Neurobehavioral Deficits, Diseases, and Associated Costs of Exposure to Endocrine-Disrupting Chemicals in the European Union

Abstract

Context:

Epidemiological studies and animal models demonstrate that endocrine-disrupting chemicals (EDCs) contribute to cognitive deficits and neurodevelopmental disabilities.

Objective:

The objective was to estimate neurodevelopmental disability and associated costs that can be reasonably attributed to EDC exposure in the European Union.

Design:

An expert panel applied a weight-of-evidence characterization adapted from the Intergovernmental Panel on Climate Change. Exposure-response relationships and reference levels were evaluated for relevant EDCs, and biomarker data were organized from peer-reviewed studies to represent European exposure and approximate burden of disease. Cost estimation as of 2010 utilized lifetime economic productivity estimates, lifetime cost estimates for autism spectrum disorder, and annual costs for attention-deficit hyperactivity disorder.

Setting, Patients and Participants, and Intervention:

Cost estimation was carried out from a societal perspective, ie, including direct costs (eg, treatment costs) and indirect costs such as productivity loss.

Results:

The panel identified a 70–100% probability that polybrominated diphenyl ether and organophosphate exposures contribute to IQ loss in the European population. Polybrominated diphenyl ether exposures were associated with 873 000 (sensitivity analysis, 148 000 to 2.02 million) lost IQ points and 3290 (sensitivity analysis, 3290 to 8080) cases of intellectual disability, at costs of €9.59 billion (sensitivity analysis, €1.58 billion to €22.4 billion). Organophosphate exposures were associated with 13.0 million (sensitivity analysis, 4.24 million to 17.1 million) lost IQ points and 59 300 (sensitivity analysis, 16 500 to 84 400) cases of intellectual disability, at costs of €146 billion (sensitivity analysis, €46.8 billion to €194 billion). Autism spectrum disorder causation by multiple EDCs was assigned a 20–39% probability, with 316 (sensitivity analysis, 126–631) attributable cases at a cost of €199 million (sensitivity analysis, €79.7 million to €399 million). Attention-deficit hyperactivity disorder causation by multiple EDCs was assigned a 20–69% probability, with 19 300 to 31 200 attributable cases at a cost of €1.21 billion to €2.86 billion.

Conclusions:

EDC exposures in Europe contribute substantially to neurobehavioral deficits and disease, with a high probability of >€150 billion costs/year. These results emphasize the advantages of controlling EDC exposure.

The central nervous system is uniquely sensitive to adverse effects of chemical exposures during early (especially fetal, but through pubertal also) development (1, 2), and endocrine disruption has emerged as an important mechanism by which chemicals may have adverse effects on the developing brain, whether by interfering with thyroid hormone or sex steroid actions, or via other hormonal modes of action (3).

Thyroid hormone is particularly important for normal brain development, and both clinical and animal research provides confidence in the assertion that thyroid disruption will affect brain development (4–7). Predictable outcomes of thyroid disruption include global IQ deficits and neurodevelopment disabilities such as autism spectrum disorder (ASD) and attention-deficit hyperactivity disorder (ADHD) (7–12). Classes of chemicals such as polychlorinated biphenyls (PCBs), polybrominated diphenyl ethers (PBDEs), and organophosphate (OP) pesticides have been shown to interfere with thyroid hormone action in humans and in laboratory studies (13). The human population is ubiquitously exposed to these chemicals (14), and several high-quality studies have documented adverse outcomes. Moreover, ASD and ADHD are common disorders with global prevalence rates in 2010 of 6.2–7.6/1000 (15, 16) and 6.1% (17), respectively. Both outcomes are complex and heterogeneous, with both genetic and environmental origins.

Environmental exposures, including lead, methylmercury, arsenic, certain drugs, tobacco smoke, and pesticides, have been linked to ASD as well as ADHD (18–23). Although these conditions may be identified in individual children, developmental neurotoxicity may also affect brain function in more subtle and insidious ways, resulting in cognitive deficits that affect much larger numbers of undiagnosed children, sometimes referred to as a “chemical brain drain” (1).

Resulting neurodevelopmental disabilities are costly to the affected individuals, to their families, and to society. For example, in 2010, the lifetime societal cost per individual with ASD without intellectual disability was estimated to be about $1.4 million in the United States and the United Kingdom, driven chiefly by adult productivity loss (24). Likewise, national annual costs for ADHD in The Netherlands were found to be between €1.04 billion and €1.53 billion, including lost parental productivity and income (25). In the European Union (EU), costs for environmentally attributable childhood and adolescent disorders (defined to include ASD, ADHD, and conduct disorder) and intellectual disability were previously estimated to be nearly $10 billion (26), although these estimates preceded very recent studies documenting IQ loss in association with exposures to PBDEs and OPs (27–29).

In the context of emerging evidence regarding EDC contribution to neurodevelopmental disease and disability and well-developed methods for calculating economic impacts of cognitive deficits and neurobehavioral disorders (30), the present report attempts to utilize current epidemiological and mechanistic data linking EDC exposure to neurobehavioral outcomes to estimate the attributable disease burden and costs to society. Because environmental contributions to the burden of disease may be easily underestimated due to uncertainties in the evidence (31), our goal was to generate realistic estimates based on the strength of evidence using a framework first developed in regard to climate change (32, 33). We focused on costs attributable to exposures in Europe in the context of active regulatory decision-making on EDCs.

Materials and Methods

Overall approach

Based on the above evidence, the expert panel focused on four exposure-outcome relationships with the greatest evidence for causation: PBDE exposure with reduced cognition, OP exposure with reduced cognition, endocrine disruptor exposures (including phthalates) with ASD, and endocrine disruptor exposures (including OP and PBDE) with ADHD. The panel selected these exposure-outcome relationships because of the presence of well-conducted, longitudinal human and animal studies to assess developmental neurotoxic effects of these EDCs. We adhered to the approach described in the accompanying overarching manuscript (34) in evaluating the strength of the epidemiological (using the World Health Organization GRADE Working Group criteria) (35, 36) and toxicological literature (using the Danish Environmental Protection Agency criteria) (37) and to assigning probability of causation (adapting the Intergovernmental Panel on Climate Change [IPCC] criteria) (32). The Supplemental Data describes exposure biomarker inputs used to model exposure in the EU and approaches to valuing costs of reduced cognition, ASD, and ADHD, whereas subsequent sections describe estimation of affected populations and attributable prevalence/incidence.

Modeling PBDE-47 and OP-associated IQ loss and intellectual disability

The number of births in the year 2010 in the EU were calculated to represent the percentile ranges 0-ninth, 10–24th, 25–49th, 50–74th, 75–89th, and 90–99th to allow application of estimates of exposure across each range. The lowest grouping was assumed to have no exposure, whereas the other groups were assumed to have levels corresponding to the lowest extreme (eg, 10th percentile for all births in the 10–24th percentile grouping). The panel selected the exposure-response relationship from the longitudinal birth cohort study with the prenatal PBDE levels most closely corresponding to levels found within the EU (38) and applied a reference level corresponding to the 10th percentile in the longitudinal study to model IQ loss in each of the five higher exposure groups. Births for each country in 2010 were obtained from Eurostat (39). The exposure-associated IQ point loss was multiplied by the number of births in each percentile range to calculate the total IQ loss within each exposure interval. The sum of these losses produced the best estimate of the IQ loss due to prenatal PBDE exposures. The expert panel also applied an alternate higher exposure-response relationship from one of the other two longitudinal cohort studies associating IQ loss with prenatal exposure (28) as input to a sensitivity analysis, thus recognizing the uncertainty in the exposure-response relationship that remains despite substantial research.

We also modeled increases in intellectual disability, defined as IQ < 70, assuming a normal distribution with mean 100 and SD 15. Within each exposed group, the NORMDIST function in Microsoft Excel 2010 was used to identify increases in intellectual disability associated with the associated decrement in IQ. The increase in percentage with intellectual disability across percentiles of exposure was multiplied by each country's population estimate to quantify attributable cases of intellectual disability to PBDE exposure, and these estimates were aggregated to quantify the total EU burden.

We followed a nearly identical approach with OPs, in which the population of births was divided into identical percentile ranges. The exposure-response relationship represented a sample size-weighted average of results from two longitudinal birth cohorts (27, 40) with results from each cohort used as inputs for extrapolation in sensitivity analyses. After applying the exposure-response relationship described by Bellinger (41), estimates of total IQ points lost utilized the above-mentioned birth data from Eurostat (39).

Estimating EDC-attributable autism

The expert panel chose a longitudinal study of prenatal phthalate exposure from which to extrapolate ranges for potential burden of autism attributable to EDC exposures (42). This study was selected because of its unique measurement of autistic behavior, confounders, and biomarkers of EDC exposure in a prospective longitudinal design. No similar studies on this issue were identified. As with OP- and PBDE-attributable IQ loss, the population was divided into percentile ranges with levels assigned to groups at the lowest extreme of the percentile range.

Increments in the social responsiveness score (SRS), a quantitative scale for measuring the severity of social impairment related to ASD in the general population, were calculated by multiplying the log(base exp) of the ratio of the maternal urinary low molecular weight phthalate in each percentile to the presumed reference level (the maternal urinary low molecular weight phthalate estimated for the 10–24th percentile exposure group) by the increment per log unit increase identified by Miodovnik et al (42) (ie, 1.53). We also modeled shifts in the distribution of SRS, assuming a normal distribution of mean of 29.7 and an SD of 16.8, as identified from normative data (43), using the NORMDIST function. Severe social impairment is typically identified as SRS ≥ 75. Incremental increases in autism attributable to phthalates were determined by subtracting the percentage of SRS ≥ 75 in each exposed group minus the percentage of SRS ≥ 75 in the unexposed scenario. These increments were aggregated and divided by 0.62%, the more conservative of the two recent global autism prevalences, because of its exclusion of Asperger's syndrome (15), to quantify one data input for the ASD attributable fraction (AF). The expert panel then considered a range of AFs including a base case estimate, accounting for other EDCs plausibly associated with ASD but for which no similar studies were available. These AFs were multiplied by the number of 8 year olds with autism, which was estimated by multiplying the 0.62% prevalence by each country's population estimate of 8 year olds in 2010 obtained from Eurostat (44). These estimates assume that prevalence is equivalent to cumulative incidence of ASD, positing that early life EDC exposures manifest at various time points before age 8.

Estimating EDC-attributable ADHD

In attributing ADHD to EDC exposures, the expert panel identified positive longitudinal studies for OP pesticide and PBDE exposure in pregnancy (22, 23). To avoid double counting, the expert panel chose to extrapolate a range of attributable burdens of disease using studies of these two exposures, rather than assuming additive effects of these two exposures in their contribution to ADHD. As a conservative measure, the panel chose to use a cross-sectional study of OP exposure (19) that identified a more modest exposure-response relationship than the longitudinal study (22). Following the approach described for ASD, odds ratios (ORs) in the Bouchard et al study (19) were exponentiated by the log(base 10) unit of the ratio of the total dialkyl phosphate (DAP) concentration in each EU exposure group by 65.0 nmol/L, the reference level used to calculate IQ loss. Recognizing that ADHD is more prevalent than ASD and that the use of an OR rather than a relative risk in the Levin equation (45) could produce overestimation of the AF, adjustment was applied to estimate the relative risk (46). The OR for each group was input, along with the percentage of the population with that range of exposure, into the Levin equation to calculate AFs (45). The AFs were then summed across the exposed groups to identify an aggregate OP AF for ADHD. The Gascon et al (23) study identified an OR of 1.80 for ADHD among the 20% with detectable levels compared to others with nondetectable levels. After correction to relative risk as above, these values were used to generate an AF for PBDEs.

The expert panel used the OP AF as a base case estimate and the PBDE AF as a sensitivity analytic input. Following the same approach made for ASD, the estimated AFs for ADHD were multiplied by the 6.1% global prevalence of ADHD (17). As with ASD, we utilized prevalence as a proxy for cumulative incidence.

Results

PBDE-attributable IQ loss and intellectual disability

The expert panel identified moderate-to-high epidemiological evidence for IQ loss attributable to PBDE exposure. The panel identified four well-designed, longitudinal observational studies (birth cohorts). Three of the studies (39, 47) identified consistent, exposure-response relationships with IQ, with carefully collected data on many potential confounders. The fourth study (23), from Spain, suffered from modest sample size with few detectable PBDE levels. Although this study showed substantial directionality toward cognitive and motor dysfunction at age 4 (23), IQ was not measured. It should be noted that the three other studies were of US populations. This is relevant because exposure levels in the United States are much higher than in the EU. The congener distribution is different in the EU, with substantially higher levels of PBDE-47, the congener most consistently associated with cognitive deficits, than in the United States.

Exposure-response relationships for PBDE-47 from these longitudinal birth cohorts were selected to develop a range of exposure-response relationships, with a reference level of 2.82 ng/g, corresponding to the 10th percentile in the longitudinal cohort study (38) that the panel selected for extrapolation in the base case scenario. The Menorca, Spain, cohort had a median cord blood concentration of 2.1 ng/g; it found positive directionality, although it did not reach statistical significance (P > .05), suggesting that the selection of a reference level of 2.82 ng/g is conservative. This reference level is also supported by data from animals that document interference with thyroid function by PBDE-47 in fetal lambs at exposures of 0.2 μg/kg birth weight per day (48), ie, serum concentrations only slightly above those that are prevalent in the EU (49–51). Additional support comes from the US Environmental Protection Agency use of a benchmark dose approach to develop a reference dose of 0.1 μg/kg birth weight per day based on neurotoxicity in mice (52, 53). Furthermore, the European Food Safety Authority calculated that current exposures are less than one-tenth of the benchmark dose for various population groups at risk, thus leaving a fairly small margin of exposure (54).

The panel concluded that there is strong evidence for an endocrine mode of action of PBDEs contributing to IQ loss. The foundation of this conclusion is the evidence that PBDE exposure interferes with thyroid hormone action during development, and that interference with thyroid hormone action causes IQ loss. Supplemental Table 1 tabulates the evidence from humans, animals, and in vitro studies that PBDE interferes with thyroid hormone action. In addition to modulating the interaction of thyroid hormone with its receptor, these substances may also affect the metabolism of thyroid hormone (55). Thyroid hormone is essential for normal brain development (5); thyroid hormone insufficiency during development can produce different effects on the brain depending on the timing of the insult (6, 56), and such deficits persist into adulthood (57). However, PBDE-associated developmental neurotoxicity may occur without detectable changes in thyroid function (58). Also, several potential non-EDC mechanisms have been highlighted (59), but the preponderance of the literature suggests that thyroid disruption plays an important role in the pathogenesis of PBDE-induced developmental neurotoxicity.

IQ loss was estimated to occur only in the two highest percentile fractions of the EU population in base case analyses (0.52 to 0.84 IQ points), with alternative scenarios ranging from IQ loss only in the highest quantile (0.27 points) to larger IQ losses in the two higher quantiles (1.19 to 1.94 points). In total, 873 000 IQ points were estimated to be lost annually (sensitivity analysis, 149 000 to 2.02 million), with associated productivity loss of €8.40 billion (sensitivity analysis, €1.43 billion to €19.4 billion). An additional 3290 (sensitivity analysis, 544 to 8080) cases of intellectual disability were estimated, with €1.19 billion (sensitivity analysis, €148 million to €2.93 billion) in associated social costs.

Together, the evaluation of the epidemiological and toxicological evidence led to the assessment of a 70–100% probability that PBDE neurotoxicity costs the EU €9.59 billion (sensitivity analyses, €1.58 billion to €22.4 billion) annually, using the IPCC criteria (Table 1).

Table 1.

PBDE-Associated IQ Loss, Intellectual Disability, and Costs of European Children Born in 2010

Expert panel evaluation of epidemiological evidence	Moderate-to-high
Expert panel evaluation of toxicological evidence	Strong
Probability of causation	70–100%
Percentile of exposure	0–9	10–24	25–49	50–74	75–89	>90
Percentile assumed	0	10	25	50	75	90
Cord blood PBDE (base case), ng/g	0.00	0.00	0.00	2.60	4.61	6.27
Cord blood PBDE (sensitivity analysis), ng/g	0.00	0.00	0.53	1.60	2.68	3.66
IQ loss (low)	0.00	0.00	0.00	0.00	0.00	0.27
IQ loss (base case)	0.00	0.00	0.00	0.00	0.52	0.84
IQ loss (high)	0.00	0.00	0.00	0.00	1.19	1.94
Births	541 000	812 000	1 350 000	1 350 000	812 000	541 000
IQ points lost (low)	0	0	0	0	0	149 000
IQ points lost (base case)	0	0	0	0	418 000	454 000
IQ points lost (high)	0	0	0	0	968 000	1 050 000
Lost economic productivity (low)	€1.43 billion
Lost economic productivity (base case)	€8.40 billion
Lost economic productivity (high)	€19.4 billion
Attributable intellectual disability (low)	544
Attributable intellectual disability (base case)	3290
Attributable intellectual disability (sensitivity analysis)	8080
Cost of intellectual disability (low)	€148 million
Cost of intellectual disability (base case)	€1.20 billion
Cost of intellectual disability (high)	€2.93 billion
Total costs (low)	€1.58 billion
Total costs (base case)	€9.59 billion
Total costs (high)	€22.4 billion

OP pesticide-associated IQ loss, economic productivity, and intellectual disability

The expert panel identified moderate-to-high epidemiological evidence for IQ loss attributable to OP exposure. The panel found three well-designed longitudinal observational studies (birth cohorts) (28, 40, 60) that identified consistent, exposure-response relationships with carefully collected data on many potential confounders. These studies were of populations with much lower exposure levels than in the EU, although it should be noted that two of the three studies examined urinary DAP, an indicator of recent (acute exposure approximately 1 wk before sampling) exposure to OP pesticides. Although chlorpyrifos is the chief OP pesticide used in Europe, in the United States (where these studies were conducted), other pesticides may have contributed to the DAP levels, raising the issue of modest exposure imprecision.

The panel also identified strong toxicological evidence for effect via an endocrine-disrupting mechanism. The principal mode of action of chlorpyrifos is through acetyl cholinesterase (AChE) inhibition, although many reports indicate neurotoxicological effects independent of AChE inhibition. Developmental exposure of mice to levels of chlorpyrifos that had no effect on AChE activity in mice adversely affected thyroid hormone levels (61). A number of other animal studies have shown chlorpyrifos to affect thyroid hormone signaling. Jeong et al (62) carried out a one-generation reproductive toxicity study on chlorpyrifos using doses of 1, 10, and 100 mg/kg birth weight/d. They found significant effects on thyroid signaling (either thyroid histology or circulating levels of T₄ and/or TSH) after long-term exposure in utero and through lactation. In terms of neurotoxicity, Levin et al (63) showed that similar exposures (1 or 5 mg/kg birth weight/d) in late gestation (gestational day 17–20) also produced significant, long-lasting effects on behavior in pups tested in adolescence and adulthood. Interestingly, effects were greater at the lower dose. Furthermore, the 4-day period of exposure studied is well known to be a phase of peak neurogenesis, dependent on maternal thyroid hormone supply (64).

Exposure-response relationships already established were selected to develop a exposure-response relationship in the base case scenario, with a reference level of 65 nmol/L urinary total DAP (41). Bouchard et al (63) reported that a 10-fold increase in total urinary DAP was associated with a loss of 5.6 IQ points (28, 40), whereas Engel et al (41) reported that a 10-fold increase was associated with a loss of 1.39 points. Weighting the effect estimates by sample size produces an expected loss of 4.25 full-scale IQ points for a 10-fold increase in urinary DAP over this range.

Detectable OP exposure levels in the EU population ranged from 79.9–1160.8 nmol/L in base analyses (Table 2) and 34.2–444.8 nmol/L in sensitivity analyses. These levels were associated with decrements in IQ between 0.38 and 5.32 points in base case scenarios and a range of 0.12–7.01 lost IQ points in sensitivity analyses. Each year, 13.0 million IQ points are lost (sensitivity analysis, 4.24–17.1 million), with an associated productivity loss of €124 billion (sensitivity analysis, €40.8 billion to €164 billion). In addition, 59 300 additional cases of intellectual disability were attributed to prenatal OP exposure (sensitivity analysis, 16 500 to 84 400) across the EU, with an additional €21.4 billion (sensitivity analysis, €5.96 billion to €30.5 billion) in social costs. The evaluation of the epidemiological and toxicological evidence led to an assessment of 70–100% probability of OP neurotoxicity that costs the EU between €46.8 billion and €195 billion annually.

Table 2.

OP-Associated IQ Loss, Intellectual Disability, and Costs of European Children Born in 2010

Expert panel evaluation of epidemiological evidence	Moderate-to-high
Expert panel evaluation of toxicological evidence	Strong
Probability of causation	70–100%
Percentile of exposure	0–9	10–24	25–49	50–74	75–89	>90
Percentile assumed	0	10	25	50	75	90
Urinary total DAP (base case), nmol/L	0.00	79.92	175.55	280.58	741.31	1160.78
Urinary total DAP (sensitivity analyses), nmol/L	0.00	34.20	97.30	200.00	370.00	444.79
IQ loss, base case scenario	0.00	0.38	1.83	2.70	4.49	5.32
IQ loss, Low	0.00	0.12	0.60	0.88	1.47	1.74
IQ loss, High	0.00	0.50	2.42	3.56	5.92	7.01
Births	541 000	812 000	1 350 000	1 350 000	812 000	541 000
IQ points lost (base case)	0	310 000	2 480 000	3 650 000	3 650 000	2 879 000
IQ points lost (low)	0	101 000	811 000	1 190 000	1 190 000	942 000
IQ points lost (high)	0	408 000	3 270 000	4 810 000	4 800 000	3 790 000
Lost economic productivity (base case)	€125 billion
Lost economic productivity (low)	€40.8 billion
Lost economic productivity (high)	€164 billion
Attributable intellectual disability (base case)	59 300
Attributable intellectual disability (low)	16 500
Attributable intellectual disability (high)	84 400
Cost of intellectual disability (base case)	€21.4 billion
Cost of intellectual disability (low)	€5.96 billion
Cost of intellectual disability (high)	€30.5 billion
Total costs (base case)	€146 billion
Total costs (low)	€46.8 billion
Total costs (high)	€195 billion

EDC-attributable autism

Due to the paucity of epidemiological studies, the expert panel identified low epidemiological evidence for ASD incidence attributable to EDC exposure. The panel identified two informative longitudinal observational studies (21, 42); although both controlled well for potential confounders, they identified different EDC exposures linked to autism-associated behaviors. Although exposure-response relationships were identified, only a single urinary measure of phthalates in one study raises substantial exposure imprecision concerns that may cause bias toward the null. Phthalates are also a mixture of chemicals with variable androgen and thyroid antagonism. These studies have not followed children to the teenage years, when ASD behaviors would ideally be measured to confirm associations identified in the toddler years. SRS is a nonspecific association, and so this outcome imprecision also reduces confidence with regard to causal attribution.

As a basis for extrapolation to estimate AF of disease for EDCs, the panel used data from a longitudinal birth cohort study identifying increases in prenatal phthalate exposure with increases in SRS, an index used to evaluate autism (42). Using the 10th percentile of exposure in the EU as a reference level, 0.34–1.97 point increases in SRS were identified (Table 3), with increases in SRS values over and above the 75 reference level typically associated with autism between 0.021 and 0.143%. In total, the AF was estimated to be 8.88%, using these assumptions.

Table 3.

Estimates of Phthalate-AFs for Autism of European Children in 2010

Percentile of exposure	0–9	10–24	25–49	50–74	75–89	>90
Percentile assumed	0	10	25	50	75	90
LMW phthalates, ng/mL	0.00	21.30	35.40	60.00	212.00	416.00
Increase in responsiveness score, assume 10th percentile reference level and 1.53 increase/log(LMW)	0.00	0.00	0.34	0.69	1.53	1.97
Increase in SRS > 75, assuming mean 30 and SD 13	0.00%	0.00%	0.021%	0.045%	0.106%	0.142%
AF, assuming baseline 0.62% prevalence	0.00%	0.00%	1.45%	3.03%	7.21%	9.68%
Overall AF	8.88%

Abbreviation: LMW, low molecular weight.

The panel was beset with the difficulty that SRS is an intermediate index of autism spectrum features and that SRS alone is not diagnostic for autism. Therefore, as a conservative measure, the panel chose to use the above estimate as a guide in estimating total EDC AF for autism, rather than attributing the estimated burden of autism directly to phthalates. Recognizing other EDCs besides phthalates as contributors to autism, the panel chose 5% as a base case estimate of AF with 2–10% as inputs for sensitivity analyses. The panel noted that the Miodovnik increments in SRS (42) were much smaller than those noted in sex-stratified analyses by Braun et al (21). The sex predilection of ASD further suggests the biological plausibility of hormonal mechanisms disrupted by EDCs. The panel also noted twin studies suggesting that approximately 50% of ASD can be attributed to environmental factors (66), although gene-environment interactions can occur through EDC mechanisms. A National Academy of Sciences panel in 2005 identified 28% of neurodevelopmental disabilities including autism to be due at least in part to environmental factors (67).

The panel also noted strong toxicological evidence for autism-associated pathologies via different endocrine-disrupting mechanisms. The panel identified moderate evidence for causation for each chemical category. Because maternal hypothyroidism increases ASD risk 4-fold (68), it is logical to examine links with those EDCs associated with ASD risk in human cohorts with data on thyroid hormone signaling pathways and brain development. This is the case for PBDEs and OPs, as discussed in the previous sections. It is also the case for phthalates, a chemical class for which exposure has been related to increased ASD risk (69). Although most work on the endocrine-disrupting effects of phthalates has focused on antiandrogenic effects (70), many epidemiological and animal studies show effects on thyroid hormone signaling. An early rat study (71) reported significant effects of dibutylphthalate on circulating levels of the active form of thyroid hormone, T₃, and these effects were seen at 250 mg/kg/d, levels half the lowest dose modifying circulating T. The findings in animals associating phthalates with reduced thyroid levels has also been shown in humans through nationally representative cross-sectional studies in the United States (72), as well as Danish children between 4 and 9 years old (33).

After applying a 0.62% autism prevalence (Table 4), an estimated 12 300 8 year olds in the EU were autistic, with 316 cases attributable to EDCs (sensitivity analysis, 126–631), after reducing by 48.5% to account for coexisting intellectual disability and to avoid double counting. Together, the findings suggest a 20–39% probability that EDC-associated ASD costs between €79.8 million and €399 million annually.

Table 4.

Estimates of Endocrine Disruptor-Attributable Autism and Costs of European Children in 2010

Expert panel evaluation of epidemiological evidence	Low
Expert panel evaluation of toxicological evidence	Moderate
Probability of causation, %	20–39
AF (sensitivity analysis), %	5 (2–10)
Prevalence of autism, %	0.62
No. of autism cases, 8 y olds	12 300
No. of attributable autism cases after accounting for coexistent intellectual disability, 2010 (sensitivity analysis)	316 (126–631)
Attributable lifetime autism costs, 2010 (sensitivity analysis)	€199 million (€79.7 million to 399 million)

EDC-attributable ADHD

The panel identified three longitudinal studies and one cross-sectional epidemiological study examining EDCs and ADHD, leading to the consensus of low-to-moderate epidemiological evidence for ADHD attributable to EDC exposures. The cross-sectional study identified strong exposure-response relationships of DAP, with ADHD diagnosis based upon validated parental questionnaires (19). One longitudinal birth cohort identified an exposure-response relationship of PBDE-47 in child blood at age 4 with attention-deficit symptoms (but not with the few cases with ADHD diagnosis), having carefully controlled for many potential confounders (23). The second longitudinal study, by Chen et al (29), also found increased hyperactivity scores in children born to mothers with higher PBDE-47 levels than the Gascon et al (23) study and was also well controlled for confounders. The third longitudinal study, by Marks et al (22), identified increased frequency of symptom-based ADHD diagnosis, although not with visual attention, in a exposure-response relationship to maternal urinary DAP levels. The Gascon et al (23) study controlled for PCB, dichlorodiphenyldichloroethylene, dichlorodiphenyltrichloroethylene and hexachlorobenzene (but not OP) exposures, whereas the Chen et al (29) study was unable to control for other EDC exposures. Only the Gascon et al (23) study is of European origin, although exposures in the other two US-based studies are of much lower exposures, and EU exposures to alkyl phosphates are thought to be mainly chlorpyrifos. Exposure-response gradients were defined in the Chen et al (29) and Marks et al (22) studies, but not in the Gascon et al (23) study, which included very few cases. Although the substances involved are EDCs, the studies did not consider EDC-related pathogenesis.

The panel concluded that there is strong evidence for the ability of endocrine disruption to contribute to ADHD incidence in humans. Humans with generalized resistance to thyroid hormone have a high risk of ADHD (73), and this is related to a genetic mutation in TRβ (74). A reproduction of this mutation in mice also leads to ADHD-like behaviors (75). Moreover, low serum thyroid hormone levels in pregnant women are linked to ADHD (76), and low thyroid hormone in pregnant rodents also leads to ADHD-like behaviors in the offspring (77). Although animal models for ADHD are difficult to establish, chemicals that interfere with thyroid hormone action (including PCBs and bisphenol A) also produce ADHD-like activity in rodent studies (78, 79). Corroborating evidence from pregnancy cohorts in humans demonstrate that maternal subclinical and clinical hypothyroidism both contribute to ADHD (21, 53, 66). Because PBDE and OPs have been previously described as thyroid inhibitors, the mechanistic link for an EDC-mediated effect on ADHD risk is strong.

As a basis for extrapolation to estimate AFs for EDCs, the panel decided to calculate two AFs, using the Bouchard et al (19) and Gascon et al (23) studies, respectively. The panel decided against the use of the Marks et al (22) study, which showed stronger ADHD associations, whereas association with attention tests could not be demonstrated. Using data from a cross-sectional study of urinary DAPs (19) to extrapolate potential EDC-attributable burden, effects were identified in the most highly exposed half of the population with estimated relative risk of ADHD between 1.03 and 1.42 (Table 5). AFs of 10.8–17.3% were identified, with 19 400 to 31 200 children with ADHD after excluding 44.3% of potentially attributable cases due to coexistent intellectual disability. The costs of these cases were estimated to be €2.40 billion (sensitivity analyses, €1.21 billion to €2.86 billion). Using data from Gascon et al (23), a higher AF of 12.5% was applied, resulting in an estimated €1.74 billion (Table 6; sensitivity analyses, €1.41 billion to €2.07 billion). Together, these analyses suggest a 20–69% probability that EDC-associated ADHD costs the EU between €1.21 billion and €2.86 billion annually.

Table 5.

Extrapolating EDC-Attributable ADHD and Costs for OP of European Children in 2010

Percentile of exposure	0–9	10–24	25–49	50–74	75–89	>90
Percentile assumed	0	10	25	50	75	90
Total urinary alkyl phosphate (base case), nmol/L	0.00	79.92	175.55	280.58	741.31	1160.78
Total urinary alkyl phosphate (sensitivity analyses), nmol/L	0.00	34.20	97.30	200.00	370.00	444.79
Estimated relative risk, ADHD (base case)	1.00	1.03	1.13	1.19	1.34	1.42
Estimated relative risk, ADHD (sensitivity analyses)	1.00	1.00	1.05	1.15	1.24	1.26
AF (base case), %	17.28
AF (sensitivity analyses), %	10.76
ADHD prevalence among 12 y olds, %	6.10
No. of attributable ADHD cases after accounting for coexistent intellectual disability, 2010 (base case)	31 154
No. of attributable ADHD cases after accounting for coexistent intellectual disability, 2010 (sensitivity analyses)	19 388
Cost of attributable cases (sensitivity analysis)	€2.40 billion (€1.21 billion to 2.86 billion)

Table 6.

Extrapolating EDC-Attributable ADHD and Costs for PBDE of European Children in 2010

Percentage exposed	20
ADHD OR	1.80
AF, %	12.53
ADHD prevalence among 12 y olds, %	6.10
No. of attributable ADHD cases after accounting for coexistent intellectual disability, 2010	22 600
Cost of attributable cases (sensitivity analysis)	€1.74 billion (€1.41 billion to 2.07 billion)

Discussion

Economic calculations have been used to help prioritize societal investments in health care, environmental protection, and other important sectors (80). Adverse effects of neurodevelopmental toxicity have recently emerged as an important public policy concern (1, 2). For example, the global costs due to pollutants, such as lead (26, 81) and methylmercury (30, 82), are very substantial, and this evidence has helped inspire global efforts to phase out the use of leaded fuel and to control the release of mercury and other air pollutants to the environment (83, 84).

The present study builds upon this experience and extends the calculations of societal expenses due to neurodevelopmental toxicity associated with exposure to endocrine disruptors. The calculations on costs due to cognitive deficits show that the few substances that were suitable to analyze are associated with costs that total over €150 billion per year within the EU. This is likely an underestimate due to the exclusion of neurotoxicant effects of EDCs banned by Europe (eg, under the Stockholm Convention), such as PCBs. The main cost is due to widespread occurrence of cognitive deficits expressed in terms of IQ points lost, whereas a sizeable, though smaller, amount is associated with specific diagnoses of ASD and ADHD. This finding is in accordance with results obtained for, eg, mercury, where costs associated with cognitive deficits are much greater than those associated with specific diagnoses (85).

We endeavored to make our estimates as precise as possible, but were limited due to sizable uncertainties in the evidence. Thus, although these calculations were based on a small number of chemicals for which we have data, they illustrate for the first time that neurotoxicity associated with endocrine disruption is very costly to society. Insofar as other PBDEs have effects beyond those of PBDE-47, our estimates may be substantially conservative. These calculations do not include the costs due to potential cognitive deficits associated with exposures to, eg, phthalates (86) or perchlorate, a thyroid hormone disruptor (87). We also did not examine the potential for synergistic effects between EDCs that might heighten the effects of a single EDC exposure. Given that social responsiveness score is an intermediate index of autism spectrum features and that SRS is not a diagnostic test for autism, the panel chose to err on the conservative side, use a lower base estimate for attributing autism, and use this estimate for autism attributable to all EDCs. We were also limited by a paucity in European data on exposure-outcome relationships; whereas control for confounding was strong in many of the studies used, some of the extrapolations are from subpopulations (eg, Mexican American), and our results are predicated on the generalizability of exposure-outcome relationships to European populations. Many of the exposure-outcome relationships used to extrapolate disease burden were log-linear and therefore had a supralinear component, although non-monotonic relationships may also exist, just as they have been identified in animal studies. Finally, the extrapolation of attributable ADHD from a cross-sectional study bears some emphasis. Although current exposures could be a proxy for exposures earlier in development, it is unlikely that ADHD is caused by exposure occurring at the time of diagnosis. Using the prospective studies would have presented the same weakness as our extrapolation for ASD in that the prospective studies of ADHD only look at symptoms, rather than diagnosis of ADHD.

The strength of the approach taken includes the transparent use of available data to define dose-related outcomes and the distribution of exposures in EU countries, and such estimates will become more precise as better evidence becomes available. The causal attribution is supported by experimental data, and the judgment in regard to reference levels, impact of covariates, and steepness of the dose-dependence of the outcomes was based on consensus among the authors. Likewise, biomarker data were not available for all EU countries, and judgment was used in extrapolating to the EU as a whole. By this approach, the authors attempted to avoid underestimating the burden of disease simply because of insufficient or missing data (31). On the other hand, the calculations could not take into account potential differences between exposure levels in the member states.

A previous independent effort to calculate the EU costs due to environmentally attributable intellectual disability estimated that these costs were $61.9 billion (27) based on the effects of lead and methylmercury poisoning only. Our calculated costs associated with several industrial chemicals are of the same order of magnitude. These estimates are likely to be additive rather than duplicative and testify to the substantial societal impact of cognitive deficits.

Although endocrine disruption is defined generally as the chemical disruption of endocrine systems (88), the exact mechanisms of neurotoxic effects are usually not known in detail. For all of the substances reviewed in this study, more than one mechanism is likely. However, disruption of thyroid hormone action is a commonality among the substances we included in the current analysis. This is important because the strength of the data linking thyroid disruption to cognitive and neurobehavioral disorders (4–7) provides confidence in our analysis. However, the fact remains that these chemicals are known to exert effects also through other pathways, and it is not clear which is the most important. Proof of a mechanism in humans is difficult if not impossible to establish, and the authors therefore relied on the EU definition of an EDC and a judgment based on weight of evidence and plausibility. Again, it bears emphasis that the exposure-outcome relationships extrapolated to disease burden are but a subset with the greatest evidence of the EDCs for which evidence of developmental neurotoxicity has been identified. Our calculations therefore suggest that prevention of EDC exposure would result in substantial societal benefits.