Improving the Lead Impact Model

Abbreviations

Executive summary

This exploratory report, prepared by Rethink Priorities (RP), presents a suggested revision of the Lead Impact Model (LIM), originally developed by Pure Earth (PE). The model remains a work in progress and is not ready to guide resource allocation decisions. We share it to promote transparency, encourage collaboration, and highlight key data and knowledge gaps.

The LIM was developed by PE to better understand the scale, drivers, and priorities of the global lead poisoning challenge. Its long-term goal is to quantify national health burdens from individual sources of lead exposure, thereby informing policy priorities and supporting cost-benefit analyses of interventions. The LIM uses cumulative population BLLs (cpBLLs)—the average BLL of a population multiplied by its size—as the key metric for quantifying lead poisoning burdens (Fuller et al., 2025). 

RP’s contributions focus on one component of the LIM: estimating national health burdens from individual lead exposure sources using biokinetic modeling. These contributions were developed in close collaboration with PE, which continues to refine the model and progress health burden attribution of lead exposure sources. The approach outlined in this report is based on a simple, transparent linear biokinetic model, adapted from the US Environmental Protection Agency’s (US EPA’s) Adult Lead Methodology. However, the utility of the model is currently limited by major data and knowledge gaps. First, the LIM currently relies on data from PE’s Rapid Market Screening (RMS) study (Sargsyan et al., 2024) and some home-based assessments (Pure Earth, 2023). The RMS study measured the lead concentration of 5,000 consumer product samples across 25 low- and middle-income countries, but was not nationally representative. Second, the LIM does not include all sources of lead exposure, and is best suited for sources involving direct ingestion, such as food and spices. Indirect lead exposure sources are modelled using highly uncertain assumptions relating to the pathways of lead transfer, such as the rate of leaching from cookware and the rate of lead shedding from paint to dust. The modeling of exposures from environmental lead pollution remains undeveloped. Finally, data on how frequently populations are exposed to different lead sources is currently very limited. 

The issues of unrepresentative data and uncertain assumptions are compounded when extrapolating to national burden estimates. As a result, the model’s health burden outputs are currently too uncertain to inform prioritization or resource allocation decisions.

While recognizing these shortcomings, we hope that this research will be helpful to the broader community. We expect that multiple iterations of the LIM will be needed to achieve reliable burden attribution; by regularly and transparently sharing versions for feedback, we hope to speed progress and ultimately arrive at more accurate outcomes. We encourage others to use our low-confidence parameter estimates as starting points from which to hone their own intuitions and identify promising avenues for further research and data collection. 

We emphasize that there is no safe threshold for lead exposure and advocate for the regulation of all sources, regardless of modeled burden. Nonetheless, quantifying burdens may help shape priorities. To make this feasible, we highlight key areas for further research:

• Nationally representative screening of lead in consumer products, especially foods
• Investigating ingestion pathways from industrial pollution (e.g., crop contamination)
• Conducting representative lead leaching tests on foodware and cookware
• Alternative approaches to source burden attribution using BLL data.

Biokinetic modeling for lead exposure attribution

The model structure can be viewed in this spreadsheet.

We estimate the BLL impact of individual lead exposure sources using a modified version of the US EPA’s Adult Lead Methodology (ALM) (U.S. EPA, 1996, p. 2). The ALM is a biokinetic model that estimates individual BLLs based on lead concentrations in ingestion media such as soil, dust, water, air, and food. 

We adapt the ALM so that we can: 

• Incorporate indirect exposure sources (e.g., cookware, paint, toys)
• Model BLL impacts separately for adults and children
• Scale estimates to the population level

Our three primary modifications to the ALM are:

• Leaching rate: We add a new parameter that quantifies the fraction of lead transferred from a source (e.g., cookware) into a medium that can be ingested (e.g., food).
• Age-specific modeling: We apply the model separately to children (0–6 years) and adults (7+ years), consistent with the age categories used in the IEUBK model.
• Adjusted absorption fractions: We revise the ALM’s absorption fraction parameter to better reflect age-specific bioavailability and variation across different lead sources.

While the original ALM estimates BLL for an exposed individual, we extend it to estimate the cumulative population-level burden by multiplying the individual-level BLL impact by the number of people exposed. This enables comparison of total lead burden across sources.

For most sources, our model centers around two equations. Equation 1 shows the individual BLL impact from a given source; Equation 2 scales this to the population level based on the number of people exposed. The variables and parameters are explained in Table 1 below.

Equation 1. Incremental increase in blood lead level (PbBS,I) for an individual exposed to a given lead source. Subscript S indicates that a parameter is specific for each lead exposure source, and subscript I indicates that a parameter can vary across adults vs. children, i.e., I ∈ {adult, child}.

Equation 2. Cumulative population blood lead level (PbBP,S,I) for a population exposed to a given lead source.

Table 1: Overview of model variables and parameters

There are two sources for which we deviated from the model structure, with further detail provided in the respective sections:

• • Cosmetics: We model lead exposure from cosmetics by grouping products into four categories (lipsticks, nail polishes, eyeliners, and skin products) and estimating exposure through skin absorption, hand-to-mouth ingestion, and (for lipstick) direct ingestion. Our approach is adapted from Julander et al. (2020), which models lead exposure pathways in an occupational context and draws partly on the EPA’s ALM.

• Informal used lead-acid battery (ULABs) recycling: As a point source of industrial lead pollution, informal ULAB recycling is modeled separately from other exposure sources. Our approach is based on the model developed by Ericson et al. (2016) and focuses exclusively on inadvertent ingestion of lead-contaminated soil.
  

Model inputs and assumptions by exposure source

Table 2 below summarizes the key modeling assumptions, levels of confidence, and modeling approaches used for each lead exposure source included in our model. Some potentially important sources are not represented due to data limitations or scope, and confidence in most estimates is low. Given the substantial uncertainties in many of these inputs, we encourage readers to make a copy of the model and experiment with adjusting the estimates to explore how results might change. Details for each source follow in the subsections below.

Table 2: Summary of assumptions for exposure sources 

Foods

This section presents our estimates of average adult ingestion rates for each food source included in the RMS (i.e., the categories for which we have data). Other food groups, such as fish, may also significantly contribute to elevated BLLs but are not included here.

Spices

[Confidence: Low-medium. Our estimate is largely based on data from India and a rough classification of countries into high- and low-spice-ingestion categories. We think it’s unlikely that further desk research would uncover significantly better estimates of spice ingestion.]

We found the following estimates of spice ingestion in the literature:

• Helgi Library (2025) reports 2019 annual spice ingestion rates that imply the following daily rates: 9.0 g in Bangladesh, 0.7 g in Georgia, 9.7 g in India, 5.1 g in Morocco, 20.4 g in Nepal, and 3.3 g in Pakistan.
• The UN Food and Agriculture Organization’s (2024) “apparent intake” statistics—based on household surveys from 2012 to 2020—show significantly higher daily ingestion rates of 70.5 g in Bangladesh (2016), 24.3 g in India (2012), and 13.1 g in Pakistan (2019).
• Srivastava et al. (2024) estimate, based on Indian national surveys, that average ingestion of spices is 271 g per month (8.91 g/day).
• Pradeep et al. (1993) find, from a survey of 20 households, an average daily ingestion of 9.54 g of spices.
• Bhathal et al. (2020) find, from a survey of 100 households, an average daily ingestion of 10.04 g of spices among urban adult women and 7.68 g among rural adult women.

We estimate that adult spice ingestion is ~8 g per day in the countries where contamination of spices could be considered a major issue: Bangladesh, Georgia, India, Morocco, Nepal, Pakistan, Bhutan, Egypt, Libya, Sri Lanka, Tajikistan, and Tunisia. In other countries, we estimate that adult spice ingestion is ~1.5 g per day. We did not find child-specific estimates of spice ingestion. Across all countries, we assume that children’s spice ingestion is 40% the amount of adult spice ingestion (i.e., 3.2 g in high-risk countries and 0.6 g in all other countries).

Sweets

[Confidence: Low. Our estimate is based on data from the US, which we adjusted intuitively downward to represent LMICs. We may have missed existing estimates of sweet ingestion in LMICs that could be more accurate.]

In the absence of empirical data on LMICs, we estimate low- and middle-income countries’ per capita ingestion of sweets as one-half of that in the United States. Duyff et al. (2015) estimated candy ingestion among US kids (2-18 years old, 2007-2010) to be 11.9 g/day. For US adults (≥19 years), candy ingestion was estimated to be 9.4 g/day. Reducing these estimates by half for LMICs gives 6 g/day for children and 4.7 g/day for adults.

Staple dry foods

[Confidence: Low-medium. We used data for food supply rather than food ingestion. We make estimates to account for the fraction of the food supply lost as food waste, but with low confidence. However, we think it is unlikely that further desk research would uncover significantly better estimates.]

Pure Earth’s RMS protocol selects the most common nonperishable carbohydrate food(s) typically purchased in each country. Based on Our World in Data’s (2017) “share of energy from cereals, roots, and tubers vs. GDP per capita, 2019,” we visually estimated that, in LMICs, ~55% of energy intake comes from staple foods. We then converted this percentage into grams.

• • We do not directly refer to food supply data, which does not account for wastage or other losses of food. Instead, we take a simple estimate of the energy content of the average adult’s actual food intake as 2,000 kcal. Using our assumption that 55% of energy comes from staple foods, we estimate that staple dry foods make up 1,100 kcal of daily food intake.
  • Assuming staple dry foods are 100% carbohydrates and that carbohydrates have an energy density of 4 kcal/g (Wikipedia, 2025a), we estimate the adult ingestion rate of staple dry foods as 275 g/day.

• We assume that the child ingestion rate is 55% of the adult ingestion rate, roughly following the US Department of Agriculture’s caloric intake guidelines by age (USDA & HHS, 2020, Appendix 2), which yields a child ingestion rate of 151 g/day. 

Herbal and traditional medicine

[Confidence: Low. Our estimate for the rate of ingestion is based on a single, out-of-date report on India, which we adjusted intuitively downward to represent all LMICs. More accurate estimates of traditional medicine use in LMICs may exist in sources we didn’t find, but this seems unlikely.]

 The only data we found was a 2005 report commissioned by the Indian government’s National Medicinal Plants Board (Ved & Goraya, 2007), which estimates the domestic demand for “botanical raw drugs” to be 103,700 tonnes. India’s population in 2006 was 1.172 billion, so annual ingestion per capita (assuming ingestion is equivalent to demand) was 88 g: 0.24 g/day. We suspect India has higher rates of herbal and traditional medicine ingestion than other LMICs, given its long history of traditional Indian medicine. Based on an estimated 30% increase over the LMIC average, this gives an LMIC daily intake of 0.18 g. In the absence of other information, we assume children’s intake is 55% that of adults (0.16 g in India, 0.1 g in all other LMICs).

Foodware and cookware

We consider (1) the amount of food from the cookware & foodware consumed (leachate ingestion rate); (2) the rate of lead leaching from the cookware & foodware into food; and (3) the fraction of the population using the cookware & foodware. We consider four separate types of cookware and foodware: metallic cookware, heated ceramics, non-heated ceramics, and plastic foodware.

Leachate ingestion rates

[Confidence: Very low. We consider our current estimates a temporary placeholder until better evidence is published, as we have not found any studies that would allow us to calculate ingestion rates from specific foodware and cookware types.]

In the absence of empirical data, we make the following estimates for the weight of leachate (i.e., food cooked or contained in each type of foodware or cookware) ingested per person per day among the fraction of the population who uses each type (see Table 3).

Table 3: Foodware and cookware usage estimates

Instead of assuming universal daily ingestion, we assume daily ingestion from foodware/ cookware occurs among a defined fraction of the population, and that this fraction (usage rate) varies by cookware or foodware type. In the absence of empirical data, we devised intuitive placeholder estimates of these fractions (see specific estimates in later subsections). 

Metallic cookware

Leaching rate

[Confidence: Medium. Our estimate is based on two leachability studies carried out across 10-25 LMICs and refined with expert input. We expect that further desk research is unlikely to meaningfully improve this estimate, but the publication of leachability studies under real-life cooking conditions would likely provide a meaningful update.]

The most relevant studies informing our estimate are:

• Weidenhamer et al. (2016) tested 42 aluminum cookware items from 10 LMICs for lead leachability during cooking, boiling with 4% acetic acid for 2 hours. We took the XRF analysis of cookware samples and the amount of lead leached per 20 mL serving to calculate an average leaching rate of 0.00004 (see calculations).
• Binkhorst et al. (2025) tested 100 aluminum pots from 25 LMICs in the RMS study, also boiling for 4% acetic acid for 2 hours to assess lead leaching. We calculate the average leaching rate from the study as approximately 0.00006.

Lead concentrations in leachate can vary substantially for pots with the same total lead concentration, with differences of at least two orders of magnitude reported in the literature (Binkhorst et al., 2025). The variability in new pots is driven by factors such as differences in manufacturing techniques, the heterogeneous distribution of lead within the aluminum, the dissolution of aluminum along preferential pathways, and the presence of potential coatings. Other factors affecting the amount of lead leaching from all pots include the age of the pots, cooking times and temperatures, cooking residues, and the pH of foods. Common test protocols, such as boiling with acetic acid, are intended to mimic acidic foods such as a tomato sauce, but may significantly overestimate lead leaching into commonly consumed staples like rice, corn, and cassava that may represent the majority of foods consumed.

Based on a conversation with Binkhorst, we think it may be more realistic to assume a leaching rate that is approximately one-third of the values observed in acetic acid-based tests based on dietary considerations and other variables, although this is a rough estimate. Binkhorst also noted that cold leaching from metallic foodware or cookware is negligible, so we do not focus on non-heated metallic cook-/foodware. The vast majority of leaching studies are carried out on aluminum cookware. According to Binkhorst, this is because it is the most commonly used metal for cookware in LMICs, although other studies indicated similar issues with other metals like bronze and brass (Hore et al., 2025; Fellows et al., 2025).

Thus, we reduce the leaching rates from the literature to roughly one-third of the available estimates, which results in a leaching rate of 0.000017.

Usage rate

[Confidence: Low-medium. Our estimate is based on qualitative, anecdotal reports of widespread use, and the single study that included quantitative estimates had citations that we could not verify. We believe that it is unlikely that our estimates can be significantly improved unless metallic cookware usage rates in LMICs are published.]

We estimate with low confidence that 70% of households across LMICs use metallic cookware, based on the following literature:

• Osborn (2009): “While products made by aluminium casters are today ubiquitous in West Africa – it is almost impossible to find a household, food stall, or restaurant that is not home to a locally-made cast aluminium cooking pot […].”
• Habimaana et al. (2022): “Like most Sub-Saharan Africa households, about 73% of the rural population in Uganda use locally made cookware recycled from scrap metal [23], while very few rich urban dwellers use the factory-made pans […]. A survey in Ntungamo district reported 68% of the households to be using locally fabricated cookware, commonly referred to as ‘endongoburaya’ [24].”
• Bergkvist et al. (2010): “Cooking [in Matlab, Bangladesh] is often carried out outdoors using traditional Chulla, a mud-built cylinder, inserted into the ground, on which cooking pots and utensils are placed. By far, the most commonly used cocking pots are of low-quality aluminum.”

Ceramic foodware

We expect that heated ceramic ware is associated with significantly higher leaching rates than non-heated ceramics but is used less frequently.

Leaching rate

[Confidence: Low. Our leaching rate estimates for heated and non-heated ceramics are based on two studies of uncertain representativeness regarding real-life cooking behavior and other regions/ communities. Our understanding is that leaching from lead-glazed ceramics is heavily influenced by the temperature at which the glaze is fired, with high temperatures immobilizing lead; however, there is a lack of available research to quantify these effects. We do not expect that our estimates can be meaningfully improved through desk reviews without further leachability studies on ceramic foodware.]

Heated wares

We found a handful of studies that investigated the leachability of ceramic cookware (Gould et al., 1983; Mohamed et al., 1995; Sheets et al., 1996; Villalobos et al., 2009). Only one study (Lynch et al., 2008), investigating lead-glazed ceramic cooking vessels in a Hispanic community in Oklahoma, reported both the lead content of the ceramic and the leachate after heating, which allowed us to calculate the leaching rate:

• The mean lead content of lead-glazed ceramics was 19,231 ppm. 
• The mean lead content of the three tested foods after cooking in lead-glazed ceramic was: ~5,007 µg/L for tomatoes, 200 µg/L for black beans, and 70 µg/L for hominy.
• Assuming that each of these foods represents a third of typical food intake (which may not be realistic), we obtain an average of 1,759 µg/L, which corresponds to ~1.8 ppm.
• Thus, we estimate a leaching rate of 0.00009 for heated ceramic ware (=1.8/19,231).

Non-heated wares

Mohamed et al. (1995) studied the lead leachability of ceramic tableware in Malaysia under different temperatures. Based on Fig. 2 in the study, we estimate that leachability at the boiling point of water results in approximately 7 times more leached lead relative to room temperature. Thus, for non-heated ceramic ware, we use a leaching rate of 0.000013 (=0.00009/7).

Usage rates

[Confidence: Low. We have not found any direct estimates of ceramic foodware/cookware usage rates, but several data points we found offered some qualitative sense of usage with which we generated our highly uncertain best-guess estimates. We think it is unlikely that our estimates can be significantly improved through desk research without studies on ceramic usage rates.]

We found only three studies with relevant information:

• Belgaied (2003): “Until now, dishes, bowls and other kitchen utensils made of glazed earthenware are widely used in Tunisia for the preparation and storage of a large variety of foodstuffs.” 
  • This is a qualitative description of widespread (i.e., likely >50%) use of glazed ceramic wares in Tunisia and might be outdated.

• Welton et al. (2016): “The use of ceramic cookware decreased from over 90% during respondents’ childhood household use in Oaxaca to 47% in 2006 among households in Baja California, and further reduced to 16.8% in 2012.” 
  • In other words, there is evidence of declining use in two regions of Mexico to just under 50% and less than 20%. Since these figures are from more than 10 years ago, we think it is plausible that ceramic cookware usage rates have probably decreased further to around 20% in Mexico overall. Unlike the observation from Tunisia, this study was specifically on cookware.
• Liu et al. (2023) conducted an online survey on Chinese consumers’ food consumption using ceramic tableware, finding that the average respondent consumed 1,568 g of food and beverage daily using ceramic tableware. The study does not report the proportion of respondents who use ceramic tableware, but the relatively high average weight of food and beverage consumed suggests that ceramic tableware is widespread (likely >50%) in China. These ceramics are not necessarily lead-glazed and are not used for cooking.

Heated wares

While we did not find  any direct estimates of current usage rates of heated ceramics, our impression is that the use of heated ceramic cookware is an important cultural practice in Hispanic communities (less common in non-Hispanic LMICs), though it is generally declining. Based on Welton et al. (2016), we assume a 20% usage rate in Mexico, and our highly uncertain best guess is that 10% of households in LMICs use heated ceramic cookware on a regular basis.

Non-heated wares

While we did not find any direct estimates of the usage rate of non-heated ceramic ware in LMICs, our impression is that usage is fairly widespread. Our intuitive rough guess is that the usage rate of ceramic foodware is about the same as the usage rate for metallic cookware. Thus, we assume that  70% of households use non-heated ceramic foodware.

Plastic foodware

Leaching rate

[Confidence: Very low. We consider our current estimate a temporary placeholder until better evidence is published, as we have not found any studies that would allow us to calculate a leaching rate.] 

The most relevant study we found was Inthorn et al. (2002), who measured the lead leachability of microwavable plastic foodware items from Thailand, but did not report the total lead concentration of foodware that is needed to calculate a leaching rate. Our best guess is that the leachability of plastics is highest when heated, i.e., plastics used in a microwave. We also expect that microwave usage is typically shorter than the usage of metallic or ceramic foodware for heating up food, which may imply a comparatively lower leaching rate. Thus, as a preliminary best guess, we round down the leaching rate for metallic cookware and assume that the leaching rate for plastic foodware is 0.00001.

Usage rate

[Confidence: Very low. We consider our current estimate a temporary placeholder until better evidence is published, as we have not found any studies that would allow us to calculate a usage rate.]

We could not find any estimates of the prevalence of plastic foodware in LMICs, so we take a highly uncertain guess that 30% of all households use plastic foodware.

Paints

We model exposure to lead paint in two different ways: (1) chronic exposure to lead paint via ingestion of house dust, and (2) more acute exposure via ingestion of paint chips. Note that the RMS includes two different types of paint: paints intended for large surfaces, and paints for arts and crafts. We only included paints for large surfaces in our model, as our expectation is that these make up the vast majority of lead paint exposure. 

Ingestion of contaminated dust

We model paint exposure via the leaching of lead-based paint in homes into dust, which is then ingested. Other lead paint exposure pathways exist, such as via soil from exterior paints, via lead paint on playground equipment, and via paint in schools and other public spaces. Our general impression is that these other pathways are relatively minor compared to the daily ingestion of house dust, but we are highly uncertain about this assessment. For the time being, these alternative pathways are omitted from our model.

Leaching rate (paint to dust)

[Confidence: Low. Even though we found a few data points we could use to estimate a leaching rate, we are very uncertain about our estimate, as the extent of leaching varies considerably depending on multiple factors. In the absence of more studies on the correlation between lead-based paint and dust, we think that more work is unlikely to substantially improve this estimate.]

Paint-to-dust lead transfer depends heavily on various factors, such as paint condition, age, quality, and the climate. Thus, there seems to be a consensus that “it has not, therefore, been possible to make a direct correlation between specific concentrations of lead in paint and the resulting concentrations of lead in household dust and the blood lead concentration” (WHO, 2020, p. 16). Nonetheless, for the purposes of our exercise we used data from the following studies in the US to estimate a leaching rate:

• Beard and Iske (1995, p. 17) found that a 1-unit increase in mg/cm2 of interior lead paint was associated with an increase in dust from 2 to 12 ppm (though statistically insignificant). This corresponds to a leaching rate of 0.0002 to 0.0012. 
• We took data from Lanphear et al. (1998, Table 3) and ran a simple linear regression between dust lead loading and maximum interior paint lead content. The resulting leaching rate is 0.00007.
• Dixon et al. (2007) found that a 50% increase in lead paint on windowsills was associated with a 5% increase in floor dust lead. We cannot directly calculate a leaching rate due to the non-linear relationship, but a linear approximation yields a leaching rate of 0.004.

Thus, the very limited data points to a leaching rate between 0.00007 and 0.004. We use the midpoint of this range, 0.002, as a tentative best guess of the leaching rate. However, as we learned from several experts, paints tend to degrade somewhat faster in LMICs than in the US, so it is possible that the leaching rate is higher.

Leachate ingestion rate (inadvertent dust ingestion)

[Confidence: Low-medium. We are reasonably confident that the daily ingestion rate is within a reasonable range of accuracy. However, our estimate of the fraction of population exposed is merely a guess and not based on concrete data, though we think it is unlikely to be off by more than ±20 percentage points. We think it is possible to refine/sense-check this estimate indirectly, e.g., by using home construction characteristics as a proxy of paint usage.]

Ingestion rate:

• Children: The daily ingestion of dust is typically estimated to be between 100 and 400 mg/day in 6-year-old children (IPEN, 2020, p. 11). We use the midpoint of this range, 250 mg/day.
• Adults: The daily ingestion of dust is typically estimated to be around half of the ingestion for children (e.g., ATSDR, 2018, p. 2; Oomen et al., 2008). Thus, we assume an ingestion rate of 125 mg/day. 

Exposure frequency: 

• We assume that a typical person who lives in a home with lead paint has daily exposure, i.e., 365 days per year.

Fraction of population exposed: 

• We have not found direct estimates of the share of painted houses in LMICs. Thus, we make an intuitive guess that the fraction of the population exposed to lead paint in LMICs is ~50%.

Direct ingestion of paint chips

We model the impact of eating paint chips, limited to consideration of children with pica eating disorder. We are unsure about the best way to model this, given the “inappropriateness of use of IEUBK model for paint chip ingestion” (U.S. EPA, 1994, p. 61). The IEUBK documentation does not state exactly why it is inappropriate for paint chip ingestion, but our guess is that the assumed linear relationship (as modeled via the biokinetic slope factor) between lead exposure and BLLs only holds for comparatively low levels of daily exposure, not acute intoxication (which may result from ingesting paint chips with high lead concentrations). We have not found any alternative ways to model paint chip ingestion. To obtain a first, very rough estimate, we decided to stick with our primary model structure. 

Ingestion rate

[Confidence: Low. We only found one estimate in the literature from 1976 to calculate a daily ingestion rate. We are also highly uncertain about how often pica occurs and in how many children, and whether paint chip ingestion in children without pica is really negligible. The evidence base on paint chip ingestion seems very limited, and we don’t expect it to provide much further insight, but speaking with an expert could help sense-check our inputs and calculations.]

Exposure frequency:

• • We only consider children with pica eating disorder. We have not considered the extent to which healthy children ingest whole paint chips on a regular basis.
  • It is estimated that ~20% of children have pica eating disorder (UPMC, 2015). This is likely a substantial overestimate of children regularly eating paint chips, as: (1) pica can be a temporary condition; and (2) not all children with pica ingest paint chips. Our guess is that ~5% of children regularly ingest paint chips.
  • Jacobs et al. (2002, Table 4) found that 35% of US houses with lead-based paint are in a significant state of deterioration. We assume that the same figure holds for LMICs.
  • Our best guess is that ~50% of children in LMICs live in painted houses (see here). 

• We multiply those figures and obtain 0.9%, which is our fraction of children exposed.

Fraction of population exposed:

• Pica “is believed to be episodic, occurring perhaps two to three times per week” (ibid). Taking the midpoint of 2.5 times per week, we assume an exposure frequency of 130 out of 365 days (=2.5/days/week * 52 weeks/year).

Ingestion rate:

• We only found one empirical estimate: “The best available clinical evidence indicates that children with pica may ingest one to three grams of paint per week” (National Research Council, 1976, p. 7). We use the midpoint (2g/week) and, given our exposure frequency of 2.5 ingestion days/week, calculate an ingestion rate of 0.8g.

Toys

We have not been able to find any studies that investigate the dose-response relationship between the lead content in plastic toys and children’s BLLs,  so we devised our own approach based on the ALM, using several strong simplifying assumptions.

In our model, we focus exclusively on plastic toys. An estimated 90% of toys used in the US are made of plastic (McGrew, 2023), but we have not seen comparable statistics for LMICs. We only model lead exposure through leaching into saliva during typical mouthing behavior. Other potential exposure pathways, such as ingestion of toy fragments, are not included.

With more time, we would explore whether non-plastic toys, such as those made of metal or wood, represent a meaningful exposure route, and whether toy ingestion is rare enough to justify exclusion from the model.

Leaching rate (toys to saliva)

[Confidence: Low. We found only one leachability study that allowed us to estimate a leaching rate of lead from plastic toys into saliva during mouthing, and we believe it reflects a worst-case scenario. While we expect that a more extensive literature review is unlikely to yield significantly better data, a conversation with an expert could help us assess how the Kang and Zhu (2013) estimate might be adjusted to reflect a more realistic exposure scenario.]

The only study we found on lead leaching from plastic toys is Kang and Zhu (2013). The authors measured total lead content and its bioaccessibility in the mouth phase (RIVM-M), representing the fraction of lead that leaches from plastic toys into saliva during mouthing behavior. The leaching rate based on the RIVM-M method measured after 30 minutes was 0.037, which corresponds to ~0.005 per 4 minutes (4 minutes because we estimate below that an average child mouths plastic toys for this duration per day). Note that the RIVM-M method was designed to represent a worst-case scenario, immersing a toy in a comparatively large amount of artificial saliva (RIVM, 2004, p. 10). We expect that real-life mouthing behaviors are more intermittent with lower contact times, leading to a substantially lower rate of leaching. Our uncertain best guess is that it is about one-fifth of the estimate in Kang and Zhu (2013). Thus, we use a leaching rate of 0.001.  

Leachate ingestion rate (saliva)

[Confidence: Low-medium. We are fairly confident that the ingestion rate of saliva while mouthing plastic toys is reasonably accurate based on an empirical study on mouthing durations. However, we are unsure about the fraction of children who regularly (i.e., daily) play with plastic toys, as we have not found data on this. We expect that a longer literature review is unlikely to lead to substantially different findings.]

Exposure frequency:

• We assume that a typical child who has access to plastic toys has daily exposure, i.e., 365 days/year.

Fraction of population exposed:

• In the absence of data, we assume that 90% of children play regularly with plastic toys.

Ingestion rate:

• Average daily plastic toy mouthing duration for a child:
  • Juberg et al. (2001) found that children aged 0-18 months mouthed plastic toys for an average of 17 minutes/day, and children aged 19-36 did so for an average of 2 minutes/day. While we have not found estimates for mouthing after 36 months, we expect it is likely close to zero.
  • Assuming that children are roughly evenly spread across ages from 0 to 7 years, we estimate that the average child mouths plastic toys for 4 minutes/day.
• Amount of saliva swallowed while mouthing plastic toys per day (leachate ingestion rate):
  • Children are estimated to produce ~0.5-1.5 L of saliva per day (Raveendran, 2022, p. 4; Watanabe et al., 1995), of which we use the midpoint (1 L/day). This corresponds to ~0.7 mL per minute. Thus, a child swallows roughly 2.8 mL of saliva while mouthing plastic toys per day, which corresponds to 2.8 g.

Cosmetics

[Note: Our model for cosmetics is based on Julander et al. (2020), a study of blood lead contribution by pathway of occupational exposure to metal-cutting fluids. We do not expect our estimates of specific parameters, such as skin absorption rate, to be accurately representative of cosmetic use. We caution against taking the outputs of this section literally.]

Our indicative model for cosmetics can be viewed here. As discussed above, we deviate from our default model structure for cosmetic products in order to account for both ingestion and non-ingestion lead exposure pathways. We organized RMS cosmetic products into four groups: 1) lipstick, 2) nail polish, 3) eye product (e.g., eyeliners, mascara), 4) skin product (e.g., face powder, skin cream).  For each group, we consider multiple lead absorption pathways: skin absorption, direct ingestion, and hand-to-mouth ingestion. For each group of cosmetic products, we encapsulate the effects of all exposure pathways as a single number (β, BLL gradient) that represents the expected BLL increase (in µg/dL) after exposure to a product with 1,000 ppm. We have some uncertainty about how we have categorized the products into four groups, as this may oversimplify or misrepresent the lead ingestion pathways of some products.

Exposure pathways

We modeled the multiple lead absorption pathways based on a model developed by Julander et al. (2020) which estimates the contribution to blood lead by exposure pathway for metal cutting fluids (MCFs) among Swedish brass foundry workers. We chose this model because it is relatively transparent and flexible, and because it partially builds on the EPA’s ALM, making it a good fit with the rest of our modeling approach. 

We show the original model here, which can be compared to our adapted version (see here). Specifically, we adopted its structure for modeling skin absorption and hand-to-mouth ingestion and applied it to all four categories of cosmetics. For lipstick, we added a direct ingestion pathway due to its frequent application near the mouth. We excluded the inhalation pathway due to time constraints and because we expect it to be a relatively minor contributor to overall exposure.

Skin absorption

[Confidence: Very low. Our estimates are based on intuitive adjustments to the empirical absorption rates found in the literature for metal-cutting fluids, and do not consider the material differences to cosmetics. A quick review of lead skin absorption studies suggests the skin absorption rates we obtained are generally conservative (see e.g., Niemeier et al., 2022). While our estimates can serve as an initial guess, it is likely that conversations with experts about cosmetics and/or lead skin absorption could refine them further.]

Our skin lead absorption rates are based on empirical data from Julander et al. (2020) on lead absorption into pig skin, which we assume to have similar absorption characteristics to human hand skin. Specifically, Julander et al. (2020) found:

• Over 2 hours of exposure, 0.00197 absorption
• Over 24 hours of exposure, 0.00374 absorption

We modify these rates based on the thickness of skin in the area of application and the duration of application, shown in Table 4. We believe that skin lead absorption rates may also be significantly influenced by the specific cosmetic formulation and lead compound(s); however, this is not yet included in our model. Notably, MCFs commonly contain inorganic lead compounds, similar to cosmetics.

Table 4: Parameters and justification for cosmetic skin absorption model

Direct ingestion

[Confidence: Low. Our estimates are intuitive: no data was found quantifying lipstick ingestion. Direct lipstick ingestion is relatively insignificant compared to the hand-to-mouth ingestion pathway under the current assumptions—which is open to scrutiny. We think it’s unlikely that our estimates can be significantly improved through desk reviews unless studies on inadvertent direct lipstick ingestion are published.]

Lipstick was the only group for which we considered direct ingestion as a pathway. In the absence of empirical data, we assumed that one-quarter of the amount applied is inadvertently ingested. We used a gastrointestinal absorption rate of 0.2 in line with Julander et al. (2020).

Hand-to-mouth ingestion

[Confidence: Very low. Our estimates are intuitive: no data was found quantifying hand-to-mouth dosage. These parameters significantly affect the total dose ingested under the current assumptions. We encourage consulting experts on hand-to-mouth behavior to improve these assumptions.]

We modified Julander et al. (2020)’s anatomical and behavioral assumptions: “that 13.4 cm2 skin surface on hands is in contact with the peri-oral area, that 24% is transferred from hand-to-mouth and 20% is subsequently absorbed in the gastrointestinal tract.” Since the steel workers studied are exposed via their hands, Julander et al. (2020)’s model (rightly) assumes that hand-to-mouth ingestion would expose them to equivalent doses of lead as applied to the skin. However, for cosmetics, we think that the hands may have lower doses than the areas to which the cosmetics are applied, depending on the type of cosmetic. Hence, we introduced a new variable, “ratio of hand dose to dose in body part of application” (see Table 5 below). We are uncertain about these ratios and believe they could depend heavily on specific cosmetic formulation and behaviors such as handwashing. As with direct ingestion, we used a gastrointestinal absorption rate of 0.2.

Table 5: Parameters and justification for cosmetic hand-to-mouth ingestion model

Usage rates

[Confidence: Very low. We make generalized usage assumptions for the different categories of cosmetic products, which may not reflect all the products within each group. We are also unsure about the appropriate definition of “skin products” and how to distinguish full-body application from facial application; the RMS included samples intended for full-body application. We are very uncertain about the rate of application among regular users and the fraction of population who can be considered regular users. We think it is unlikely that a longer literature search would lead to a much-improved estimate.]

For each group of cosmetics, we estimated the skin surface area of application, rate of application among regular users, and fraction of population who are regular users (see Table 6).

Skin surface area of application:

To determine the skin surface area of application, we prompted Claude 3.5 Sonnet, a large language model developed by Anthropic (see output screenshot). We adjusted some of these estimates:

• We reduced the eye product surface area from 9 cm2 to 5 cm2 given our intuition that eye makeup is often applied more lightly.
• We asked Claude to estimate the surface area for facial cream application (631 cm2), which we then extrapolated to the total for skin products covering the whole body, given Loretz et al.’s (2005) data on the rate of cosmetic use among regular users, which yielded an estimate of 3,000 cm2.

Usage rates of each type of cosmetic product:

• Loretz et al. (2005): 0.024 g/day of lipstick; 2.05 g/day of face cream and 8.7 g/day of body lotion (we took the sum, 10.75 g/day, as total skin product use)
• Loretz et al. (2008): 0.04 g/day of eye shadow (which we took as representative of all eye product)
• YouTube (2017): 8 mL bottle of nail polish provides 50 full coatings. Assuming a density of 1 g/mL: 0.16 g per coating. Assuming a regular application is once per week, we calculated a usage rate of 0.023 g/day.

Fraction of population regularly using cosmetic products:

In the absence of data, we make assumptions:

• Our guess is that, in low- and middle-income countries, 20% of the adult population (~40% of women, ~0% of men) might qualify as regular users of makeup products (lipstick, nail polish, eye product), while 30% of the overall population (~50% of women, ~10% of men) might qualify as regular users of skin products (including creams).
• We think that young children are unlikely to be regular users of cosmetic products (10% as likely as adults), while they are half as likely to be regular users of skin products.

Table 6: Usage parameters for cosmetics

BLL gradients (β)

To summarize exposure across pathways, we calculated a BLL gradient (β) for each cosmetic group. This represents the predicted increase in blood lead level (in µg/dL) for an adult who is regularly exposed to a cosmetic product with a lead concentration of 1,000 ppm. In other words, β reflects the modeled BLL impact per 1,000 ppm of lead in the product, incorporating exposure through all relevant pathways.

We assumed zero exposure for non-regular users. For children, we applied a crude adjustment by multiplying the adult β by 1.25, reflecting the higher absorption in children (2.5x higher bioavailability) but smaller exposed surface area (0.5x compared to adults).

The resulting β and the relative contribution of each exposure pathway for both adults and children are shown in Table 7.

Table 7: Summary of cosmetic BLL gradients and exposure pathways 

Informally recycled used lead-acid batteries (ULABs)

[Confidence: Very low. Our main model for lead uptake through informal ULAB recycling is based on Ericson et al. (2016), but it carries substantial uncertainties due to a lack of data on quantities such as national ULAB waste volumes, the distribution of informal ULAB recycling sites, the number of people exposed, and international ULAB flows. Following Ericson et al. (2016), we only consider lead ingestion from inadvertent soil ingestion. A better understanding of the distribution of ULAB sites and the ingestion routes of lead pollution from point sources (particularly the contamination of agricultural plots and food crops) is essential to reveal the true magnitude of the health burden from ULABs and other industrial sources of lead pollution. The authors further note that substandard formal recycling operations are also known to cause lead exposure. As the ULAB recycling sector increasingly formalizes, this modeling approach may no longer be sufficient.]

Our proposed remodeling of lead exposure from ULABs can be viewed in this spreadsheet.

In line with Ericson et al. (2016), we only consider lead exposure from ULABs through the ingestion of contaminated soil around sites where they are informally recycled. Following the Ericson et al. (2016) approach, we:

1. Estimate the number of contaminated sites where informal ULAB recycling takes place (recycling operations);
2. Make assumptions about the size of the site, level of soil contamination, and the number of people affected by different degrees of soil contamination via soil ingestion;
3. Estimate the amount of lead uptake among the people affected by ULAB-related soil contamination.

Appendix B provides a more detailed description of our model and key uncertainties. 

Methods for estimating the number of contaminated sites

Ericson et al. (2016) use two methodologies to estimate the extent of informal ULAB recycling in 90 LMICs (subsequently tuned based on national economic statistics):

• Method 1: Estimate the total annual ULAB generation in each country, primarily from vehicle use. Subtract the amount formally recycled to derive the volume informally recycled. Then, using assumptions about the size and throughput of informal recycling operations, estimate the number of sites required to process that volume.
• Method 2: Extrapolate the per capita number of informal ULAB recycling operations from Ghana, as reported by Dowling et al. (2016), to other countries.

Method 2 assumes that the density of ULAB recycling operations in Ghana can be generalized to all countries. By contrast, Method 1 offers a more realistic and grounded approach. While Method 1 seemed clearly preferable, we retained both approaches in our work for two reasons:

1. Time constraints: Method 1 is more time-intensive, requiring country-specific data, and we did not find reliable data for many LMICs. In the time available, we only found enough data to calculate results for 34 LMICs.
2. Comparability: Method 2 was relatively simple to implement in bulk and offered a set of results to compare with Method 1. 

Surprisingly, we found that our modified versions of Methods 1 and 2 generated broadly similar findings. Nonetheless, for countries with both a Method 1 and Method 2 estimate, we would place slightly higher weight on the Method 1 estimate.

We made several modifications to both methods (more detail here):

• • Method 1: In addition to passenger cars and commercial vehicles, we include electric two- and three-wheelers and motorcycles in our estimate of ULAB generation, as these vehicles in many countries are predominantly powered by lead-acid batteries. In our spreadsheets, we call this modified method “Method 1a.” For comparison, we also include “Method 1b,” which includes motorcycles but excludes electric two- and three-wheelers. However, we do not recommend Method 1b because it clearly misses a significant portion of ULAB generation. We recommend referring mainly to Method 1a.

• Method 2: We follow Ericson et al. (2016) but reduce Ghana’s estimated site density by 25% to exclude non-ULAB-related lead-contaminated sites. We apply the adjusted lower and upper bounds (31 and 115 sites per million people) to other countries, labeling them “Method 2a” and “Method 2b” in our spreadsheet. We calculate their average as “Method 2 average”, which we recommend using; Method 2a and 2b are deprecated.

We have low confidence in the outputs of both Methods 1 and 2 to estimate cpBLLs from ULABS, and discuss further reasons for uncertainty in Appendix B.

Closing remarks and call for collaboration

The biokinetic approach to modeling cpBLLs described in this report carries substantial uncertainty resulting from the lack of representative data and significant knowledge gaps. The significant degree of uncertainty in modeling the BLL impacts for an individual is compounded and amplified when extrapolating BLL impacts across national populations. The estimates of cpBLLs from individual lead exposure sources are unreliable, and the current modeling should not be used to compare the relative burden of lead exposure sources or to influence resource allocation and national priorities. However, we hope that sharing the LIM can inspire the collaboration and research necessary to progress towards more reliable health burden analysis that may aid national lead poisoning mitigation strategies. 

We believe that our modeling could be improved by:

• Exploring the suitability of other US EPA models—for example, using the All Ages Lead Model (AALM)—rather than modifying the ALM to model BLL impacts for children 
• Sense-checking our assumptions for each exposure source with relevant experts

Recommendations for future research

We outline below a list of priority research topics that may significantly improve the accuracy of estimating the cpBLL burden of lead exposure sources. The items are presented in no particular order:

Nationally representative surveys of lead levels in consumer products: 

• These would increase the transparency of how many people are exposed to each lead source and to what level. This can be pursued through market-, shop-, and home-based assessments and should account for the significant regional variation in the prevalence of exposure sources, consumer preferences, and the market share of individual brands. 

Cookware and foodware leaching tests:

• While leachate tests based on acetic acid (4%) are a useful approach for defining conservative regulatory standards, they are much too aggressive to reflect cooking practices and overestimate lead leaching. To better understand the health burden from cookware and foodware, specific leachate test protocols should be developed to reflect local cooking practices, considering the leachate pH, contact time, cooking temperatures, and food storage practices.
• Studies should quantify the effect of glaze firing temperatures on the lead leaching rate from lead-glazed ceramics.
• Studies should investigate lead leaching from plastic foodware. 

Food lead screening:

• The low-level lead contamination of staple foods has the potential to cause substantial cumulative cpBLL burdens across populations; however, there is a lack of transparency of lead levels in foods.
• Studies should review and identify foods that have a high affinity for absorbing lead from the environment, and which foods pose significant cpBLL risks.

ULAB and industrial lead pollution point sources:

• The distribution of ULAB sites and industrial lead pollution point sources is unclear. Further research is needed to quantify the number of people exposed to industrial sources of lead pollution.
• Lead pollution persists in the environment and embeds into ecosystems and food chains. However, common models to quantify cpBLLs from industrial pollution only consider the inadvertent ingestion of contaminated soil. Studies should quantify other significant routes for the ingestion of industrial lead pollution, such as the contamination of local agricultural sites and food crops. 

Other sources of lead exposure:

• The lead sources initially included in the LIM are based on the available data and are not exhaustive. Additional significant sources of lead exposure should be added to the LIM, such as smoking, drinking water, and other industrial point sources of pollution, such as coal power plants.

Other approaches to estimating cpBLLs

Biokinetic modeling offers a structured framework for estimating BLL impacts when the amount of lead ingested by an individual can be quantified with reasonable accuracy. It is particularly well-suited for sources of direct ingestion (such as food), provided variables like ingestion quantity and exposure frequency are known. However, it is data-intensive and less reliable for sources with indirect ingestion pathways, where data are limited and lead transfer rates between media are uncertain. 

For example, with cosmetics, there is substantial uncertainty around several key variables, such as the rate of dermal absorption and bioavailability of the specific lead compound, the amount of inadvertent ingestion (such as from hand-to-mouth behavior or lipstick use), the surface area of application, contact time with the skin, and the typical quantity of product used. Similarly, lead exposure from paint is highly variable depending on the condition and deterioration of the painted surface. Lead leaching from foodware also depends heavily on factors such as the manufacturing process and the presence and integrity of surface coatings.

An alternative approach is to use multivariate regression analysis to statistically compare the BLLs of users and non-users of a given product, thereby estimating the average BLL impact associated with that exposure. When conducting such studies, recording the lead concentration of the exposure source would enable dose-BLL response relationships to be quantified. We expect that much of the relevant data may already exist but has not been published in a form that facilitates such analysis. Re-analyzing raw data from existing studies could therefore be a promising avenue. However, when establishing and interpreting dose-response relationships and attributing BLL impacts to specific lead sources, the substantial variability in exposure patterns and pathways should be taken into account.

Appendices

Appendix A: Bioavailability by source in the IEUBK model

Figure A1: Bioavailability data entry for all gut absorption pathways

Note. From U.S. EPA (2007, p. 13). “Bioavailability” in the IEUBK model corresponds to “absorption fraction” in the ALM.

Appendix B: More detail on ULAB methodology

Methodology for estimating the number of informal recycling operations

We mostly followed Ericson et al.’s (2016) two methods, described above, to obtain estimates of the number of informal ULAB recycling operations. We also spoke with Bret Ericson, the lead author of the paper, to better understand his methodology and hear his suggestions for our replication and update. We modified his original methodology, taking into account several of his suggestions, and used the most recent country data we could find. Next, we describe the methods as we implemented them. (Any errors in this section are our own.)

Method 1: We estimated the informal “secondary production” (henceforth “recycling”) of lead as the difference between (1) reported formal recycling of lead and (2) estimated total recycling of lead.

• We obtained data on formal recycling from the United States Geological Survey.
• We estimated recycling from vehicle batteries using vehicle use data (various sources).
  • This involved identifying the number of vehicles in use by type in each country, and assuming each vehicle type generates a certain amount of lead per year, to calculate the weight of recyclable lead generated from vehicle use per year. See Table B1 for the specific assumptions we used.
  • For electric two- and three-wheelers, we also assumed the proportion that run on lead-acid (as opposed to lithium-ion) batteries for each country. 

Table B1: Annual weight of lead generated by vehicle type

• We then extrapolated to total recycling from vehicle battery recycling, assuming the share of lead generated by each lead-acid battery application equals its market share. Specifically, we took a rough guess that 70% of the total weight of recyclable lead comes from automotive and motorcycle SLI batteries based on several market research reports. We made a further adjustment to 72.6% to account for the use of deep-cycle batteries in electric two- and three-wheelers.
• We then calculated the weight of informally recycled lead as the difference between formal recycling and total recycling.
• Finally, we followed the paper’s approach in assuming the weight of informally recycled lead is distributed into three classes of operation (small [100 t recycled per year], medium [500 t], large [1,000 t]) to establish an initial estimate of the total number of informal recycling operations by country.

Method 2: This was the previous approach taken by the LIM and assumes that the density of contaminated sites is equivalent to that of Ghana (as reported by Dowling et al., 2016). Since the Dowling estimate is expressed as a range (31-115 contaminated sites per million inhabitants), we calculated lower- and upper-bound values of the total number of informal recycling operations by country according to this methodology. We call our lower- and upper-bound estimates Method 2a and Method 2b, respectively.

• We modified Dowling et al.’s (2016) and Ericson et al.’s (2016) methodology by discounting the proportion of informal ULAB recycling sites from 37% to 28%. This adjustment excludes non-ULAB lead-contaminated sites identified in Pure Earth’s Toxic Sites Identification Program. The discount of one-quarter was a rough guess based on a spot check of contaminatedsites.org.
• This methodology was feasible for all 127 countries included in the LIM. While we think this method is less accurate, we still include it for reference and for the sake of comparability.

• After establishing initial estimates of the number of informal recycling operations through Method 2, we adjusted them based on four variables that Ericson et al. (2016) argue correlate with informal ULAB recycling: country’s GDP (PPP) per capita, relative size of informal economy, rate of urbanization, and relative size of the mining, manufacturing, and utilities sector. These variables are expressed relative to those of Ghana (such that Ghana’s variables are all exactly 1), weighted, then combined as a multiplier. (We did not apply these adjustments to Method 1.)

Once the number of informal ULAB recycling operations was estimated, we distributed them into size classes following Ericson et al.’s (2016) assumptions (the same referred to above). We modeled each size class of operation as creating three exposure scenarios—such that 15% of affected people experience soil lead concentrations of 5,000 ppm, 35% of them experience 2,500 ppm, and 50% of them experience 850 ppm (Figure B1).

Figure B1: Ericson et al.’s (2016) soil contamination exposure scenarios

Note. Figure created by Rethink Priorities; not to scale. Circles represent radii of 100 m, 200 m, and 300 m.

We then modeled these soil lead concentrations, generating elevated BLLs through soil ingestion per dose-response models defined by the US EPA—IEUBK for children under 7 years old and ALM for adult women aged 17+ (see Table B2). The study assumed “250 to 450 mg/day” of soil ingestion but did not specify the specific values used for each age cohort (IEUBK’s default values of intake are in the range of 52-94 mg/day, having been revised downward from 85-135 mg/day). As we could not determine the exact dust and soil ingestion rates used in the study, and to take into account the downward revision of US ingestion rates, we opted to universally apply a 250 mg/day ingestion. We did not spend much time on this decision and are highly uncertain as to whether this accurately reflects ingestion rates in LMICs.

(Note our soil lead uptake modeling deviates from Pure Earth’s current LIM, which uses the Lead Exposure Reduction Impact Calculator [LERIC]; LERIC assumes a 135 mg/day soil ingestion rate for children and 50 mg/day for adults. Pure Earth also assumed child ingestion rates for the whole population, while we segmented the population by age.)

Table B2: Individual BLL from exposure to lead-contaminated soil

Note. Blood lead levels are expressed in µg/dL. The soil/dust ingestion rate is configured as 250 mg/day in line with the lower bound suggested by Ericson et al. (2016) for all age groups. See spreadsheet for more detail.

Finally, we calculated cpBLLs (by age group and total) by:

1. Applying the country’s age structure (World Population Prospects) to affected populations in each exposure scenario;
2. Multiplying each age group’s population by relevant BLL elevations (Table B2); and
3. Summing across age groups and scenarios.

Room for improvement

Compared to Method 2, Method 1 has a sounder methodological structure and uses relatively more reliable country-level empirical data. However, we think that Method 1 remains relatively inaccurate at the individual country level due to poor data quality and because international flows of lead scrap are not considered. We think it would be useful to prioritize improvements to (and expand country coverage of) Method 1.

We were initially concerned about the validity of Method 1 because it suggests that many HICs, such as the US, have extremely high amounts of unaccounted-for lead (and thus, informal recycling), although in reality, we expect HICs such as the US to have near 100% formal recycling. In addition, the model suggests some countries have more formal recycling of lead than estimated total recycling, resulting in invalid negative estimates of informal recycling volume. Therefore, we used the US as an example to check the model’s validity. (The LIM does not include HICs like the US, but more reliable data is available for the US.)

Using Method 1, we initially estimated that the US generates ~9 million tonnes of ULABs annually, but it only formally recycles ~1 million tonnes according to the United States Geological Survey (USGS, 2022). A report commissioned by Battery Council International (BCI, 2023) estimated that the US lead battery industry has a 99% recycling rate, and estimated the total weight of lead recycled from batteries in the US was 12.2 billion pounds (5.5 million tonnes) from 2017 to 2021—thus, 1.1 million tonnes per year, which is similar to the USGS figure.

This discrepancy can be almost fully explained by three factors:

1. How OICA classifies vehicles: We realized that OICA reports the US having an unusually high number of commercial vehicles (176 million) and fewer passenger cars (116 million) than one would expect based on the US car ownership rate and population size. Since our calculations assumed commercial vehicles are equivalent to trucks, we estimated a very large amount of battery use from US commercial vehicles. The US Bureau of Transportation Statistics (2024) indicates that there are 258 million “light duty vehicles” and 15 million trucks and buses in the US.
2. Longer battery service life in HICs: We used battery service life estimates from Tür et al. (2016), who write that most of Africa’s “lead acid battery market is dominated by inexpensive batteries with a limited lifetime. Furthermore, high temperatures and bad road conditions affect the lifetime of batteries negatively. Due to this, the service life of a lead-acid battery in use in Africa is assumed to be around two years” (p. 13). By contrast, BCI (2023) estimates that batteries last four years in passenger cars and light trucks and three years in trucks and heavy-duty trucks.
3. Exports and imports: The BCI (2023) report states that US net exports of battery scrap in the 2017-2021 period amounted to 6.9 billion pounds (3.1 million tonnes) of lead (p. 6)—i.e., around 620,000 tonnes of lead per year. A relevant expert suggested that we look into data from UNCTAD.

When we adjusted the US inputs by replacing the OICA vehicle data with higher-quality government data and tuned the battery service life assumptions to be more in line with HIC values, we found that the gap closed to approximately 540,000-600,000 tonnes per year. That is in the ballpark of the US net export volume of 620,000 tonnes, meaning that essentially all of the recyclable lead is accounted for. We did not have time to repeat this process with other countries.

Given that these three factors (and potentially others) have the potential to generate significant errors, we are wary of taking the estimates based on either Method 1 or Method 2 at face value. Our low-confidence prior guess was that Method 1 results are only likely to be accurate within 1 order of magnitude, while we think the Method 2 results are unlikely to be accurate for countries other than Ghana. However, since the majority of Method 1 and Method 2 estimates of cpBLLs from ULABs are within 1 order of magnitude of each other, this generally does not update our view of the importance of ULABs.

Overall, however, we still think this model structure is likely to be a useful basis for further work. We think 15-20 extra hours of desk research to source high-quality parameters and international trade data could significantly improve Method 1, as well as potentially produce estimates for countries for which there is currently no Method 1 estimate due to a lack of data. (With more time and resources, one could also try to improve Method 2, i.e., do more studies similar to Dowling et al. (2016).)

There are several aspects of Ericson et al.’s (2016) model that we did not have time to evaluate in depth:

• • The post-initial estimate adjustments according to GDP (PPP) per capita, relative size of informal economy, rate of urbanization, and relative size of the mining, manufacturing, and utilities sector: We think the GDP (PPP) per capita adjustment may currently be applied in the wrong direction. It’s currently set up so that low-income countries generate more recyclable lead, but the lowest-income countries likely have very few vehicles with lead-acid batteries and thus generate fewer recyclable batteries. In middle-income countries, by contrast, there seems to be greater potential for lead-acid battery generation to overwhelm their limited formal recycling capacity. (We did not have time to consider this more carefully and opted to follow the methodology.)

• The completeness of the USGS data: The majority of LMICs are not mentioned as hosting any formal lead recycling, but we believe there could be some countries where formal recycling exists but is not recorded as such in the USGS dataset. We had to source a separate estimate of Bangladesh’s formal lead recycling capacity (30% of its total generation; UNCTAD, 2024) and expect that there are other countries we missed.

• Expert opinion-based model assumptions on the size distribution of recycling operations and the number of people exposed to elevated soil concentrations: We did not have time to independently check these.

There are also several out-of-model uncertainties that we think could affect a decision maker’s view of the importance of informal ULAB recycling in lead mitigation, but which we did not have time to look into:

• Persistent soil contamination from abandoned recycling operations, which were raised as an issue by Richard Fuller of Pure Earth, is not considered in the model. We lack clear data on the rate at which such operations close and open, so we did not attempt to quantify this effect. However, we would not be surprised if this significantly increases the actual number of people exposed to lead from ULABs.
• The potential effects of geophagy (especially among pregnant women) in lead-contaminated soil, and whether this means ULABs pose a unique threat to pregnant mothers.