Time series study on air pollution and mortality in Indian cities R Uma, Kaplana Balakrishnan, Rajesh kumar

Background Air quality issues are of major concern for many cities in Asia and other developing countries Increasing attention from policy makers, legal body, NGOs, research, academic institutions and funding agencies Many initiatives, but gaps in research still exist

PAPA study: Indian cities Selection based on geographical representation, population, data availability, Research team capability Delhi: National capital with 13.8 million population Ludhiana: Industrial town in Punjab, North India with population of about 1.5 million Chennai: Located in coastal area of Southern India with population of 4.5 million

Aim and Objectives of the study Aim: To generate site specific database on effect of air pollution on mortality for Indian cities Delhi, India Specific objectives: –To assess the time series data on air quality parameters and mortality to study the relationship between air pollution and mortality due to respiratory diseases in Indian cities –To assess the daily change in mortality in relation with change in air quality after controlling for the exogenous parameters

Salient features of the study Multidisciplinary team Meeting ICMR guidelines on ethical aspects Review and guidance from ISOC QA/QC audit Capacity building –Training on developing exposure series –R Package –Core model for time series analysis

Study Locations

Methodology Collection of retrospective time series data (2002, 2003 & 2004) on –Ambient air quality –Mortality data –Meteorological data (Temperature, humidity, visiblity) Statistical analysis of data to study the association of age specific death (all cause mortality) with exposure to air pollution

Station name Station code No of observationsMean Std.Deviati onMinMax IHC ITO Sarojini NagarSan Town HallToh MayapuriMap NizamudinNmd SrifortSrf JanakpuriJkp ShazadabaghShb ShadaraSad Ashok ViharAsv Descriptive Statistics of RSPM in Delhi

Distribution of RSPM (PM 10) levels in Chennai

Descriptive Statistics of Air Pollutants - Ludhiana StatisticsRSPMSO2NOx Overall Overall Overall N Mean S.D Minimum Maximum

Descriptive Statistics of Mortality-Ludhiana (2004) Total deaths:9322 Number of male deaths:6103 Number of female deaths:3219 Mean SDMinMax

Mortality data descriptives in Chennai

Core model Generalized Additive Model (GAM) with penalized and natural spline smoothers in R. Quasi-Poisson function with all natural causes as the dependent variables. Smoothers for time, temp,RH Day of week terms (i.e, dichotomous variables for each day of the week from Monday through Saturday). Exposure at single-day lags of 0 to 3 days & a cumulative two-day average of lags considered.

Summary results of model -Delhi Exposure variable N of observations P %  per Δ µg/m3 in Pm10 IHC ITO AV IHC< AV< CEN CEN< Lag Lag1< Lag Lag2<

Model result-Chennai GAM output from all single monitors GAM output from the best single monitors

Gam Model Results – Ludhiana (After Selecting Base Model, Natural Spline (6,4,4)) Lag Effects %  per Δ µg/m3 in Pm10 P - value Lag Lag Lag * Lag * * RSPM Effects are significant with lag2 and lag3

Work in progress Sensitivity analyses for different exposure series Multi-pollutant models Population (Exposure) weighted assignment for individual monitors using spatial models Gender specific analysis Cause specific impact estimations

Conclusions The data processing steps for the pollutant and mortality data could be completed as envisaged as a result of co-operation and access to raw data provided by the local bodies in-charge of routine data collection. The results of models developed thus far show that the estimates for impacts of PM10 (an approximately 0.2% to 0.6% increase in all-cause mortality for every 10  g/m3 of exposure) are in the range reported by other on-going PAPA studies and earlier studies reported in North America and Europe (using similar statistical methods). More sophisticated analyses including use of spatial models to estimate population- weighted exposures from individual monitors and use of EM algorithms to address missing data and Poisson auto-correlation are being undertaken to refine these initial estimates. Data quality issues may limit development of multi-pollutant models and differential cause specific estimates. Consistency of results obtained using similar methodological approaches between Indian cities and other Asian cities indicate that routinely collected pollutant and mortality data may be reliably used in time-series analyses of air pollution related health impacts. However, significant challenges remain in making clean data readily accessible to environmental health researchers.

Thank you