Lead Below the CDC's Threshold Tied to Teen Depression

Lead came out of house paint in 1978 and out of American gasoline by 1996, and for most parents that is where the story ends. But the child living in a 1940s rental with a chipped window sill, or in a house whose renovation kicked dust into the hallway, still carries a measurable amount of it. The number on the lab report is usually small enough that nobody calls back.

A study published January 28, 2026 in JAMA Network Open asks what those small numbers do to a child's mood years later. Christian Hoover, a doctoral student at Brown's School of Public Health, and colleagues at Brown and Cincinnati Children's Hospital drew on the Health Outcomes and Measures of Environment Study, or HOME, which recruited pregnant women in Cincinnati between 2003 and 2006 — deliberately from families living in pre-1978 housing, to capture children at higher exposure risk — and followed their children from the second trimester through age 12.

Blood lead was drawn repeatedly: from cord blood at birth, then at ages 1, 2, 3, 4, 5, 8 and 12. At the age-12 visit, held between 2016 and 2019, children and caregivers completed the 160-item BASC-3 in both child- and caregiver-report forms, and the children alone completed the 28-item Children's Depression Inventory-II and the 41-item SCARED anxiety inventory. The analytic sample was 218 caregiver-child pairs: 55.5% girls, 35.8% Black, mean age 12.4, with 76.3% of mothers having at least some college. Of the 339 children with at least one lead measurement, only these 218 completed the full mental health assessment.

The exposure levels are the part worth pausing on. The median within-child average was 9.6 micrograms per liter, with an interquartile range of 7.8 to 12.6 and a full range of 4.8 to 32.4 — that is 0.96 micrograms per deciliter in the units most American labs print. The CDC's reference value, the 97.5th percentile for US children in 2015 to 2018, is 35 micrograms per liter. Not one child in this cohort reached it.

Thirty-four of the 218 children — 15.6% — scored in the at-risk range on the child-reported BASC-3 depression scale, meaning a T-score of 60 or above. Against that base rate, each doubling of a child's mean childhood blood lead was associated with a relative risk of 1.90 for landing in that range by the child's own report (95% CI, 1.00 to 3.66; P = .05). When child and caregiver reports were combined, the estimate was 1.76 (95% CI, 1.12 to 2.78; P = .02). Caregivers reporting alone saw nothing statistically distinguishable: 1.49 (95% CI, 0.77 to 2.89; P = .24).

Anxiety mostly did not move. Child-reported BASC-3 anxiety came in at 1.14 (95% CI, 0.59 to 2.21) and the SCARED total at 1.10 (95% CI, 0.73 to 1.67). The one exception cuts against the depression result rather than with it: caregivers reporting on their children's anxiety produced the largest single estimate in the paper, 2.09 (95% CI, 0.96 to 4.55; P = .06) — marginal, interval crossing 1.00, and from the informant whose depression report showed nothing. One of SCARED's five subscales — significant school avoidance — did reach significance at 1.64 (95% CI, 1.07 to 2.50; P = .02), a single positive among five looks that the paper does not flag as exploratory.

The timing analysis is what generated the headlines. A multiple-informants model, which uses the repeated blood draws to estimate each age's association jointly and test whether they differ, put the adjusted difference in child-reported depression score per standard-deviation increase in log-transformed lead at 1.82 at age 1 (95% CI, −1.10 to 4.74), 3.22 at age 8 (95% CI, 0.53 to 5.90) and 2.21 at age 12 (95% CI, 0.19 to 4.24). The authors' conclusion describes "particularly large increases in risk for exposures occurring in late childhood and early adolescence."

Read as a sequence, those three estimates tell a story: the association is quiet in toddlerhood and loud at 8. Read as intervals, the story softens considerably. The age-1 interval runs from −1.10 to 4.74; the age-8 interval runs from 0.53 to 5.90. They overlap across most of their length. What separates them is that one crosses zero and the other does not — a difference in statistical significance, which is not the same as a significant difference. The paper does not leave this to inference. It tested directly whether the association differed by exposure window, using a deliberately permissive threshold of P ≤ .15 as evidence of a difference, and got P = .33 for depression. By the authors' own test, the age-8 peak is not distinguishable from the rest.

There is also a sequencing problem at the far end. The age-12 blood draw happened at the same visit as the symptom questionnaires, so that estimate cannot establish which came first. A twelve-year-old who is withdrawn, sleeping badly, eating differently or spending more time at home is a twelve-year-old whose exposure profile may be shifting for reasons downstream of mood rather than upstream of it. And because blood lead declines with age, some values were imputed under an assumed linear decline — a reasonable modeling choice the authors name as a limitation, but one that puts a statistical assumption inside the exact variable the timing claim rests on.

The finding that carried the coverage came from the BASC-3, a broad-band behavior inventory, at exactly P = .05, with a lower confidence bound of 1.00. The CDI-II — a questionnaire built to do nothing but measure depressive symptoms in children — was administered to 217 of the same children on the same day and returned a relative risk of 1.19 (95% CI, 0.47 to 2.98; P = .71). On the CDI-II, 20 of 217 children reached the clinically significant cutoff. Two instruments, one cohort, opposite verdicts. That does not make the BASC-3 result wrong, but it means the signal is not robust to the choice of ruler.

Then there is the question of what lead is standing in for. HOME recruited families in older housing on purpose, and older, cheaper housing correlates with a great deal besides lead: neighborhood disadvantage, residential instability, financial strain — each an established risk for adolescent depression on its own. The reported adjustment set centers on maternal education, marital status, race and ethnicity, child age and sex, caregiver mental health and caregiver-child relationship quality. Household income does not appear in it. The authors note that dropping caregiver mental health and relationship quality made the associations larger and more precise, which argues those two are not manufacturing the effect; it does not address deprivation the model never measured.

Lead's effect on the developing brain is not in dispute — the cognitive and externalizing-behavior literature is decades deep, and no threshold below which it is harmless has ever been established. What has been thin is the internalizing side: depression and anxiety, which mostly announce themselves in adolescence, long after the exposure and long after anyone stopped looking. The authors' proposed mechanism — altered neurotransmitter function, reduced neurogenesis, disrupted synaptic plasticity in mood-regulating regions — is biologically plausible and entirely unmeasured here.

The practical inference is narrower than the headline and more useful. A blood lead result under the CDC reference value means a child is not in the top 2.5% of American children; it does not mean the exposure is zero, and this cohort's entire range sat under that line. If there is something to the late-childhood signal, the relevant exposure is not a distant infancy event but the house the child is living in now — which is, at least, the kind of thing that can still be changed.

Coverage of the paper generally reported that childhood lead exposure increases the risk of depressive symptoms. What the design supports is an association, in 218 children in one city, detected by one depression instrument and not the other, with a P value sitting on the line.

This is a careful study with a thin result, and the gap between those two things is where the coverage went wrong. HOME is a genuinely valuable cohort — twelve years of serial blood draws is rare and expensive — and the question it asks, whether below-reference lead shows up in adolescent mood rather than in IQ scores, is one somebody needed to ask. But the headline association has a confidence interval whose lower bound is 1.00, it appears on one depression instrument and vanishes on the other, and the much-repeated age-8 window failed the paper's own test for a difference across exposure windows at P = .33. This is best understood as a serious hypothesis that has earned a larger cohort and a second look, not as a finding about any particular child. The one durable takeaway needs no statistics at all: a lead result below the CDC's reference value is not the same as no exposure, and this cohort's entire range sat beneath that line.