All prevalence outcomes were log-transformed to reduce skew and normalize distributions. We used simple and multivariable linear regression to assess associations between log-prevalence of neurodevelopmental disorders and lead exposure, adjusted for log-transformed population density. Though we initially included income and psychiatric institution density in exploratory models to assess their independent contributions, neither variable was statistically significant in the overall model (both p > 0.80), and their inclusion did not materially change the effect estimate for lead. Inclusion of income and psychiatric density resulted in a negligible (< 1%) change in the lead coefficient (−0.988 to −0.981). Nested model comparisons via ANOVA confirmed that these variables did not significantly improve model fit (income: F = 3.63, p = 0.04; psychiatric density: F = 3.65, p = 0.04; both: F = 2.56, p > 0.08 versus base model: F = 5.75, p = 0.01). Thus, to preserve statistical power and avoid overfitting in this small-sample (n = 19 counties) ecological analysis, both variables were excluded from the final model.

To identify influential observations, we calculated standardized residuals, Cook’s distance, and leverage values. Telemark county was identified as an outlier with a standardized residual of 2.64, exceeding the 2.5 threshold. While Oslo and Vest-Agder exhibited high leverage, their low residuals (both < | 0.9|) and low Cook’s distance (< 0.2) confirmed they were consistent with the overall model trend. Visual inspection of added-variable plots confirmed Telemark county’s deviation (see Supplementary Figure 2).

We conducted a series of sensitivity analyses. First, Telemark county was identified as a potential outlier based on the above analysis and excluded. Second, ASD prevalence of children aged 2–8 years (born between 2006 and 2014) from a follow-up study by Surén et al. (2019), was extracted, and analyses conducted with dental lead and population density covariates. This allowed us to assess whether observed associations between lead and ASD persisted in a temporally distinct cohort, despite the lag between exposure and outcome measurement. Finally, we conducted sensitivity analyses to assess whether variation in sample representativeness affected our results. The representativeness of deciduous teeth varied substantially across counties (0.02%–0.31% of the population). To evaluate whether counties with smaller samples (and potentially less precise lead measurements) influenced findings, we first examined the correlation between sampling proportion and the geometric standard deviation of dental lead within each county; no significant correlation was observed (r = −0.13, p = 0.59), suggesting measurement precision was not systematically related to sample size but more likely intra-county variance. We then scaled the county-level geometric mean dental lead concentration by one geometric standard deviation in each direction to simulate scenarios where exposure was higher or lower than measured—effectively testing whether results were sensitive to measurement error that might be more pronounced in poorly sampled counties. These analyses demonstrated that the observed associations remained significant across both scenarios, indicating that findings were robust to potential measurement variability and were not driven by extreme values from counties with fewer samples or wider standard deviations. We also conducted correlation analysis of regression residuals against teeth representativeness and found no significant correlation (r = 0.49, p = 0.26).

No missing data were present for any variables included in the primary analysis. Variance inflation factors (VIFs) were calculated to assess multicollinearity and were below 2.0 in all models. All regression analyses were conducted with SAS^® OnDemand for Academics. Given the ecological design and small sample size, findings should be interpreted with caution and viewed as hypothesis-generating.

This analysis used publicly available, aggregated data and did not require institutional review board approval.

Norway, much like the rest of the developed world, has experienced a steep decline in lead exposure due to the phase-out of leaded gasoline as well as other successful environmental remediation efforts (Tvinnereim et al., 1997). For instance, mean lead levels in deciduous teeth in Norway fell from 2.92 μg/g in the 1970s to 1.27 μg/g in the early 1990s (Tvinnereim et al., 1997). Simultaneously, ASD prevalence in Norway has increased by an order of magnitude from children born in the 1970s to those born in the early 2000s (Surén et al., 2013).

Global trends in reduced lead exposure, particularly with respect to the phase-out of leaded gasoline, aligns with worldwide variations in ASD prevalence. Nations which had banned leaded gasoline early, such as the US, Japan, Canada, South Korea, Sweden, and the UK, reported ASD prevalence exceeding 100 per 10,000 by the early 2010s (Lovei, 1998; Zeidan et al., 2022). Conversely, a culturally diverse group of nations with historically higher lead exposure due to delayed bans or ongoing industrial pollution have tended to report lower ASD prevalence: for example, Bangladesh, Ecuador, Israel, Oman, and Venezuela (Zeidan et al., 2022). While these geographic associations are intriguing, they should be interpreted cautiously given the potential for diagnostic, cultural, and health system differences.

Previous research suggests substantial portions of historical increases in ASD were due to diagnostic shifts; for instance, one-third of the 1980–1991 prevalence increase in Denmark (Hansen et al., 2015). However, these factors do not explain prevalence increases during periods of stable diagnostic criteria nor do they explain regional differences in ASD prevalence, such as observed in the study by Suren, from which we sourced our data (Surén et al., 2013). In the present study, neither socioeconomic status nor access to psychiatric facilities played a determining role in ASD risk. The only factors of consequence were environmental.

Although population density was moderately correlated with dental lead levels (r = 0.46), its effects on ASD prevalence diverged in multivariable models. Lead was inversely associated with ASD prevalence, while population density showed a positive association—statistically significant when the outlier county of Telemark was excluded. This suggests that population density is not simply a proxy for lead but rather likely reflects a distinct set of urban-related vehicle emission exposures that may independently increase ASD risk. The association between population density and ASD only emerged after adjusting for lead levels and with incorporation of the later birth cohort, suggesting that lead exposure may suppress the apparent effect of urban-related emissions exposures. This lends further support to the hypothesis that declining lead exposure may have unmasked the harmful effects of vehicle emissions associated with urbanicity on ASD risk.

A recent systematic review of 37 studies found that more than half reported positive associations between lead exposure and ASD, while none reported an inverse association (Wang et al., 2019). However, these studies relied on postnatal biological matrices—such as hair, blood, and urine—collected from children after diagnosis. Similarly, primary teeth have also been employed to compare metal levels, but results have been inconsistent, ranging from null findings (Abdullah et al., 2012; Adams et al., 2007) to reports of higher fetal and postnatal dental lead in ASD twins compared to their non-ASD siblings (Arora et al., 2017).

Importantly, because these studies inferred lead exposure using matrices collected from children, they may be susceptible to reverse causation. Children with ASD often exhibit altered dental enamel formation, marked by stress-induced accentuated lines (Kurek et al., 2020), and have been reported to have lower bone mineral density (Ekhlaspour et al., 2016). Since bone is a critical repository of lead (Bjørklund et al., 2020), a reduction in bone mass can increase the relative percentage weight of sequestered lead within the skeletal and dental matrix.

Furthermore, decreased bone density may facilitate the redistribution of stored lead back into the bloodstream and peripheral tissues like hair. Such physiological differences suggest that elevated postnatal biomarkers may reflect altered mineral homeostasis and storage dynamics inherent to the disorder rather than the primary etiologic exposure. These systemic alterations likely affect the measured concentrations of not only lead but also other metals, including essential minerals. In contrast, the present study estimates exposure using deciduous teeth collected from children born well before the ASD cohorts under study, reducing the potential for bias arising from physiological alterations inherent to the multi-systemic nature of ASD (Kohane et al., 2012).

While clinical studies often focus on the child’s internal lead burden, our results imply that lead’s primary influence on ASD prevalence may occur via competitive interactions at the maternal-fetal interface. Crucially, this mechanism would be entirely invisible to clinical studies utilizing child biomarkers (e.g., blood or hair), as those matrices reflect the child’s own lead accumulation. This divergence in methodology—external environmental proxies versus internal child biomarkers—is likely the primary reason our findings differ from the existing clinical literature.

To date, few studies have examined maternal lead levels and subsequent ASD risk, largely due to the substantial cohort sizes required to observe sufficient outcomes. To our knowledge, this is the first study that has examined fetal exposure to lead using an external proxy. Interestingly, a recent Norwegian case-control study found lower, though not statistically significant, maternal lead levels among ASD cases compared to controls (Skogheim et al., 2021). However, the authors did not adjust for population density or urbanicity, factors that the present study suggests may confound or modify the relationship between lead exposure and ASD prevalence.

Metal dynamics are key to understanding the putative role of lead in modifying risk of ASD. We have previously suggested that maternal intake of iron may reduce ASD risk not by correcting deficiency per se, but by competitively inhibiting absorption of other metals with neurotoxic potential (Grossman, 2017). Iron is known to reduce intestinal absorption and mitochondrial uptake of metals such as manganese and zinc (Arnich et al., 2004; Jouihan et al., 2008; Thompson et al., 2006; Whittaker, 1998). Lead may behave similarly within shared metal-transport pathways. For example, experimental studies demonstrate that lead exposure alters zinc homeostasis, increasing urinary excretion and reducing tissue concentrations in bone and brain (Victery, 1987). The divalent metal transporter-1 (DMT-1), which mediates intestinal absorption of iron and other divalent metals, has been shown to transport lead and is expressed in both placenta and mammary tissue (Bannon et al., 2002; Georgieff et al., 2000). Importantly, DMT-1 activity is dynamically regulated by systemic metal levels, allowing competitive inhibition and hierarchical transport among metals (Meltzer et al., 2010; Shawki et al., 2015).

Within this framework, higher maternal or environmental lead exposure may reduce fetal or early-life bioavailability of other metals implicated in neurodevelopmental toxicity by competing for shared transport mechanisms. At the population level, this could manifest as a suppressor effect, whereby counties with historically higher lead exposure exhibit attenuated associations between urban-related exposures—proxied here by population density—and ASD prevalence. As environmental lead levels declined nationally, this competitive buffering may have weakened, allowing associations between population density–related exposures and ASD prevalence to emerge more clearly in later birth cohorts.

This proposed mechanism is consistent with the ecological and temporal patterns observed in the present study but should be interpreted cautiously. It highlights the possibility that metal–metal interactions and shared transport pathways may influence population-level susceptibility to neurodevelopmental risk factors, warranting further investigation in mechanistically focused and prospective studies.

The absence of an inverse association for ADHD and CP suggests a degree of specificity in the etiologic pathway of ASD, rather than a generalized neurodevelopmental effect. However, this interpretation warrants caution for two reasons. First, in Norway, ascertainment of CP is high, with 89% of affected children registered by 8 years (Andersen et al., 2020). Given this high level of diagnostic coverage, the null result for CP is likely robust and consistent with the understanding that CP is primarily driven by distinct mechanical or vascular insults (Patel et al., 2024). In contrast, it is estimated that only half of children with ADHD are diagnosed within typical primary care, which may limit the statistical power to detect associations for this specific outcome and increase the risk of Type II error (Duric and Elgen, 2011). ASD diagnoses fall between these two extremes; they are generally ascertained at an early age, with a survey of Norwegian clinics reporting that all children were diagnosed by 71 months (Larsen, 2015). Second, there remains the possibility that to inhibit ADHD via a similar mechanism as ASD, a higher dose of lead is required to mitigate exposure to neurotoxicants causative of the condition. This suggests that while the CP null result is a reliable indicator of etiologic specificity, the ADHD result should be interpreted with caution, as both under-ascertainment and potential differences in dose-sensitivity may mask a similar underlying mechanism. As noted above, the county with the highest level of dental lead (Vest-Agder) also had the lowest ASD and ADHD prevalences.

There are several limitations to this study. First, it is subject to ecological fallacy. Our findings describe associations at the aggregate county level, and we cannot directly infer that individual children with higher lead exposure have a lower risk of ASD; any inference regarding individual risk requires further verification through longitudinal cohort studies. While we have attempted to mitigate this risk by adjusting for population-level variables—such as income and psychiatric service availability—this does not eliminate the possibility of unmeasured confounding or the risk of misattributing group-level trends to individuals. Consequently, these conclusions are strictly applicable to the county-level population in Norway and should be viewed as hypothesis-generating. Second, the sample of 19 counties reduces statistical power and limits the generalizability of the findings to different geographical or temporal contexts. While a significant association was detected, the limited number of counties increases risk of effect size inflation. Third, there was a time lag of up to 15 years between dental lead levels and the birth of the cohort of interest, which may have permitted relative levels of lead exposure across counties to have changed. Finally, though we suggest that increased exposure to pollutants may ultimately enhance risk of ASD, a full exploration is outside the scope of this study. These results should, therefore, be interpreted as hypothesis-generating rather than as evidence of a definitive causal or individual-level relationship.

These findings raise the possibility that declines in environmental lead may have inadvertently unmasked or exacerbated other ASD-related risk factors. While these results do not call for a reversal of lead remediation efforts, they underscore the need for a shift from single-metal thresholds toward comprehensive metal-mixture monitoring—specifically incorporating potential suppressors like lead or iron into risk assessments of potential toxicants.

Furthermore, due to differential development trajectories and the risk of reverse causation in child biomarkers, policy should prioritize the monitoring of maternal and prenatal exposure profiles. Because biological testing for certain essential metals can be inadequate (O’Neal and Zheng, 2015), proxy measures of environmental exposure—such as population density for vehicle emissions—should be integrated into public health surveillance, particularly when other forms of pollution monitoring—such as PM_2.5—are lacking. Finally, our results suggest that ecological analyses provide unique value in capturing broad environmental interactions that would be infeasible to track with the same geographic scope at the individual level. Such a scope is essential when studying lead exposure, as localized studies often lack the necessary variance in exposure levels to detect meaningful associations.