Federal Housing Assistance and Blood Lead Levels in a Nationally Representative US Sample Age 6 and Older: NHANES, 1999–2018

Abstract

Background:

Federal housing assistance is an important policy tool to ensure housing security for low-income households. Less is known about its impact on residential environmental exposures, particularly lead.

Objectives:

We conducted a quasi-experimental study to investigate the association between federal housing assistance and blood lead levels (BLLs) in a nationally representative US sample age 6 y and older eligible for housing assistance.

Methods:

We used the 1999–2018 National Health and Nutrition Examination Survey (NHANES) linked with US Department of Housing and Urban Development (HUD) administrative records to assess BLLs of NHANES participants with concurrent HUD housing assistance (i.e., current recipients, $n=3,\mathrm{0}71$) and those receiving assistance within 2 y after the survey (i.e., pseudo-waitlist recipients, $n=\mathrm{1,235}$). We estimated BLL least squares geometric means (LSGMs), odds ratio (OR) for BLL $\ge 3.5\mathrm{\mu }\mathrm{g}/\text{dL}$, and percent differences in LSGMs by HUD housing assistance status adjusting for age, sex, family income-to-poverty ratio, education, country of birth, race/ethnicity, region, and survey year. We also examined effect modification using interaction terms and stratified analyses by program type [i.e., public housing, multifamily, housing choice vouchers (HCV)], and race/ethnicity.

Results:

Current HUD recipients had a significantly lower LSGM [$1.07\mathrm{\mu }\mathrm{g}/\text{dL}$; 95% confidence interval (CI): 1.02, 1.12] than pseudo-waitlist recipients ($1.21\mathrm{\mu }\mathrm{g}/\text{dL}$; 95% CI: 1.14, 1.28), with an adjusted OR of 0.60 (95% CI: 0.42, 0.87) for BLL $\ge 3.5\mathrm{\mu }\mathrm{g}/\text{dL}$. Some effect modification were observed: The protective association of HUD assistance on BLL was strongest among public housing ($-19.5\%$ LSGM; 95% CI: $-27.5\%$, $-1\mathrm{0}.7\%$), multifamily ($-12.5\%$ LSGM; 95% CI: $-2\mathrm{0}.7\%$, $-3.5\%$), and non-Hispanic White ($-2\mathrm{0}.6\%$ LSGM; 95% CI: $-29.8\%$, $-1\mathrm{0}.3\%$) recipients. It was weaker to null among HCV ($-5.7\%$ LSGM; 95% CI: $-12.7$, 1.7%), non-Hispanic Black ($-1.6\%$ LSGM; 95% CI: $-8.1\%$, 5.4%), and Mexican American (−12.5% LSGM; 95% CI: $-31.9\%$, $12.5\%$) recipients.

Discussion:

Our research underscores the importance of social-structural determinants like federal housing assistance in providing affordable, stable, and healthy housing to very low-income households. More attention is needed to ensure housing quality and racial equity across HUD’s three major housing assistance programs. https://doi.org/10.1289/EHP12645

Introduction

Lead is a ubiquitous environmental toxin with deleterious effects on the nervous, hematopoietic, endocrine, renal, and reproductive systems. Higher blood lead levels (BLLs) in adults have been consistently associated with elevated blood pressure and risk of cardiovascular disease, renal insufficiency, and cognitive impairments.^1–5 A nationally representative longitudinal study of US adults from 1988 to 2011 found that even at low BLLs, an increase from $1.\mathrm{0}\mathrm{\mu }\mathrm{g}/\text{dL}$ to $6.7\mathrm{\mu }\mathrm{g}/\text{dL}$ was associated with increased risks of all-cause, cardiovascular, and ischemic disease mortality.⁶ Young children are particularly susceptible to the effects of lead because their nervous systems are still in development and they more readily absorb lead in comparison with adults.⁵ Elevated BLLs among children have been associated with neurocognitive and intellectual impairments,⁴ poor school performance,⁷ behavioral problems,⁷ and criminality later in life,⁸ even at low levels of exposure.

The major sources of blood lead are residential, with lead-based paint the most prevalent source of lead exposure, particularly in homes built before lead-based paint was banned in 1978.⁹ Currently, $\sim \mathrm{29}$ million households have lead-based paint hazards, such as deteriorated paint and residues in lead-contaminated dust and soil, and close to 2.6 million of these households have young children.^9,10 Other residential sources of lead include the corrosion of lead water pipes in drinking water; lead-containing consumer products (e.g., toys, jewelry); and lead in foods (e.g., candies), cosmetics, and traditional home remedies.⁹ Lead can also be brought home from occupational activities such as ore mining, smelting, and manufacturing of lead-containing products and pesticides.^2,5,11 There are also anthropogenic sources, such as coal and oil combustion, waste incineration, and the use of leaded jet fuel.⁵

The reduction of BLLs in the general US population is a national public health priority.^9,12–14 Although BLLs have declined in the United States since the 1980s,¹⁵ the US Centers for Disease Control and Prevention have deemed that there are no safe levels of lead in the body for children.¹⁶ Based on the 2011–2016 National Health and Nutrition Examination Survey (NHANES) data, Egan et al.¹⁷ found that 386,775 children age 1 to 11 y still had BLLs $\ge 5\mathrm{\mu g}/\text{dL}$. In the same data source, Ettinger et al.¹⁸ found that more than 500,000 women of childbearing age had BLLs $\ge 5\mathrm{\mu g}/\text{dL}$.

In addition, socioeconomic and racial inequities in lead exposure persist, with the risk of lead exposure significantly higher among families living in older homes, families living in poverty, and non-Hispanic Black households.^17,19,20 Low-income families have limited options for securing safe and affordable housing. Housing stock that is affordable to rent in the private market usually consists of older homes built before 1978 that are more likely to have physical deficiencies and environmental hazards.^10,21,22 As a result, low-income families often must sacrifice housing quality for affordability, which could then increase their risk of lead exposure.^19,20,23 Furthermore, persistent racial/ethnic disparities in lead exposure are driven by historical and current-day racism at the institutional and interpersonal levels that limit access to quality housing and neighborhoods for racially minoritized families. These practices include redlining, zoning, land use restrictions, housing discrimination, and government and corporations’ decisions to place pollution-emitting developments like highways, waste sites, and industries in primarily low-income Black and Brown neighborhoods.^24–27 These forms of structural racism contribute to disproportionately higher levels of exposure to contaminants like lead among racially minoritized populations.

To address racial/ethnic and socioeconomic disparities in BLLs, existing research has focused primarily on identifying proximate drivers of lead exposure, such as lead dust,^28–30 age of housing,¹⁹ and neighborhood characteristics.^19,31 Few studies have focused on structural determinants such as housing policy and housing assistance programs. Housing assistance programs that provide low-income families access to quality and affordable housing may reduce unsafe lead exposure. The US Department of Housing and Urban Development (HUD) provides affordable housing assistance to nearly 5 million families, including about 3 million children, through three major programs administered by local public housing agencies: public housing (0.84 million households), tenant-based housing choice vouchers (HCVs) (2.3 million households), and multifamily income-restricted housing (1.4 million households).³² Tenant-based HCV is HUD’s largest program that provides tenants with subsidies to secure housing units in the private market that meet unit size eligibility and pass a visual assessment. HUD requires that 75% of vouchers be reserved for families with incomes below 30% of the area median income (AMI). In contrast, project-based housing are subsidized units at a particular site that are either publicly or privately owned and managed. Public housing programs provide project-based units owned and administered by local public housing authorities for low-income families, elderly participants, and persons with disabilities, with 40% of units reserved for families below 30% of AMI. Multifamily income-restricted programs provide mostly project-based Section 8 units in privately owned developments where owners receive subsidies to offer affordable rates for assisted families.³²

HUD housing assistance has been shown to be an effective platform for decreasing the financial burden of housing and improving residential stability.^33–35 It has also been associated with positive mental and physical health outcomes in quasi-experimental studies of children and adults.^36–40 Regarding lead exposure, federal housing assistance may facilitate lower BLLs due to the enforcement of residential lead laws in HUD housing units.^39,40 In recent years, HUD’s Office of Lead Hazard Control and Healthy Homes has allocated over $\text{USD\hspace{0.17em}}\$125\text{\hspace{0.17em}million}$ to protect families from lead and other home hazards.⁴¹ The 2005–2006 American Healthy Homes Survey found that households receiving government rental assistance were less likely to have lead-based paint hazards than those not receiving such support (12% vs. 22%).²³ Similarly, another observational study of children ages 1 to 5 y from the 2005–2012 NHANES found that BLLs were lower among HUD-assisted children compared to BLLs among a propensity score–matched sample of non-HUD–assisted children.⁴² However, other studies have found elevated lead levels in public housing sites across the United States,^43–46 which suggests that HUD’s regulatory enforcement of lead laws and/or property maintenance may not be uniformly effective.

This study’s objective is to better understand the impact of HUD housing assistance on BLLs in a nationally representative sample of the US population eligible for HUD housing assistance. We conducted a quasi-experimental study to investigate differences in BLLs among current and pseudo-waitlist HUD recipients. In addition, to the best of our knowledge, no study to date has evaluated BLLs by HUD assistance program type, despite variability in housing quality, affordability levels, and regulatory oversight across these programs that may contribute to differences in lead exposure. Prior studies have also not examined racial/ethnic disparities in BLLs by HUD housing assistance. Therefore, our secondary objective is to examine the effect modification of HUD housing assistance and BLLs by program type and race/ethnicity.

Methods

NHANES Sample Design and HUD Data Linkage

We analyzed data from 10 waves of NHANES data from 1999 to 2018 linked to HUD administrative records from 1999 to 2019.⁴⁷ NHANES is a nationally representative, cross-sectional, household survey of civilian, noninstitutionalized US populations conducted in 2-y waves using a complex, stratified, multistage probability sampling design. NHANES ascertains health and nutritional status information of children and adults through interviews and physical examinations, including biomarker analyses of environmental chemicals. HUD administrative records provide information about housing assistance status, income, and program type for recipients in the 50 US states and Washington, DC.³² The linkage of NHANES participants and HUD recipients is based on deterministic and probabilistic matching using personally identifiable information, such as Social Security Number, first and last names, and date of birth. Linkage-eligible NHANES participants matched with HUD administrative records data had to have provided consent as well as sufficient personally identifiable information.⁴⁷

Our primary analytic sample included all 1999–2018 NHANES participants who were linkage-eligible and present in the HUD administrative records at any time from 1999 to 2019, regardless of their HUD assistance status on the date of their NHANES interview. The HUD administrative records provide a longitudinal record of entry and exit dates for participants on HUD assistance and the specific types of assistance program they entered (e.g., public housing, multifamily income-restricted, and HCV), which were then used to generate housing histories for each NHANES participant. For example, if a participant lived in HUD housing from 2001 to 2008 and interviewed for NHANES in 2006, the NHANES-HUD linked dataset would provide information about the year the participant entered (i.e., 2001) and exited (i.e., 2008) HUD housing and any events involving HUD during that time period (e.g., if the person exited and reentered HUD housing multiple times between 2001 and 2008). Due to confidentiality and disclosure policies, the linked NHANES-HUD data is restricted-use and can only be accessed through the National Center for Health Statistics (NCHS) Research Data Center.⁴⁷ No institutional review board protocol approval was needed for this study because secondary data was used.

Housing Assistance Status

The receipt of HUD housing assistance is not randomly assigned due to characteristics that may select families into housing assistance, such as economic hardship, knowledge of government programs, and motivation to participate.^48,49 Therefore, we used a quasi-experimental, pseudo-waitlist approach to examine the association between HUD housing assistance and BLLs. A naïve comparison of participants with and without housing assistance would be insufficient to address potential selection biases in HUD eligibility and entry. We leveraged the timing of entry into HUD housing assistance with respect to the NHANES interview date to create our main comparison groups: current recipients of HUD housing assistance at the time of the NHANES interview and pseudo-waitlist recipients who would enter a HUD assistance program within 2 y after the NHANES interview (i.e., “pseudo-waitlist”) (Figure S1). As such, our study restricts the study population to only HUD recipients, i.e., individuals with similar eligibility for and experiences of receiving HUD housing assistance during or after the interview. This pseudo-waitlist approach has been used in prior studies to assess HUD housing assistance.^{35–38,50,51} In addition, we selected a 2-y waiting list duration to reflect the national average wait time for HUD housing assistance at the time of our analysis.³² Sensitivity analyses conducted in prior studies confirmed that over 70% of adults in the pseudo-waitlist group were on the actual waiting list and had sociodemographic characteristics similar to those on the actual waiting list.^38,51

Study Sample

Our study sample was restricted to NHANES participants ages 6 y and older at the time of the survey (Figure S2). We excluded children ages 1–5 y, given known heterogeneity in BLL exposure and effects in this age group in comparison with older age groups.¹⁵ The association of HUD assistance and BLLs has also been previously studied in children ages 1–5 y.⁴² Of participants in the NHANES 1999–2018 surveys linked to a HUD administrative record, 5,129 were either current or pseudo-waitlist HUD recipients (i.e., within 2 y after the NHANES survey). We then excluded recipients without BLL measurements ($n=823$, 16%). Our final analytical sample included 4,306 participants: 3,071 current and 1,235 pseudo-waitlist HUD recipients ages 6 y and older at the time of the NHANES survey. Participants were also categorized by assistance program type: HCV ($n=\mathrm{2,4}\mathrm{0}1$), public housing ($n=92\mathrm{0}$), and multifamily income-restricted ($n=985$). Because HUD recipients may have multiple entry and exit points in various HUD housing programs, for current HUD recipients we selected the program type concurrent with their NHANES participation, and for pseudo-waitlist recipients we selected the first program type recorded for our analysis. Our study sample also included a small proportion of pregnant and postpartum women.

BLLs

NHANES has been the cornerstone for lead exposure surveillance in the United States since 1976.⁵² Details about NHANES analytical methodology for BLLs are available elsewhere.^53,54 In brief, whole blood specimens were collected by phlebotomists at the NHANES mobile examination center (MEC) in an ethylenediaminetetraacetic acid–coated tube via venipuncture. Samples were processed on site and stored frozen before being shipped on dry ice to the US CDC National Center for Environmental Health (NCEH) in Atlanta, Georgia, where heavy metal assays were performed. Blood specimens remained frozen until they were analyzed. Lead concentrations were quantified in whole blood samples at NCEH using atomic absorption spectrometry with Zeeman background correction for survey periods from 1999 to 2002 and inductively coupled mass spectrometry for later survey periods from 2003 to 2018. The limit of detection (LOD) for BLLs has improved over time: $0.30\mathrm{\mu }\mathrm{g}/\text{dL}$ in the period 1999–2002, $0.28\mathrm{\mu }\mathrm{g}/\text{dL}$ in 2003–2004, $0.25\mathrm{\mu }\mathrm{g}/\text{dL}$ in 2005–2012, and $0.07\mathrm{\mu }\mathrm{g}/\text{dL}$ in 2013–2018. For BLLs below the LOD ($<1.\mathrm{0}\%$), we imputed values by dividing the LOD by the square root of 2 for each survey cycle as recommended by US CDC.⁵⁵ NHANES also has rigorous protocols to monitor and address outliers and other data quality concerns in its laboratory data.^53,54

Demographic and Socioeconomic Characteristics

Demographic and socioeconomic characteristics of participants were self-reported in the NHANES interview. Age of interview was categorized as 6–11, 12–19, 20–40 [Reference group (Ref)], 41–61, and $62+$ years of age. We created a race/ethnicity indicator by combining responses to questions on race and Hispanic origin across the 10 waves. Race/ethnicity categories included non-Hispanic (NH) White (Ref), non-Hispanic Black, Mexican American, and Another race (i.e., a grouped category that includes other Hispanic, NH Asian, and NH Multiracial recipients to ensure stable estimates). Mexican American was treated as a separate category from other Hispanic ethnicities because their classification was available in the NHANES data. We treated self-reported race/ethnicity in our analysis as a proxy for structural racism, which we hypothesized is the main driver of observed racial disparities in BLLs.^5,17,20 Participants’ sex was categorized based on binary response options of male (Ref) and female. We treated self-reported sex as a proxy for both potential biological differences in the disposition and toxicity of BLLs and as a social construct (i.e., differences in gender may reflect differences in structural factors such as sexism).⁵⁶ Participants’ country of birth was categorized based on binary response options of United States (Ref) and outside the United States. We adjusted for country of birth to account for variability in HUD eligibility based on immigration status⁵⁷ and elevated BLLs observed among immigrants.⁵⁸ Socioeconomic status was assessed using educational attainment of the reference householder (who was at least 18 y of age at the time of NHANES survey) [up to high school; high school degree/some college; and college degree (Ref)] and the family income-to-poverty ratio (FIPR), which was based on the federal poverty threshold for the specific survey period. Because our study population was primarily low-income, we categorized FIPR as $<\mathrm{0}.5\mathrm{0}$, 0.50–0.99 (Ref), 1.0–1.99, and $2.\mathrm{0}+$. Approximately 2.2% of recipients were missing householder educational attainment and 6.7% were missing FIPR, which we excluded from our analyses. As such, we conducted complete case analyses because the chance of being a complete case is likely not dependent on the observed outcome, i.e., BLLs,⁵⁹ which were collected in the MEC and separate from the NHANES interview data.

Statistical Analyses

We conducted analyses in R Studio (version 1.3.9; R Studio Team) with base R (version 4.0.2; R Core Team), using the “survey” package to account for the NHANES’ nonrandom sampling design and sample population weights.⁶⁰ We used NHANES-HUD adjusted MEC weights (ADJ_MECWT) that provide representative estimates of the age, sex, and race/ethnicity distributions of linkage-eligible populations in the United States from 1999 to 2018. The weights account for unequal probabilities of selection, nonresponse, oversampling, and HUD linkage eligibility.⁴⁷ Because we combined 10 survey waves from the years 1999 to 2018, we divided the provided weights by 10 and used the Taylor Series Linearization method for variance estimation, as recommended by the NCHS.⁶¹ A two-sided $p$-value of below 0.05 was considered statistically significant for all analyses.

Sources of potential confounding were assessed a priori based on the literature. Although our pseudo-waitlist approach helped to ensure comparable groups by restricting the study population to only those receiving HUD assistance, we also took an analytical approach and adjusted for potential confounders that may have changed among pseudo-waitlist recipients during the time between their NHANES survey and entry into HUD assistance, such as age, education, and FIPR. Prior studies have documented differences in income by HUD assistance status. For example, a prior study found that residents receiving rental assistance often had higher earnings than their wait-listed counterparts—an effect referred to as the Ashenfelter Dip—because tenants often apply for assistance at their lowest economic point.⁶² However, more recent studies using the pseudo-waitlist approach have found lower FIPRs among current recipients in comparison with pseudo-waitlist HUD recipients, which may be attributed to current recipients reducing their income levels to maintain eligibility for HUD assistance. In addition, HUD assistance may be associated with a reduction in household size, which can reduce overall household income.^37,38 We also adjusted for major US Census 2000 regions (i.e., North, East, West, South), which was masked in our dataset by the NCHS to minimize disclosure risks, and for survey year as linear and quadratic terms to account for temporal variability. Because 93% of our sample lived in urban areas, we did not adjust for urbanization status but conducted a sensitivity analysis restricting to recipients in urban areas.

We computed frequency distributions of sociodemographic characteristics and weighted geometric means (GM) and geometric standard errors (GSE) for BLLs by timing of HUD housing assistance (i.e., current vs. pseudo-waitlist). BLL was treated as both a continuous and dichotomous outcome. For linear regression analyses, we log-transformed BLLs to meet the normality assumption of residuals. Our model for BLLs took the form $log\left(y_i\right)={\mathrm{\beta }}_0+{\mathrm{\beta }}_{Hlin}H+{\mathrm{\beta }}_{X}X+{ϵ}_i$, where ${ϵ}_i\sim N(0,{\mathrm{\delta }}^2)$ is a normally distributed error term; $y_i$ is the BLL for individual i; H is housing assistance status (1 = current, 0 = pseudo-waitlist); and X is a vector of covariates. We then estimated the least squares geometric mean (LSGM) for BLLs by housing assistance status as $\left(e^{leastsquaremeans}\right)$ and the 95% confidence intervals (CIs) as $\left(e^{leastsquaremeans\pm 1.96\times standarderror}\right)$, where $least square means$ is the mean of BLLs by housing assistance status after adjustment for covariates.⁶³ We also estimated the percent difference in LSGM by housing assistance status as $(e^{\widehat{{\mathrm{\beta }}_{Hlin}}}-1)\times 100\%$ and the 95% CIs as $(e^{\widehat{{\mathrm{\beta }}_{Hlin}}\pm 1.96\times standarderror}-1)\times 100\%.$

Our logistic regression model for BLL $\ge 3.5\mathrm{\mu }\mathrm{g}/\text{dL}$ took the form $\text{logit}\left(p\right)=\log \left[\frac{p}{1-p}\right]={\mathrm{\beta }}_0+{\mathrm{\beta }}_{H\mathit{log}}H+{\mathrm{\beta }}_{X}X$ , where $p$ is probability of BLL $\ge 3.5\mathrm{\mu }\mathrm{g}/\text{dL}$; H is housing assistance status (1 = current, 0 = pseudo-waitlist); and X is a vector of covariates. We estimated the odds ratio (OR) of having an elevated BLL $\ge 3.5\mathrm{\mu }\mathrm{g}/\text{dL}$ by HUD assistance status as $\left(e^{\widehat{{\mathrm{\beta }}_{Hlog}}}\right)$ and the 95% CIs as $e^{\widehat{{\mathrm{\beta }}_{Hlog}}\pm 1.96\times standarderror},$ where $\widehat{{\mathrm{\beta }}_{Hlog}}$ is the estimated logistic regression coefficient after adjustment for covariates. We set $3.5\mathrm{\mu }\mathrm{g}/\text{dL}$ as our threshold based on prior studies finding decreases in children’s IQ levels at BLLs below $5\mathrm{\mu }\mathrm{g}/\text{dL}$⁴ and the CDC’s current BLL reference value of $3.5\mathrm{\mu }\mathrm{g}/\text{dL}$ for children under 5 y of age.¹⁶ We also examined statistical interactions using two-way product terms of HUD housing assistance status and program type and HUD assistance status and race/ethnicity in separate models. The statistical significance of interaction terms was set at a two-sided alpha of 0.05. Results with samples sizes below 5 were suppressed to ensure confidentiality.

Sensitivity Analyses

In sensitivity analyses, we compared characteristics of participants with and without BLLs. We also evaluated the robustness of our findings under several scenarios: the use of a longer, 4-y waiting list duration to create our pseudo-waitlist recipient group; restricting our sample to recipients living in urban areas; and restricting our sample to renters. Although most HUD recipients were in renter-based housing programs, a few were in HCV homeownership programs.³² In addition, data on NHANES child participants who entered a HUD assistance program after their 18th birthday were not accessible in the linked dataset.⁴⁷ Pseudo-waitlist participants in the 2017–2018 NHANES survey who received HUD assistance after 31 December 2019 (i.e., the latest date in the linked HUD administrative files) would also not be captured in our sample. These two data limitations may have underestimated the number of pseudo-waitlist recipients in our study sample and contributed to its lower sample size in comparison with the current recipient group. As a sensitivity analysis, we excluded children in NHANES who were 16–18 y of age at the time of the interview and all participants in the 2017–2018 NHANES survey wave. We also tested for effect modification of HUD assistance and BLLs by age.

Results

The distribution of sociodemographic characteristics and BLLs for HUD recipients by timing of HUD housing assistance are shown in Table 1. Overall, HUD recipients were primarily low-income (i.e., within 200% of FIPR), unemployed, without a college degree, and living in urban areas. In comparison with pseudo-waitlist recipients, current recipients had similar distributions of program assistance type and most sociodemographic characteristics (e.g., sex, race/ethnicity, education, country of birth). However, age, FIPR, household size, partnership status, and urbanicity status differed across the two groups ($p<\mathrm{0}.\mathrm{00}1$–0.031), with current recipients having a larger percentage of older residents age $62+$ y (20.1% vs. 11.4%) and those living below poverty (65% vs. 54%). The unadjusted GMs for BLL for current and pseudo-waitlist recipients were $1.\mathrm{0}4\mathrm{\mu }\mathrm{g}/\text{dL}$ (GSE: 1.03) and $1.32\mathrm{\mu }\mathrm{g}/\text{dL}$ (GSE: 1.03), respectively. Approximately 4.9% ($n=15\mathrm{0}$) of current recipients had BLL $\ge 3.5\mathrm{\mu }g/\text{dL}$ in comparison with 9.2% ($n=\mathrm{114}$) of pseudo-waitlist recipients (Table 1). The distribution of unadjusted GMs for BLL by sociodemographic characteristics can be found in Table S1.

Table 1.

BLL and sociodemographic characteristics by timing of HUD housing assistance, NHANES-HUD linked data 1999–2018 ($n=\mathrm{4,3}\mathrm{0}6$). Unless otherwise noted, values are shown for percentage (SE).

	Total ($n=\mathrm{4,3}\mathrm{0}6$)	Currenta ($n=3,\mathrm{071}$)	Pseudo-waitlistb ($n=\mathrm{1,235}$)
	Count	% (SE)	% (SE)	$p$-Valuec
Geometric mean BLL (GSE) (µg/dL)	—	1.04 (1.03) µg/dL	1.32 (1.03) µg/dL	$<\mathrm{0}.\mathrm{00}1$
BLL $\ge 3.5\mathrm{\mu }\mathrm{g}/\text{dL}$	264	4.9 (0.5)	9.2 (1.3)	0.002
BLL $\ge 5.0\mathrm{\mu }\mathrm{g}/\text{dL}$	111	2.0 (0.3)	4.0 (1.0)	0.030
HUD program type
Housing Choice Voucher	2,401	56.2 (2.5)	48.7 (2.8)	0.113
Public housing	920	19.6 (2.1)	22.8 (2.6)	—
Multifamily, income-restricted	985	24.2 (2.1)	28.4 (2.7)	—
Age categories (y)				$<\mathrm{0}.\mathrm{00}1$
6–11	1,040	16.3 (0.9)	18.9 (1.4)	—
12–19	1,062	15.1 (0.9)	13.7 (1.3)	—
20–40	783	28.1 (1.3)	37.2 (2.3)	—
41–61	633	20.5 (1.0)	19.0 (1.7)	—
$62+$	788	20.1 (1.7)	11.4 (1.5)	—
Sex				0.088
Male	1,673	32.1 (1.1)	35.3 (1.5)	—
Female	2,633	67.9 (1.1)	64.7 (1.5)	—
Race/ethnicity				0.491
Non-Hispanic White	715	31.7 (2.9)	34.8 (3.4)	—
Non-Hispanic Black	2,458	45.7 (2.7)	43.2 (2.9)	—
Mexican American	459	5.9 (0.8)	7.5 (1.2)	—
Another race/ethnic group (e.g., Asian, Other Hispanic, Multiracial)	674	16.7 (2.1)	14.6 (2.6)	—
Federal Income-to-Poverty Ratio (FIPR)d				0.002
$<\mathrm{0}.5\mathrm{0}$	1,016	23.3 (1.5)	21.4 (2.2)	—
0.50–0.99	1,557	42.2 (2.2)	32.5 (2.3)	—
1.0–1.99	1,128	27.5 (1.6)	34.6 (2.7)	—
$2.\mathrm{0}+$	318	7.0 (1.1)	11.5 (1.3)	—
Missing	287	—	—	—
Education of household referenced				0.697
Up to high school	1,760	37.5 (1.8)	40.1 (3.1)	—
High school degree, some college	2,260	57.4 (1.7)	55.4 (3.0)	—
College degree (bachelor’s degree or graduate level)	193	5.2 (0.7)	4.4 (1.0)	—
Missing	93	—	—	—
Household sized				0.031
1–2 persons	1,397	46.5 (2.3)	37.1 (2.9)	—
3–4 persons	1,452	39.4 (2.2)	46.3 (2.5)	—
$5+$ persons	588	14.0 (1.4)	16.6 (1.9)	—
Missing	869	—	—	—
Country of birth				0.297
Outside of United States	595	14.6 (1.5)	12.4 (1.9)	—
United States	3,711	85.4 (1.5)	87.6 (1.9)	—
Partnership statusd				0.003
Single, widowed, or divorced	2,949	74.3 (1.9)	63.9 (2.9)	—
Married or coupled	1,183	25.7 (1.9)	36.1 (2.9)	—
Missing	174	—	—	—
Employment status				0.057
Unemployed	3,552	76.7 (1.2)	72.5 (1.8)	—
Employed	754	23.3 (1.2)	27.5 (1.8)	—
Urban statusd				0.023
Urban	4,016	96.0 (2.0)	91.4 (2.7)	—
Rural	144	4.0 (1.2)	8.6 (2.0)	—
Missing	146	—	—	—

Note: Percentages may not add up to 100% due to rounding. Sociodemographic characteristics are self-reported and refer to the participant unless specified. —, no data; BLL, blood lead level; GM, geometric mean; GSE, geometric SE; HUD, US Department of Housing and Urban Development; NHANES, National Health and Nutrition Examination Survey; SE, standard error.

^a Receiving housing assistance at NHANES interview.

^b Will receive housing assistance within 2 y after NHANES interview.

^c $p$-Values were calculated using the two-sample $t$-test of difference in geometric means or Wald chi-square test of difference in proportions.

^d Percentages reported exclude missing values. A complete case analysis was used for Federal Income-to-Poverty Ratio and Education of household reference.

In fully adjusted models, the LSGM of BLL was significantly lower for current recipients ($1.\mathrm{0}7\mathrm{\mu }\mathrm{g}/\text{dL}$; 95% CI: $1.\mathrm{0}2\mathrm{\mu }\mathrm{g}/\text{dL}$, $1.12\mathrm{\mu }\mathrm{g}/\text{dL}$) than for pseudo-waitlist recipients ($1.21\mathrm{\mu }\mathrm{g}/\text{dL}$; 95% CI: $1.14\mathrm{\mu }\mathrm{g}/\text{dL}$, $1.28\mathrm{\mu }\mathrm{g}/\text{dL}$) (Figure 1), with HUD assistance associated with an 11.4% lower LSGM (95% CI: $-16.4\%$, $-6.1\%$) and an OR of 0.60 (95% CI: 0.42, 0.87) for BLLs $\ge 3.5\mathrm{\mu }\mathrm{g}/\text{dL}$ (Table 2). We also assessed effect modification by program type and race/ethnicity. In stratified analyses, percent differences in LSGM between current and pseudo-waitlist recipients were most pronounced among public housing ($-19.5\%$; 95% CI: $-27.5\%$, $-1\mathrm{0}.7\%$) and multifamily income-restricted ($-12.5\%$; 95% CI: $-2\mathrm{0}.7\%$, $-3.5\%$) recipients in comparison with HCV recipients ($-5.7\%$, $-12.7\%$, 1.7%) (Table 2; Figure 1). However, the two-way interactions of HUD assistance and program type for BLLs were not statistically significant ($\text{global\hspace{0.17em}}{p}_{\text{interaction}}=\mathrm{0}.16\mathrm{0}$) (data not shown). In stratified analyses by race/ethnicity, percent differences in LSGM by HUD housing assistance were strongest among NH White recipients ($-2\mathrm{0}.6\%$; 95% CI: $-29.8\%$, $-1\mathrm{0}.3\%$) but weaker and nonsignificant among NH Black ($-1.6\%$; 95% CI: $-8.1\%$, 5.4%) and Mexican American ($-12.5\%$; 95% CI: $-31.9\%$, 12.5%) recipients (Table 2; Figure 2). Notably, the two-way interactions of HUD assistance and race/ethnicity for BLLs was statistically significant ($\text{global\hspace{0.17em}}{p}_{\text{interaction}}=\mathrm{0}.\mathrm{00}6$), with the pairwise interaction for being a current recipient (Ref: pseudo-waitlist) and NH Black (Ref: NH White) statistically significant ($p=\mathrm{0}\mathrm{.001}$) (data not shown).

Figure 1.

LSGM and 95% CIs of BLLs ($\mathrm{\mu }\mathrm{g}/\text{dL}$) among current and pseudo-waitlist HUD recipients, overall and stratified by HUD program type, NHANES-HUD linked data 1999–2018 ($n=\mathrm{4,3}\mathrm{0}6$). Note: LSGMs are marginal effect estimates based on the multivariable linear regression for BLL taking the form: $\text{log}\left(y_i\right)={\mathrm{\beta }}_0+{\mathrm{\beta }}_{Hlin}H+{\mathrm{\beta }}_{X}X+{ϵ}_i$, where ${ϵ}_i\sim N\left(0,{\mathrm{\delta }}^2\right)$ is a normally distributed error term, $y_i$ is the BLL outcome for individual $i$; H is housing assistance status (1 = current, 0 = pseudo-waitlist); and X is a vector of covariates. Covariate adjustment included age, sex, family income-to-poverty ratio, country of birth, education, race/ethnicity, US Census region (masked), and survey year (linear and quadratic). Sample sizes for HUD program type subgroups are as follows: Public housing ($n=92\mathrm{0}$), Multifamily, income-restricted ($n=985$), Housing Choice Voucher ($n=\mathrm{2,4}\mathrm{0}1$). P is $p$-value comparing LSGM of current versus pseudo-waitlist recipients in adjusted models stratified by HUD program type. Lines on bars indicate 95% CIs. BLL, blood lead level; CI, confidence interval; HUD, US Department of Housing and Urban Development; LSGM, least squares geometric mean; NHANES, National Health and Nutrition Examination Survey; Ref, reference group.

Table 2.

Percent difference in LSGM of BLL, OR of BLL $\ge 3.5\mathrm{\mu }\mathrm{g}/\text{dL}$, and corresponding 95% CIs comparing current vs. pseudo-waitlist HUD recipients overall and for each stratum of (A) program type and (B) race/ethnicity, NHANES-HUD linked data 1999–2018 ($n=\mathrm{4,3}\mathrm{0}6$).

	Timing of HUD assistance: Current vs. pseudo-waitlist (Ref)
	LSGM of BLL		BLL $\ge 3.5\mathrm{\mu }\mathrm{g}/\text{dL}$
	Percent difference (95% CI)a	$p$-Value	OR (95% CI%)b	$p$-Value
Overall: All recipients ($n=\mathrm{4,3}\mathrm{0}6$)	$-11.4$ ($-16.4$, $-6.1$)	0.016	0.60 (0.42, 0.87)	0.007
Strata A. HUD program type
Public housing ($n=92\mathrm{0}$)	$-19.5$ ($-27.5$, $-1\mathrm{0}.7$)	$<\mathrm{0}.\mathrm{00}1$	0.43 (0.23, 0.81)	0.010
Multifamily income-restricted ($n=985$)	$-12.5$ ($-2\mathrm{0}.7$, $-3.5$)	0.009	0.60 (0.34, 1.08)	0.089
Housing Choice Vouchers ($n=\mathrm{2,4}\mathrm{0}1$)	$-5.7$ ($-12.7$, 1.7)	0.126	0.85 (0.52, 1.39)	0.517
Strata B. Major race/ethnicity
Non-Hispanic White ($n=715$)	$-2\mathrm{0}.6$ ($-29.8$, $-1\mathrm{0}.3$)	0.001	0.16 (0.06, 0.42)	$<\mathrm{0}.\mathrm{00}1$
Non-Hispanic Black ($n=\mathrm{2,458}$)	$-1.6$ ($-8.1$, 5.4)	0.637	0.91 (0.56, 1.48)	0.689
Mexican American ($n=459$)	$-12.5$ ($-31.9$, 12.5)	0.273	0.45 (0.07, 3.00)	0.382

Note: —, no data; BLL, Blood lead level; CI, confidence interval; HUD, US Department of Housing and Urban Development; LSGM, least squares geometric mean; NHANES, National Health and Nutrition Examination Survey; OR, odds ratio; Ref, reference group.

^a Percent differences were estimated based on the multivariable linear regression model for BLL taking the form: $\text{log}\left(y_i\right)={\mathrm{\beta }}_0+{\mathrm{\beta }}_{Hlin}H+{\mathrm{\beta }}_{X}X+{ϵ}_i$, where ${ϵ}_i\sim N\left(0,{\mathrm{\delta }}^2\right)$ is a normally distributed error term, $y_i$ is the BLL outcome for individual i; H is housing assistance status (1 = current, 0 = pseudo-waitlist); and X is a vector of covariates.

^b ORs were estimated based on the multivariable logistic regression models for BLL $\ge 3.5\mathrm{\mu }\mathrm{g}/\text{dL}$ taking the form: $\text{logit}\left(p\right)=$ $\text{log}\left[\frac{p}{1-p}\right]={\mathrm{\beta }}_0+{\mathrm{\beta }}_{H\mathit{log}}H+{\mathrm{\beta }}_{X}X$, where $p$ is probability of the BLL outcome $\ge 3.5\mathrm{\mu }\mathrm{g}/\text{dL}$; H is housing assistance status (1 = current, 0 = pseudo-waitlist); and X is a vector of covariates. Covariates for multivariable models included age, sex, family income-to-poverty ratio, education, country of birth, race/ethnicity (if not already stratified), US Census region (masked), and survey year (linear and quadratic).

Figure 2.

LSGM and 95% confidence intervals of BLL ($\mathrm{\mu }\mathrm{g}/\text{dL}$) among current and pseudo-waitlist HUD recipients, overall and stratified by major race/ethnicity categories, NHANES-HUD linked data 1999–2018 ($n=\mathrm{4,3}\mathrm{0}6$). Note: LSGMs are marginal effect estimates based on the multivariable linear regression for BLL taking the form: $\text{log}\left(y_i\right)={\mathrm{\beta }}_0+{\mathrm{\beta }}_{Hlin}H+{\mathrm{\beta }}_{X}X+{ϵ}_i$, where ${ϵ}_i\sim N\left(0,{\mathrm{\delta }}^2\right)$ is a normally distributed error term, $y_i$ is the BLL outcome for individual i; H is housing assistance status (1 = current, 0 = pseudo-waitlist); and X is a vector of covariates. Covariate adjustment included age, sex, family income-to-poverty ratio, country of birth, education, US Census region (masked), and survey year (linear and quadratic). Sample sizes for racial/ethnic subgroups are as follows: Non-Hispanic (NH) White ($n=715$), NH Black ($n=\mathrm{2,458}$), Mexican American ($n=459$); $p$ is the $p$-value comparing LSGM of current versus pseudo-waitlist recipients in adjusted models stratified by race/ethnicity. Lines on bars indicate 95% confidence intervals. BLL, blood lead level, LSGM, least squares geometric mean; HUD, US Department of Housing and Urban Development; NHANES, National Health and Nutrition Examination Survey; Ref, reference group.

Sensitivity Analyses

Sensitivity analysis comparing characteristics of participants with and without BLLs suggested no significant differences in their HUD housing assistance status nor in the distribution of HUD program type and most sociodemographic characteristics ($p>\mathrm{0}.\mathrm{0}5$). Though, participants missing BLLs had a slightly higher percentage of having a college degree (8.2% vs. 4.9%; $p=\mathrm{0}.\mathrm{0}4$) (Table S2). Comparisons of the fully adjusted linear and logistic regression models under various scenarios (i.e., use of a pseudo-waitlist group with a 4-y waiting list duration, exclusion of homeowners, exclusion of rural residents, exclusion of children ages 16–18 y and NHANES 2017–2018 survey) yielded percent differences in LSGM of BLL and ORs of BLL $\ge 3.5\mathrm{\mu }\mathrm{g}/\text{dL}$ for current vs. pseudo-waitlist recipients that were of similar magnitude, direction, and statistical significance as those from the main models (Table S3). There were also no significant interactions between HUD assistance and age ($\text{global\hspace{0.17em}}{p}_{\text{interaction}}=\mathrm{0}.99\mathrm{0}$) (Figure S3).

Discussion

Our study is the first to examine BLLs by HUD housing assistance status in a nationally representative sample of HUD-eligible US population using a quasi-experimental, pseudo-waitlist study design. We found that entry into HUD housing assistance was associated with an 11.4% reduction in LSGM of BLL and an OR of 0.60 for BLL $\ge 3.5\mathrm{\mu }\mathrm{g}/\text{dL}$ among current vs. pseudo-waitlist recipients. Our findings have significant clinical, public health, and health equity implications as they suggest that HUD housing assistance is an important social-structural safety net for very low-income households to access both affordable and safe, healthy housing. However, we also found disparities in the protective association of HUD housing assistance on BLLs by HUD program type and race/ethnicity that warrant further attention in future environmental health research and housing interventions.

Past studies have suggested that federal housing assistance may reduce the risk of lead-based paint exposure²³ and BLLs among children,⁴² but sample size issues, data availability, and ascertainment of comparable groups by HUD assistance status limited these analyses. The 2005–2006 American Healthy Homes Survey study did not specify housing assistance type and relied on self-reported data for assistance status,²³ which is susceptible to misreporting.⁶⁴ The Ahrens et al. study used income thresholds and propensity score matching to select their control group,⁴² which may have included participants not eligible for or who never entered HUD housing assistance. Although our findings were generally consistent with these past studies, we improved on their work by implementing a quasi-experimental technique that isolated the impact of HUD housing assistance by comparing BLLs of current and pseudo-waitlist HUD recipients. This approach limited potential selection biases by HUD eligibility and entry. We also leveraged the nationally linked NHANES-HUD dataset from 1999 to 2018, which included more years of linked data and increased our study’s power to examine effect modification and conduct stratified analyses.

The protective association of HUD housing assistance and BLLs is likely attributed to HUD’s stricter compliance and enforcement of federal residential lead-paint laws in their housing portfolios in comparison with nonassisted housing in the private market.^65–67 These laws include the Lead-Paint Poisoning Prevention Act of 1971 for public housing and the Residential Lead-Based Paint Hazard Reduction Act of 1992 (Section 1012/1013 of Title X), which led to the Lead-Safe Housing Rule enacted in 2000 for all HUD-assisted units built before 1978, including those receiving HCVs. These regulations led to a range of lead hazard control strategies adopted in HUD housing assistance programs, such as visual assessments, inspections, and ongoing maintenance; risk assessment, abatement, and clearance of lead-based paint; and safe work practices during rehabilitation.

However, our study also found variations in the effectiveness of HUD housing assistance on BLLs by program type, which could be due to differences in the types and implementation of lead hazard control strategies across programs, as well as differences in housing type (e.g., renter, family based) and age.^68,69 In comparison with HCV and multifamily income-restricted units, HUD enforces more stringent lead control strategies in public housing units, which provide more long-term/permanent solutions to reduce lead exposure, such as requiring lead-based paint inspections and lead abatement for affected units. The Housing Act of 1949 also promulgated the new construction of public housing developments and the removal of old, deteriorating public housing in urban areas that likely had high lead-based paint hazards.⁶⁷ In contrast, tenant-based rental assistance programs like HCVs only require that HCV-eligible units undergo a visual assessment and lead-paint stabilization if lead is found, which are considered short-term/interim controls and thus less effective. In context of other studies, the stronger association of HUD housing assistance and BLLs that we observed among public housing recipients is consistent with much of the affordable housing literature documenting positive health outcomes for public housing recipients in comparison with those in other program types.^35,37,38,43

In addition to differences in lead control strategies, the null association of HUD housing assistance on BLLs among HCV recipients may also be due to the wide variability in the quality of affordable housing options and neighborhood conditions in the private housing market. As HUD’s largest assistance program, the aim of the HCV program is to provide tenants access to higher-opportunity and lower-poverty neighborhoods. The randomized, multisite Moving To Opportunity (MTO) study of families with children in high-poverty areas found that those who used their HCV subsidies to relocate to lower-poverty neighborhoods (i.e., experimental group) reported higher unit satisfaction than families that remained in their project-based housing (i.e., control group) (the study did not measure lead exposure).⁷⁰ However, many families in the MTO study did not move (53% in the MTO experimental group) or they relocated to similar high-poverty neighborhoods.^70,71 McClure and Johnson⁷² found that HCVs generally did not improve residents’ neighborhood quality, though Fenelon et al.⁷³ found that it did for non-Hispanic Black and Hispanic/Latino/Latina children. The limited mobility of HCV recipients to move into higher-opportunity neighborhoods can be attributed to compounding systemic challenges, such as the limited affordable housing supply, discrimination of HCV holders by landlords and oftentimes by suburban and high-opportunity communities, and programmatic inefficiencies such as short voucher expiration periods and long waiting lists.^71,74 Approximately 30% of households that received a voucher could not secure a unit by the time their voucher expired (i.e., usually 60–120 d).⁷¹ Only 12 states currently prohibit HCV discrimination by landlords. In addition, HCV recipients face bureaucratic hurdles if they request to move outside their public housing authority jurisdiction.⁷¹ Families may also be hesitant to move because of trade-offs in loss of social networks, transportation access, and affordability.^70,75–77

Furthermore, in competitive housing markets, rents in high-opportunity areas may exceed the local Fair Market Rent payment standards,⁷⁴ and in turn tenants are left to pay out of pocket to cover rent differences and/or make trade-offs in housing conditions, size, or quality. As a result, the available affordable housing stock in competitive housing markets is likely older and of poorer quality, with greater risk for lead hazards. The MTO study found that HCV households were more likely to double up with friends/relatives to save on costs in comparison with households in project-based housing (which has unit size restrictions).⁷⁰ Crowding at the neighborhood-level has also been associated with higher BLLs among children.⁷⁸ Last, changes in HUD’s HCV program over the study period (i.e., 1999–2018), such as changes to the rent ceiling, fair market rent values for voucher limits, and the availability of voucher subsidies across public housing agencies,^79,80 could inject noise in our analyses that attenuated the observed association between current HUD housing assistance and BLLs among HCV recipients.

Our study also found significant heterogeneity in the association of HUD housing assistance and BLLs by race/ethnicity. After adjustments for demographic and socioeconomic measures, the association of HUD housing assistance and BLLs was weaker for NH Black and Mexican American recipients in comparison with NH Whites, which could be due to their exposure to other sources of lead unrelated to HUD housing, such as lead-contaminated drinking water, diet, occupational exposures, ceramics, jewelry, folk remedies, and living in proximity to mining, smelting, and Superfund sites.⁸¹ Lanphear et al.’s study of 205 urban children in Rochester, New York,⁸² found that water lead levels were associated with elevated BLLs among Black children but not White children. Although many of these sources are correlated with socioeconomic status, which we tried to account for in our analysis, residual confounding could persist.

Moreover, Black households continue to face significant barriers to high-quality housing and high-opportunity neighborhoods that may have fewer lead hazards because of legacies of racist housing policies and urban planning practices in the United States. These practices include redlining, zoning and land use restrictions, gerrymandering of school and census boundaries, predatory lending, and urban renewal initiatives in Black and Brown neighborhoods that displaced families and built highways, airports, and other large pollution-emitting sources in their neighborhoods through eminent domain.^24–27 Black families receiving vouchers tend to live in more disadvantaged, racially segregated, and overcrowded neighborhoods in comparison with White families receiving vouchers.^71,74 They are also more likely to live in older housing units with more frequent housing code violations.⁸³ In addition, despite the Fair Housing Act of 1968, Black, Asian, and Latino/Latina renters continue to face discrimination in their housing searches.²⁷ Notably, Black home-seekers are more likely than White home-seekers to be shown units with housing quality problems.²⁷ As such, it is crucial that future research account for measures of structural racism as well as nonhousing-related sources of lead that may explain persistent Black–White disparities in BLLs.

Our study has several limitations. Given the cross-sectional nature of the NHANES data, we were unable to follow HUD recipients over time and observe changes in their BLLs. Second, because HUD assistance was not a random assignment, it is possible that our observed association may be driven by correlated underlying factors like housing policy changes, though we did account for survey year in all analyses. Third, although most of our study population resided in urban areas and we adjusted for US Census regions in our analyses, there may be residual geographical differences that we did not control for, such as state of residence or public housing authorities. Prior works suggest that the quality of HUD-assisted housing and enforcement of lead regulations vary by region and municipality.^43,74 However, our study was designed to be nationally representative, and state-specific estimates could not be generated due to the NHANES study design and potential disclosure risks. In addition, a previous study of HUD housing assistance in the National Health Interview Survey (NHIS) found that accounting for state residence did not influence study conclusions.³⁷ Fourth, because our study period was 1999–2018 and occurred before the COVID-19 pandemic, findings may not be generalizable to HUD programs and housing needs during the pandemic. Fifth, $\sim \mathrm{16}\%$ of our current and pseudo-waitlist HUD recipients were missing BLL data, which were not accounted for in the NHANES-HUD linkage-eligible adjusted MEC weights used in our analyses. The majority of those missing were from survey waves 2013–2014 and 2015–2016 when BLLs were measured in only one-half sample from participants 12 y of age and older. Sixth, we did not have access to information about duration of HUD assistance, which could influence the concentrations of blood lead measured. Blood lead has a half-life of $\sim \mathrm{1}-\mathrm{2}$ months.⁵³ Last, NHANES is conducted in English and Spanish, and findings may not be generalizable to US populations that speak other languages.

Our study also had several strengths. First, we applied a quasi-experimental pseudo-waitlist approach that allowed us to identify more comparable groups (i.e., current and pseudo-waitlist recipients) and isolate entry into HUD housing assistance by minimizing potential selection biases based on HUD eligibility. To do this, we leveraged the NHANES-HUD linked data from 1999 to 2018 to access a large, nationally representative sample of US HUD housing assistance recipients. Our study is also the first to assess effect modification of HUD housing assistance and BLLs by HUD’s three major housing assistance program types and by race/ethnicity. In addition, HUD administrative records and NHANES BLL data were collected independently and for unique purposes, which minimized the risk of selection bias and nondifferential exposure misclassification. Data on HUD housing assistance and program type were also obtained from HUD administrative records, which minimized participant recall bias.³² Last, BLL measurements were collected and analyzed by trained NHANES staff using standardized and data quality assurance protocols.

Our findings contribute to the broader literature on federal housing assistance and health and provide a foundation for future opportunities to investigate the role of BLL reduction in mediating associations between housing assistance and positive health outcomes across the lifecourse.^36–40 For example, Fenelon et al.’s³⁷ analysis of the 1999–2012 NHIS-HUD linked data found that children in public housing reported better mental health outcomes and lower risk of psychological problems and emotional difficulties, which could be related to lead’s influence on children’s neurocognitive, behavioral, and emotional development.^4,7,84 For adults, using the same 1999–2012 NHIS-HUD linked data, Fenelon et al.³⁸ found that residence in public and multifamily housing was associated with better self-reported health and lower psychological distress, which could be related to lead’s cardiovascular health effects, like hypertension, coronary heart disease, and stroke.^1,85

In conclusion, reducing lead exposure in the US population, and particularly among high-risk groups, should continue to be a national public health priority. Our research demonstrates that federal policies that invest in affordable and stable housing for very low-income families can also be a means for improving health and health equity across the life course through safer living environments. Housing assistance programs may reduce residents’ exposure to lead and other environmental hazards by providing better-quality housing, routine maintenance practices, compliance with residential lead-paint laws, and/or access to higher-opportunity neighborhoods. In addition, our study expands the “healthy homes” paradigm beyond public housing to include HUD’s other major housing programs such as multifamily income-restricted units and tenant-based HCVs. It provides some of the first systematic comparisons of residential environmental hazards like lead by HUD program type. Our null findings for tenant-based HCV holders suggest that HUD may need to evaluate its inspection and de-leading processes for private market housing to better avoid or reduce lead exposures for HCV holders. Although the number of affordable housing units managed by HUD has increased since 1997, the proportion of public housing and multifamily income-restricted units (i.e., project-based) available is far less than HCVs.⁸⁶ The protective association of HUD assistance among project-based housing suggest that increasing newly constructed or renovated project-based housing portfolios may help to reduce residential environmental hazards like lead. Environmental hazards could also be mitigated by investing in improvements necessitated by years of deferred maintenance in public housing.^45,46

Furthermore, our findings underscore the need to better elucidate mechanisms driving racial disparities in the association of federal housing assistance on BLLs, even among a very low-income population. Black NH households comprise the largest proportion of HUD-assisted households and thus have the most to gain from improvements in federal housing policies and programs. Specifically, future work could address racial equity in access to and the administration of these programs, as well as other potential residential sources of lead exposure, to ensure that safe and affordable housing options are accessible by all residents.

Conclusions and opinions are those of the individual authors and do not necessarily reflect the policies or views of EHP Publishing or the National Institute of Environmental Health Sciences.