1	Visualization and Analytic Methods for the Tracking of Birth Outcomes and Traffic Exposures
2	Overview Needs of Visualization for Tracking Mapping Disease Rates Failure of mapping in discrete areas Density estimation – Strengths and Weaknesses Modeling of Traffic-Exhaust Pollution Methods: Cost-Benefit Analysis
3	Needs of Data Visualization for Tracking We need a system that is continuous and ongoing We need web-based tools that allow public access and ability to interface with data We need to preserve data confidentiality and privacy The system is not complete until those who need information Know the information exists Know where to find it Know what it’s good for Are able to access and interpret it
4	Mapping Disease Rates Failure of mapping rates in discrete areas Sample size problem: law of small numbers Instability of rates with small denominators As areas get smaller, variability increases Visualization fails Political boundaries change over time (e.g. ZIP codes)
5	Density Estimation: Strengths Methods of Openshaw, Rushton Produce continuous surface of rates Preserves data confidentiality More accurately reflect reality
6	Data Restrictions Restricted births to: completed birthweight geocoded address downtown San Diego and nearby areas compatible birthweight and gestational age 16,385 births 1980 24,274 births 1990
7	Spatial Distribution Method (1) Generated Uniform Grid 0.5 miles apart (spatial filters)
8
9	Spatial Distribution Method (2) Identified all births within a 0.5 mile spatial filter (min. of 40 births to compute rate) 3) Compute LBW rates and made contour maps
10
11
12
13
14	Representation of Data Example: LBW Birth-rate Map
15	Statistical Significance (Monte Carlo Simulations) Assign each birth equal probability of being a LBW birth Assign a random number (1-1000) for each birth Classify each birth as LBW or non-LBW Compute LBW rate at each grid point 1000 times
16	Statistical Significance (Monte Carlo Simulations) Compute the % of simulated rates which are less than the observed rate. Make map of statistical significance
17
18	Density Estimation: Weaknesses Need for geocoded denominator Need for sensitivity analysis: size of filter needs to change in relation of density of health events? Spatial autocorrelation an issue in analysis: Spatial regression
19	Modeling of Traffic Exhaust Pollution Proximity Analyses Dispersion Models Land-Use Regression Integrated-Meterological Models
20
21	Proximity Analysis Strengths: Easy to use (adapt with Gaussian weights) Distance of residence to road correlates well with personal and ambient NO₂ monitoring (Rijnders,et al 2001) Weaknesses: Exposure misclassification likely without wind direction data Does not model actual level of pollutants
22
23	Dispersion Models (e.g. CALINE) Strengths: More accurately measuring dispersal of pollutants Model actual pollutant levels Weaknesses: Is gaussian plume realistic model?
24
25	Land-use regression (e.g. Briggs) Strengths: Use land use, met data, DEMs, traffic to predict pollutant concentrations Easily obtained data Weaknesses: Need enough monitoring locations for callibration/validation Need to replicate in new areas; models developed in one geographic location may not be predictive in other areas.
26	Integrated Meterological-Emission Models (e.g. ADMS-Urban, Cal-PUFF) ADMS-Urban: Strengths: Incorporates mobile, point, and area sources Ability to model gridded emissions and terrain simultaneously model 10,050 receptor points boundary layer effects and dispersal behaviors over complex terrain photochemistry. seamlessly integrated with ArcView GIS
27	Integrated Meterological-Emission Models (e.g. ADMS-Urban, Cal-PUFF) ADMS-Urban: Weaknesses:: Significant training and expertise necessary Cost Multiple data inputs What is the bang for the buck?
28
29	Incorporating Time Activity/Personal Monitoring into Tracking Cost/sample size makes these activities prohibitive for tracking Subsample analysis? Survey data on time activities Commuting important exposure time: Use of Transportation demand models?
30	Conclusions Integrating density estimation techniques important for visualization and analysis of health tracking data – increased method development necessary Various approaches for traffic-exhaust modeling for tracking – Need to capture most accurate method over the most population at least cost