Three serious objections stand against it, and each comes with a name.

The actuaries have looked, at least once. Rong Yi and Jeffrey Milton-Hall, actuaries at Milliman, tested for the Society of Actuaries Research Institute what social risk data add to Medicaid risk adjustment models. The social factors “explained much less variation in estimated risk” than existing diagnosis categories, the financial impact of integrating them was “effectively undetectable” under an alternate sampling method, and individual level social data are “currently not reliably populated” (Yi, Milton-Hall et al., 2025, Society of Actuaries Research Institute). That is the profession’s answer to the charge of blindness: we ran the numbers and the gain was too small to see.

Amy Finkelstein ran the only randomized test of the adjacent promise, that targeting high cost patients with social support lowers costs. The Camden Coalition program had no effect on readmissions (Finkelstein et al., 2020, New England Journal of Medicine 382(2):152-162). Her broader point reaches every before-after number in this field, including UnitedHealth’s 60 percent: interventions start when people are at their sickest and most expensive, so costs fall afterward with or without the program. “They’re almost by construction going to be plagued by the issue of regression to the mean” (Finkelstein, MIT News, January 2020).

This objection reaches UnitedHealth directly. Jeffrey Brenner, who built the Camden model Finkelstein tested, went on to lead myConnections, UnitedHealth’s housing program for homeless members. A randomized evaluation of that program was planned; no results have been published. The California Health Care Foundation reviewed its Phoenix pilot in 2019 and counted 41 participants, with the program cost effective only for the 25 most expensive members and costs rising for 16 of them (CHCF, November 2019).

Len Nichols and Lauren Taylor add the business objection: benefits from investments in social conditions are public goods. The insurer that pays cannot keep the benefit, because members switch plans and every competitor profits from a healthier pool (Nichols and Taylor, 2018, Health Affairs 37(8):1223-1230). The National Academies committee on permanent supportive housing, chaired by Kenneth Kizer, found “no substantial published evidence” that such housing reduces health care costs (NASEM, 2018).

Where US insurers price individuals, the ACA forbids any factor beyond the four listed above. The building is missing from premium models not because no one thought of it, but because the law does not allow it in.

Position

The usual reading is that insurers are blind to building quality. I hold that half right. In pricing they are not blind, they are fenced in by law, and in Europe by the solidarity principle. The blind spot is real but sits where the fences are not: in care management, in the risk models behind it, in large group and self funded business and in the investment book. The UnitedHealth case shows it precisely. Thirteen years and 1.2 billion dollars into housing, the company still counts homes, not building conditions, and borrows its effect numbers. The actuarial counter-evidence is weaker than it looks: Yi and Milton-Hall fed their models area level proxies and never a measured building, so their null result argues for better building data rather than against the idea. Finkelstein’s null and the Kizer committee tested housing access for the very sick, not building quality for the insured population; asthma in damp homes is a different mechanism with stronger dose response behind it. The first insurer that links measured building conditions to its own claims will learn within a few years whether Mudarri’s 22.4 billion shows up in its book. Nobody can buy that knowledge today at any price; the dataset does not exist. That is the advantage, and it is slower and narrower than I first assumed: it pays first in care management and investment selection, last in premiums.

What the evidence can carry

The cost estimates above are syntheses and models. The program evaluations that do exist measure housing access, not building quality, and they disagree. John Lovelace and colleagues, all at UPMC’s insurance division, reported emergency department use down 50.1 percent after members moved into supportive housing (Lovelace et al., 2025, Health Services Research 60(Suppl 3):e14411), an insurer evaluating its own program. Rahma Mkuu and colleagues at the University of Florida evaluated Florida’s Medicaid housing pilot and found transitional housing support associated with 15 percent more emergency visits, while tenancy support was associated with 51 percent lower odds of death (Mkuu et al., 2026, Health Affairs 45(1):68-75). Mixed results in access studies say little about the building quality question, because none of these studies measured the buildings.

That is the actual research gap. There is no dataset anywhere that links measured indoor conditions of homes to insurance claims at scale. The certification stock that could seed one is smaller than its marketing: the International WELL Building Institute reports more than 6 billion square feet “engaged” across roughly 100,000 locations in 137 countries, but engaged means enrolled, not certified, and in housing the certified or precertified count is about 3,000 units (IWBI, July and December 2025). IWBI sells the certification, so these are vendor numbers. The research gap is also the business gap: whoever builds the linked dataset closes both.

Implications

For health insurers

The pricing route is closed in regulated individual markets and was never open in solidarity systems. The open routes are care management, the large group and self funded business and the design of Medicare Advantage supplemental benefits. A workable pilot is narrow: take asthma, the best priced condition in the literature, link members’ claims to measurable home conditions (damp, mold history, radon zone, certification status) in one regional book and test whether Mudarri’s attributable fraction appears. The actuarial method exists; it has just never been fed building level data.

For reinsurers

Waiting is rational until a primary carrier links measured building conditions to its own claims, and cheap optionality is available now: re-run the Yi and Milton-Hall design with building level instead of area level variables before pricing any of this into treaties.

For asset owners with certified stock

Certification becomes insurance negotiating material only when an insurer can read it. Sharing building performance data early, under terms, costs little and sets the standard others will have to meet. The certified residential stock is so small that early movers are visible.

For sensor makers and health tech

The slot is empty as of June 2026. Chubb’s step into the Well Living Lab Alliance is advisory (Well Living Lab, March 2024), and where insurers already use building sensors, at Munich Re’s Hartford Steam Boiler, the logic is water damage, not health. The product that creates the market is not another air quality dashboard, it is the consented data link between a home’s measurements and a claims file.

For developers

Do not price insurance compensation for healthy buildings into pro formas yet. On the evidence above it is a five year possibility, not a bankable line.

Stop doing

Stop citing the 115 dollar and 60 percent pair as proof that housing investment pays; the first number is not the insurer’s and the second has no methodology behind it. Stop reading WELL’s 6 billion engaged square feet as certified stock. Stop waiting for premium discounts on healthy buildings in regulated individual markets; the law forbids them.

Methodology

I researched this report manually in June 2026 from primary sources: peer reviewed cost studies, the US Code, Swiss federal documents, company statements and press archives. I did not query the TBR study corpus. Every figure is bound to its primary document, and I traced the two effect figures circulating around UnitedHealth’s housing program to their actual origins, one to a 2016 Providence study, one to an archived 2019 press release. Twenty-two named sources. Limits: the US cost figures are model estimates, not measured claims; the absence of underwriting pilots is a search result and can miss unannounced programs; I verified Mkuu’s Florida results against the journal listing, not the paywalled full text, and Nichols and Taylor’s argument at abstract level. The cost burden of poor buildings is robust across methods. The capture mechanism for insurers is untested, which is why this report is exploratory.