Estimating the Global Public Health Implications of Electricity and Coal Consumption

Background: The growing health risks associated with greenhouse gas emissions highlight the need for new energy policies that emphasize efficiency and low-carbon energy intensity. Objectives: We assessed the relationships among electricity use, coal consumption, and health outcomes. Methods: Using time-series data sets from 41 countries with varying development trajectories between 1965 and 2005, we developed an autoregressive model of life expectancy (LE) and infant mortality (IM) based on electricity consumption, coal consumption, and previous year’s LE or IM. Prediction of health impacts from the Greenhouse Gas and Air Pollution Interactions and Synergies (GAINS) integrated air pollution emissions health impact model for coal-fired power plants was compared with the time-series model results. Results: The time-series model predicted that increased electricity consumption was associated with reduced IM for countries that started with relatively high IM (> 100/1,000 live births) and low LE (< 57 years) in 1965, whereas LE was not significantly associated with electricity consumption regardless of IM and LE in 1965. Increasing coal consumption was associated with increased IM and reduced LE after accounting for electricity consumption. These results are consistent with results based on the GAINS model and previously published estimates of disease burdens attributable to energy-related environmental factors, including indoor and outdoor air pollution and water and sanitation. Conclusions: Increased electricity consumption in countries with IM < 100/1,000 live births does not lead to greater health benefits, whereas coal consumption has significant detrimental health impacts.


Methods-Data Description
When small gaps in data were present (< 5 years), these were filled using linear interpolation between the two nearest data points in time. In the case of IM, data were only available in 10 year increments from 1960 to 1990 and 5 year increments from 1990 to 2005. The missing data were estimated by applying the spline function in Matlab 2008a to interpolate between the data points. Particularly for models using the IM datasets, the nonindependence of the interpolated datapoints leads to greater uncertainty in terms of the confidence intervals of the model results. Data for several other factors that may predict LE or IM were sought during the preliminary stages of this project. However, none were available for the time period under analysis, as many relevant statistics on education level, vaccination rates, improved water sources, female literacy rates, and health care access and spending are not available across all 41 countries until the 1990s (See Supplemental Material, Table 1). This limitation led us to choose the autoregressive instead of a full multivariate regression methodology, which would require datasets on all potential explanatory variables. The autoregressive method is described in more detail below.

Methods-AR Model Description
To separate the dependencies of LE or IM solely due to coal and electricity consumption patterns, at each time point until time t, Q(t), with the dependencies due to all other reasons, P(t) that can be modelled by its exponential functional form and those associated with the errors of predicting LE or IM at each time point until time t, that could not be captured by either P(t) or Q(t) at each time point until time t: Observation of the estimates of the parameters in the model provided in Table 1   [2]

Methods-GAINS Model Description
In the GAINS model, a linked sequence of calculations leads to estimates of health impact. First, the effects of energy sources and policies on air pollution emissions are estimated. The calculation is based on emission factors and control technologies for specific activities such as electricity generation. Resulting emission inventories for air pollutants along with weather data are used as inputs to a global-regional chemistry transport model. The atmospheric model is used to estimate the functional relationships between emissions of air pollutants in a given

Limitations of AR Models
In the analysis of historical trend data, one must be concerned with confounders. For example, electricity or coal consumption could serve as a surrogate for wealth, which may be the ultimate reason for increased health through any of a host of mechanisms such as increased availability and access to health care, increased vaccination rates, and increased education levels. It is noted that incorporating the linear time trend (a 0 ) as well as previous year's IM (LE) captures the effects of unspecified variables that vary linearly with time. The AR methodology employed here is commonly used to account for unmeasured confounders (Kale et al. 2004;Kovats et al. 2004;Levine et al. 2001) and see also discussion in methods section). The uniqueness of the present study is analysis of  Figure 5).
The a 0 parameter could be considered to be confounded with the coal and electricity consumption parameters, a 1 and b 1 if the cumulative coal consumption and cumulative electricity consumption also varied linearly over time. However this is not the case as only 2/41 countries (Romania and Bulgaria) had insignificant (at the 95% confidence level) quadratic terms for coal consumption and 1/41 (Romania) for electricity consumption.  Note consistency in overall IM reduction rates across models with initial low, mid, and high starting IM. Results are alphabetically sorted, and every 5 and 6 th country (the 1, 6,7,12,13,18,19,24,25,30,31,36, and 37 th ) are not presented due to graphical space limitations.