Building on the Ontario Health and Environment Spatial Surveillance (OHEIS) Project Project Eric J. Holowaty, Professor, Dalla Lana School of Public Health, University of Toronto. May 15, 2011. 1 Geographic Information Systems and Population Health interest and use of GIS computing power and software availability georeferenced data 2 georeferenced data rapid hazard appraisal and more granularity in community health profiling advances in spatial analysis 2 How can GIS help Public Health? Research, Surveillance and Planning Hypothesis gen./testing maps, correlations, clusters Spatial and ST models of disease risk Service planning and optimisation Making predictions e.g. Health Impact Asst Spatial Decision Support Systems Spatial Decision Support Systems Infrastructure roads, towns, HC services/avail. Census population statistics; socio-demographics Emergency/Pandemic Response Systems 911 services Disease/event registers, incl. infectious diseases 3 Deprivation Census geography Area classification Popn counts/estimates Cancer Deaths Births Hospital admissions Congenital anomalies Stillbirths and perinatal deaths Health event data Oracle/Access Oracle/Access, GIS Postcode EA/DA, CT Link data Postcode EA/DA Integrated GIS Name Street address Postal code Municipality Historic Popn Res. Files Census data (1986, 1991, 1996, 2001, 2006) 4 Environmental and geographic data Roads, Railways, Rivers Road Traffic Chemical Release Radiation release or exposure Locations of refineries, incinerators, dumps Water supplies Oracle/Access, GIS Oracle/Access Postcode EA/DA CT, CMA/CA CSD, CD boundaries airborne waterborne foodborne soil Pathway Analysis and Exposure Modeling Smoking Screening Confounders Tools Methods RIF ArcGIS R WinBUGS 4 Estimating Disease Risk in Small Areas Small areas with counts of 0 or 1 produce highly variable/implausible SIRs. Spatial dependence : areas close together have similar risks. Detecting areas at truly higher risk: 5 Detecting areas at truly higher risk: Must allow for uncertainty due to low counts; Use spatial dependence to pool info. from neighbouring areas. One solution: Hierarchical random effects models : actual risk is unobserved, and case counts are Poisson distributed. 5 Deterministic model Stochastic uncertainty Fixed vs. Random Effects 6 6 And now, with some trepidation: a Bayesian Mapping Model! From Besag, York and Mollie, 1991. 7 Markov Chain Monte Carlo Sampling How to solve such a complex hierarchical model? Area of Circle (est.) = Area of Square X dots inside circle all dots (20X20) X 39 = 312 10 cm. 8 From: Brighton Statistical and Data Services. Accessed at: http://www.brighton-webs.co.uk/montecarlo/concept.asp (20X20) X 39 = 312 50 Area of Circle (exact) = r 2 (3.1416) X 10 2 = 314.16 10 cm. 8 What if the formula for the area of a circle was unknown? Consider 10 simulations of 10,000 dots each 9 Consider 100 simulations of 10,000 dots each 9 Fundamental Spatial Surveillance Questions L What is the risk of PH outcomes among residents of a particular geographic area or neighbourhood? L Are there areas of unusually high (or low) risk? 10 L Are there areas of unusually high (or low) risk? L Is the observed pattern of risk similar to the pattern of known RFs and other antecedent determinants? From Pickle, 2002. 10 CDNAME Rate SIR Obs Algoma District 73.29 1.21 571 Brant County 65.74 1.08 443 Bruce County 57.78 0.96 253 Chatham-Kent Division 62.06 1.02 386 Cochrane District 84.52 1.41 370 Dufferin County 59.93 0.98 128 Durham Regional Municipality 62.11 1.03 1329 Elgin County 67.67 1.11 299 Essex County 69.67 1.15 1330 Frontenac County 70.97 1.17 566 Greater Sudbury Division 81.78 1.36 714 Grey County 55.94 0.91 341 Middlesex County 61.45 1.02 1283 Muskoka District Municipality 65.51 1.08 246 Niagara Regional Municipality 62.25 1.03 1646 Nipissing District 75.70 1.25 367 Northumberland County 76.79 1.27 404 Ottawa Division 63.80 1.06 2350 Oxford County 54.75 0.90 311 Parry Sound District 70.43 1.15 208 Peel Regional Municipality 46.97 0.76 1710 Perth County 58.45 0.96 240 Peterborough County 69.16 1.14 589 Tabular Summary of Rates and SIRs Grey County 55.94 0.91 341 Haldimand-Norfolk RM 68.46 1.13 421 Haliburton County 74.45 1.19 101 Halton Regional Municipality 49.39 0.81 919 Hamilton Division 64.52 1.07 1767 Hastings County 83.38 1.38 657 Huron County 57.20 0.93 222 Kawartha Lakes Division 74.41 1.21 365 Kenora District 69.76 1.14 195 Lambton County 73.75 1.22 579 Lanark County 75.03 1.24 281 Leeds and Grenville United Cnt. 74.42 1.23 462 Lennox and Addington County 67.88 1.13 164 Manitoulin District 52.18 0.89 45 Peterborough County 69.16 1.14 589 Prescott and Russell United Counties 85.54 1.43 314 Prince Edward Division 63.00 1.02 120 Rainy River District 74.36 1.23 98 Renfrew County 71.47 1.18 417 Simcoe County 69.92 1.16 1399 Stormont, Dundas and Glengarry 84.88 1.39 571 Sudbury District 71.99 1.14 92 Thunder Bay District 74.07 1.23 622 Timiskaming District 102.7 1.71 231 Toronto Division 50.23 0.83 6626 Waterloo Regional Municipality 52.77 0.88 1063 Wellington County 52.52 0.87 493 York Regional Municipality 47.07 0.76 1405 11 Disease Mapping Hamilton Census Division 12 13 Hamilton Steel & Iron Industry 14 14 Disease Mapping: Female Lung Cancer in Hamilton 15 15 Disease Mapping: Male Lung Cancer in Hamilton 16 16 SENES Approach to Dispersion Model 14 17 Risk Analysis: Modeled Concentration Exposure Bands Males - Unadjusted and Adjusted 1.75 Females - Unadjusted and Adjusted 1 1.25 1.5 1.75 2 S I RFemales - unadjusted Females - adjusted 16 0.5 0.75 1 1.25 1.5 0 0.02 0.04 0.06 0.08 0.1 Average Concentration (ug/m 3 ) S I R Males - unadjusted Males - adjusted 0.5 0.75 0 0.02 0.04 0.06 0.08 0.1 Average Concentration (ug/m 3 ) Homogeneity 2 p value Linearity 2 p value Homogeneity 2 p value Linearity 2 p value Males unadjusted 79.72 <0.0001 75.06 <0.0001 77.61 <0.0001 66.44 <0.0001 adjusted for QAIPPE 26.63 <0.0001 21.31 <0.0001 21.73 <0.0001 17.41 0.0006 Females unadjusted 67.79 <0.0001 65.40 <0.0001 54.51 <0.0001 31.26 <0.0001 adjusted for QAIPPE 36.38 <0.0001 33.45 <0.0001 22.47 <0.0001 5.85 0.02 Dispersion Model Distance as a Proxy for Exposure 18 19 19 20 20 Thyroid cancer in women living in the Greater Toronto area 2004-2008 Raw SIRs 998 CTs Overall SIR=1.18 4705 cases 21 Thyroid cancer in women living in the Greater Toronto area 2004-2008 Smoothed SIRs SaTScan cluster 243 CTs Overall SIR=1.46 1582 cases 22 Systematic approach to identifying hot spots 23 24 WDG Lung Cancer in males 1999-2003 25 WDG Lung Cancer in males 1999-2003 26 WDG Prostate cancer 1999-2003 27 WDG Prostate cancer 1999-2003 28 Special challenges L Accuracy, granularity and completeness of exposure, health and population data, and boundary files L Geocoding, i.e., accurately assigning spatial coordinates to location info. L Current place of residence may not be good proxy for exposure exposure L Problems adjusting for known confounders L Necessity of using aggregated counts L Scale/zone translation problems (MAUP) L Spatial autocorrelation L Data access and confidentiality restrictions 29 Future enhancements Quality, timeliness and accessibility of georeferenced data Faster MCMC simulations Spatio-temporal analysis Modeling small area estimates of RF and service utilization Comparing similar maps beyond visualization More flexibility within RIF 30 L Effective stakeholder engagement Preparatory discussions/training; user readiness surveys L Policy/legs./regs. changes may be required L Significant up front work in data Moving Forward - Key Issues enhancement & harmonization L Impt distinction between levels of complexity - rapid surveillance vs. designed research L Varying levels of stakeholder readiness technical infrastructure, data sharing, data discovery, stat. methods 31 Resources The GIS Primer web-based http://www.innovativegis.com/basis/primer/primer.html Health and Environment Information Systems for Exposure and Disease Mapping and Risk Assessment mini-monographs Jarup et al. Environmental Health Perspectives. June 2004. Vol. 112: 995-1045. Elliott et al. Environmental Health Perspectives. Aug 2008. Vol. 116: 1098-1130. Vol. 116: 1098-1130. GIS and Public Health Cromley EK and McLafferty SL. Guildford Press. 2002. Feasibility and utility of mapping disease risk at the neighbourhood level within a Canadian public health unit Holowaty et al. Inter. Journal Health Geogr. May 10, 2010. http://www.ij-healthgeographics.com/content/9/1/21 32 Resources (contd) Spatial Epidemiology : Methods and Applications Elliott P. et al. Oxford University Press. 2000. Applied Spatial Statistics for Public Health Waller LA and Gotway CA. Wiley Interscience. 2004. Geographic Information Systems and Public Health Richards TB et al. Public Health Reports Vol.114.1999. http://www.healthgis-li.com/library/phr/phr.htm Public Health and GIS Public Health and GIS Rushton G et al. Annual Review of PH. Vol.24.2003. Putting People on the Map : Protecting Confidentiality with Linked Socio-Spatial Data Gutmann MP et al. National Research Council. 2007. http://books.napedu/catalog/11865.html 33 For more information about the RIF, contact: Dr. Judy Qualters, Chief, EPHTN Branch, CDC Ms. Gonza Namulanda, Informatics, EPHTN, CDC E-mail: EPHT@cdc.gov Website: www.cdc.gov/nceh/tracking 34