*General description*. We applied the model first used by the

Institute of Medicine (1981) to estimate the cost of environmentally mediated disease.

Although BLLs reflect mainly the exposure to lead that occurred in the previous few months, and may not reflect the burden of lead in bones (

Needleman et al. 1996), BLL is the most commonly available measure. Increases in BLL among children are associated with decrements in cognitive development, as quantified in IQ loss. We limited our economic analysis to the neurodevelopmental impact of lead, assessed as decrements in IQ point loss estimated over three ranges of BLLs: 0.513 IQ point loss per 1-μg/dL for BLL 2–10 μg/dL; 0.19 point loss for BLL 10–20 μg/dL; and 0.11 point loss for BLL ≥ 20 μg/dL, as described by

Gould (2009). We focused on the population at risk represented by each 1-year cohort of children < 5 years of age, in whom the BLL, when measured longitudinally, is most strongly associated with neurodevelopment at school age (

Hornung et al. 2009).

We did not include mild mental retardation (MMR) in our cost estimates, because cost estimates for MMR are rarely available outside the developed world. Decrements in IQ are associated with reduced lifetime economic productivity, and associations with criminality have also been identified (

Needleman et al. 1996;

Reyes 2007), but these to date are limited to industrialized countries. Therefore, we did not include costs of increased criminality in our estimates of economic costs to LMICs.

*Environmentally attributable fraction*. We applied an environmentally attributable fraction (EAF) (

Smith et al. 1999) of 100%, consistent with scientific literature indicating that only a very small fraction of lead exposure is attributable to natural processes (

UNEP 2010). Accordingly, the attributable cost can be described as follows:

Cost = EAF × BLL × (IQ loss/BLL) × (lost economic productivity/IQ loss) × population at risk. [1]

In this equation, IQ is the IQ loss for each BLL range of values, as described above, and the population at risk is represented by each 1-year cohort of children < 5 years of age. We estimated the number in each cohort as 20% of the total number of 0- to 4-year-old children reported for each country by the most recent UN estimates (

United Nations 2012).

*Estimation of BLL distributions at the country level*. We systematically reviewed the published literature for studies estimating BLLs in LMICs, following the most recent World Bank country classification by income (

World Bank 2012a). The published literature was searched using PubMed (

http://www.ncbi.nlm.nih.gov/entrez/) and terms including “lead” in combination with the name of each country; the initial query was then refined using the “Related Citations” option. We also considered reference lists of relevant articles. We included only studies conducted from 2000 onward (or for which the recruitment period extended to the year 2000), in pediatric populations (< 18 years of age) or that included a pediatric subpopulation. Studies reporting lead exposure in heavily contaminated areas (hot spots such as areas around metal smelters and battery-recycling or gold ore–processing activities—the latter responsible for the recent outbreak of fatal lead poisoning in children in Nigeria (

Dooyema et al. 2012)—or occupational exposures were excluded, unless they included a control population not residing in the contaminated area. In these latter cases, we analyzed only data from the control population. Country-specific BLL estimates identified based on our review and used in the present analysis are provided along with the sources of these data in Supplemental Material, Table S1.

For this analysis, we did not consider urban and rural populations separately, because there is a global trend toward urbanization, and more than half of the world’s population now lives in urbanized areas, with urban growth concentrated in Africa and Asia (

United Nations 2011).

*Estimating BLL from past studies.* We developed a regression model to relate trends in BLLs over time, and to relate these to the timing of the ban in leaded gasoline in each country; this was done using BLL data retrieved through our literature search. We first examined a simple linear model with respect to trends in BLL over time, and compared our results with a linear plus quadratic model, which resulted in a modest increase of the predictive capability, measured using the *R*^{2} coefficient of determination. Therefore, our final model included a quadratic term, and is described by the following regression equation:

*y*(*t*) *=* β_{0}
*–* β_{1}*x +* β_{2}*x*^{2}
*+ e*, [2]

where

*y* is the average BLL at time

*t* (2008), β

_{0} is the intercept,

*x* is the difference between year of the study and year of leaded gasoline phaseout in the country (

UNEP 2012),

*x*^{2} is the quadratic term of the difference, and β

_{1} and β

_{2} are the coefficients being estimated. The quadratic term is also justified by experience in developed countries, in which the most rapid reductions in childhood blood levels were produced immediately after phaseout and in relationship to more rapid reductions in leaded gasoline use (

Fewtrell et al. 2004;

U.S. EPA 2003).

Parameter estimates obtained with this model are shown in , and were used to estimate BLL in each of the countries included in our analysis. Below is a working example for a specific country, Ethiopia, which has no recent BLL data available and in which leaded gasoline was phased out in 2004:

| **Table 2**Model parameter estimates for BLLs and related SDs predicted for each country in 2008. |

BLL in 2008 = [7.33 – (0.26 ×

4) + (0.01 × 16)] = 6.45 μg/dL.

We used the same model to derive SD values, but with the inclusion of BLL as one of the coefficients:

*y*(*t*) *=* β_{0}
*+* β_{1}*x*_{1} + β_{2}*x*_{2} – β_{3}*x*_{2}^{2}
*+ e*_{i}, [3]

where *y* is the average SD at time *t* (2008), β_{0} is the intercept, *x*_{1} is the average BLL, *x*_{2} is the difference between year of the study and year of leaded gasoline phaseout, *x*_{2}^{2} is the quadratic term of the difference, and β_{1}, β_{2}, and β_{3} are the coefficients being estimated. Therefore, for Ethiopia, we estimated the following SD:

SD in 2008 = [0.27 + (0.47 × 6.45) + (0.14 × 4) – (0.001 × 16)] = 3.85 μg/dL.

For countries with available data, the actual BLL and SD values were used in the regression equation; if the data were collected after 2008, we subtracted 2008 from the year of the study and used the difference. For some of these countries, more than one study reporting blood lead concentrations was available. In this case, we first estimated BLL levels in 2008 using our regression model and then combined these estimates to derive a single, sample size–weighted, geometric mean, according to a method previously described by

Fewtrell et al. (2003). An example of this is provided in the Supplemental Material, Methods.

The same procedure was followed to combine SDs. Once we estimated mean BLL and SD for each country, the percentage of children at or above predefined blood levels intervals (2–10, 11–19, ≥ 20) was estimated to determine the population at risk within each exposure interval assuming a log-normal distribution around the estimated mean BLL using the LOGNORMDIST function in Excel 2007 (Microsoft, Inc., Redmond, WA). For this study, we considered BLL < 2 µg/dL to present the lowest risk of toxic effects in children, acknowledging that a threshold level does not appear to exist.

*IQ loss.* Current evidence supports impaired cognitive development associated with lead concentrations < 10 µg/dL, and a nonlinear, inverse relationship between IQ and BLL has been established (with the greatest rate of IQ loss per unit blood lead < 10 µg/dL). Average IQ point loss was derived from an international pooled analysis (

Lanphear et al. 2005), over three ranges (0.513 point loss per 1-μg/dL for BLL 2–10 μg/dL; 0.19 point loss for BLL 10–20 μg/dL; and 0.11 point loss for BLL ≥ 20 μg/dL), as described by

Gould (2009). Because of the broad range of BLL ≥ 20 μg/dL, we also divided the ≥ 20 μg/dL group into 20–44, 45–69, and ≥ 70 μg/dL for analysis. For each of these BLL ranges, we applied the IQ point loss corresponding to the lowest BLL in the range considered (e.g., for the range 2–10 μg/dL, we applied the IQ loss corresponding to 2 μg/dL). IQ loss was calculated for each country using the BLL estimated for that country multiplied by the number of children < 5 years of age affected each year. IQ losses for each country were then summed to obtain a total for each subregion in Africa, Asia, and Latin America/Caribbean.

*Losses in economic productivity.* To estimate lead-attributable costs, the economic model developed by

Schwartz et al. (1985) was applied to the calculated prevalence distribution. This model is based on the relationship between lead exposure and dose-related decrements in IQ score, the latter in turn being associated with decreased lifetime earning power.

We estimated lost lifetime economic productivity (LEP) using average IQ point loss per microgram per deciliter BLL, percent lost LEP per IQ point, and total lost LEP. Lost LEP was derived based on a U.S. estimate (

Grosse et al. 2002) of decrements in LEP per IQ point loss. For our base-case analysis, we assumed a 2% loss in LEP–IQ point estimate, as previously done (

Trasande and Liu 2011), against LEP data from the University of California Institute for Health and Aging, which assume annual growth in productivity of 1% and a 3% discount rate (

Max et al. 2007). These data suggest that the value of lifetime expected earnings is $1,413,313 for a 5-year-old boy in 2007 and $1,156,157 for a 5-year-old girl. These data were then corrected at the country level using gross domestic product (GDP) by converting GDP per capita to international dollars using purchasing power parity (PPP) rates (

World Bank 2012b). All monetary amounts reported are in international dollars, unless otherwise specified.

*Sensitivity analyses.* Recognizing uncertainty in LEP–IQ and in trends in BLL, we performed two types of sensitivity analysis to increase the accuracy of our estimates. First, we applied the method used by

Fewtrell et al. (2004) to estimate the exposure distributions in our population of interest. Following this approach, we also accounted for lead-reduction programs that were undertaken after BLLs were surveyed. We used a reduction factor of 7.8% decrease per year (

Fewtrell et al. 2004), taking into account the year of the study and the year of leaded gasoline phaseout in each country with available data. For countries with more than one study reporting BLL, we derived a single, sample size–weighted, geometric mean BLL value and SD (for more details, see Supplemental Material, Methods.) We then obtained a subregional mean BLL by weighting country means by the size of the population < 5 years of age. For countries for which BLL data were not available, we used the corresponding subregional mean and SD to estimate the population distribution of exposure, assuming a log-normal distribution around the mean BLL for the subregion. Unlike our regression model, this method does not allow for an estimation of BLL at country level for those countries with no recent data available, and uses instead the corresponding subregional BLL mean as a substitute.

Second, recognizing the uncertainty in the relationship between IQ and economic productivity, we used the low and high ends of our estimate range based on the work of

Schwartz et al. (1985) and

Salkever (1995), who applied a range in percentages of lifetime productivity loss for each point of IQ ranging from 1.76% to 2.37%.