Modeling Concentrations of Particulate Matter in Solid Fuel Using Households From India : Epidemiology

Secondary Logo

Journal Logo

Abstracts: ISEE 22nd Annual Conference, Seoul, Korea, 28 August–1 September 2010: Air Pollution - Indoor Air Quality and Health Effects

Modeling Concentrations of Particulate Matter in Solid Fuel Using Households From India

Ghosh, Santu; Thangavel, Gurusamy; Sambandam, Sankar; Balakrishnan, Kalpana

Author Information
doi: 10.1097/01.ede.0000392031.69698.c8
  • Free



Combustion of solid fuels in rural households of developing country has been shown to produce very high concentrations of respirable particulate matter, even few folds above the available guideline. A few studies are available that document such concentrations quantitatively. There is a need to develop models that may predict household level concentrations in relation to the household level determinants which can be easily applicable in exposure and health studies.

The study aimed at developing models to estimate particulate matter concentrations in rural households using datasets that contained measured data on concentrations and survey information on a range of household level determinants.


We used 24-hour concentrations of PM2.5 (measured in the kitchen area) obtained from 600 households along with questionnaire based information on household level variables from 4 states (Tamil Nadu, West Bengal, Madhya Pradesh, and Uttaranchal) in India. First, linear regression models were used to predict kitchen area concentrations by household level variables. Subsequently, the logistic model and “classification and regression trees” (CART) technique were applied by dichotomizing kitchen area concentrations (above or below the median).


The linear regression model that included fuel and kitchen type, kitchen ventilation, state, and cooking duration as significant predictors produced an adjusted R2 of 0.33. In logistic regression and CART models, the fuel type, kitchen type, and kitchen ventilation were found to be significantly associated with overall prediction powers as 71% and 76%, respectively. Both logistic regression and CART models predicted almost 90% of the high concentration households but only about 30% of the low concentration households correctly.


Although in need of some improvement in prediction power, the models show substantial promise to be able to generate concentration data for solid fuel using households that may be aggregated to estimate population exposures at the state or national level in India.


(This study was supported by National Research Foundation of Korea [2009–0073407]).

© 2011 Lippincott Williams & Wilkins, Inc.