1.  Statistical approaches for detecting unexplained clusters of disease.  Spatial Aggregation Thomas Talbot New York State Department of Health Environmental Health Surveillance Section Albany School of Public Health GIS & Public Health Class March 3, 2009

2. Cluster A number of similar things grouped closely together Webster’s Dictionary Unexplained concentrations of health events in space and/or time Public Health Definition

3. Occupation Sex, Age Socioeconomic class Behavior (smoking) Race Time Space

4. Spatial Autocorrelation Negative autocorrelation “Everything is related to everything else, but near things are more related than distant things.” - Tobler’s first law of geography Positive autocorrelation

5. Moran’s I A test for spatial autocorrelation in disease rates. Nearby areas tend to have similar rates of disease. Moran I is greater than 1, positive spatial autocorrelation. When nearby areas are dissimilar Moran I is less than 1, negative spatial autocorrelation.

6. Detecting Clusters Consider scale Consider zone Control for multiple testing

7. Talbot

14. Cluster Questions Does a disease cluster in space? Does a disease cluster in both time and space? Where is the most likely cluster? Where is the most likely cluster in both time and space?

15. More Cluster Questions At what geographic or population scale do clusters appear? Are cases of disease clustered in areas of high exposure?

16. Nearest Neighbor Analysis Cuzick & Edwards Method Count the the number of cases whose nearest neighbors are cases and not controls. When cases are clustered the nearest neighbor to a case will tend to be another case, and the test statistic will be large.

17. Nearest Neighbor Analyses

18. Advantages Accounts for the geographic variation in population density Accounts for confounders through judicious selection of controls Can detect clustering with many small clusters

19. Disadvantages Must have spatial locations of cases & controls Doesn’t show location of the clusters

20. Spatial Scan Statistic Martin Kulldorff Determines the location with elevated rate that is statistically significant. Adjust for multiple testing of the many possible locations and area sizes of clusters. Uses Monte Carlo testing techniques

29. The Space-Time Scan Statistic Cylindrical window with a circular geographic base and a height corresponding to time. Cylindrical window is moved in space and time. P value for each cylinder calculated.

30. Knox Method test for space-time interaction When space-time interaction is present cases near in space will be near in time, the test statistic will be large. Test statistic: The number of pairs of cases that are near in both time and space.

31. Focal tests for clustering Cross sectional or cohort approach: Is there a higher rate of disease in populations living in contaminated areas compared to populations in uncontaminated areas? (Relative risk) Case/control approach: Are there more cases than controls living in a contaminated area? (Odds ratio)

32. Focal Case-Control Design CaseControl 250 m. 500 m.

33. Regression Analysis Control for know risk factors before analyzing for spatial clustering Analyze for unexplained clusters. Follow-up in areas with large regression residuals with traditional case-control or cohort studies Obtain additional risk factor data to account for the large residuals.

37. At what geographic or population scale do clusters appear? Multiresolution mapping.

38. A cluster of cases in a neighborhood provides a different epidemiological meaning then a cluster of cases across several adjacent counties. Results can change dramatically with the scale of analysis.

39. 1995-1999

43. Interactive Selections by rate, population and p value

44. References Talbot TO, Kulldorff M, Forand SP, and Haley VB. Evaluation of Spatial Filters to Create Smoothed Maps of Health Data. Statistics in Medicine. 2000, 19:2451-2467 Forand SP, Talbot TO, Druschel C, Cross PK. Data Quality and the Spatial Analysis of Disease Rates: Congenital Malformations in New York. 2002. Health and Place. 2002, 8:191-199 Haley VB, Talbot TO. Geographic Analysis of Blood Lead Levels in New York State Children Born 1994-1997. Environmental Health Perspectives 2004, 112(15):1577-1582 Kuldorff M, National Cancer Institute. SatScan User Guide www.satscan.org

46. Health data can be shown at different geographic scales Residential address Census blocks, and tracts Towns Counties State

47. Concerns about release of small area data Risk of disclosure of confidential information. Rates of disease are unreliable due to small numbers.

48. Rate maps with small numbers provide very little information. http://www.nyhealth.gov/statistics/ny_asthma/hosp/zipcode/hamil_t2.htm http://www.nyhealth.gov/statistics/ny_asthma/hosp/zipcode/pdf/hamil_m2.pdf

49. Disclosure of confidential information Census Blocks

50. Smoothed or Aggregated Count & Rate Maps Protect Confidentiality so data can be shared. Reduce random fluctuations in rates due to small numbers.

51. Smoothed Rate Maps Borrow data from neighboring areas to provide more stable rates of disease. –Shareware tools available –Empirical or Hierarchal Bayesian approaches –Adaptive Spatial Filters –Head banging –etc.

53. from Talbot et al., Statistics in Medicine, 2000

54. Problems with smoothing Does not provide counts & rates for defined geographic areas. Not clear how to link risk factor data with smoothed health data. Methods are sometimes difficult to understand - “black boxes” Does not meet requirements of some recent New York policies & legislation.

55. Environmental Facilities & Cancer Incidence Map Law, 2008 § 3-0317 Plot cancer cases by census block, except in cases where such plotting could make it possible to identify any cancer patient. Census blocks shall be aggregated to protect confidentiality.

56. Environmental Justice & Permitting NYSDEC Commissioner Policy 29 Incorporate existing human health data into the environmental review process. Data will be made available at a fine geographic scale (ZIP code or ZIP Code Groups).

57. Aggregated Count or Rate Maps Merge small areas with neighboring areas to provide more stable rates of disease and/or protect confidentiality. –Aggregation can be done manually. –Existing automated tools were difficult to use.

58. Original ZIP Codes 3 Years Low Birth Weight Incidence Ratios

59. Aggregated to 250 Births per ZIP Code Group

60. Goal Aggregate small areas into larger ones. User decides how much aggregation is needed. Works with various levels of geography. – census blocks, tracts, towns, ZIP codes etc. – can nest one level of geography in another Uses software which is readily available in NYSDOH (SAS)‏ Can output results for use in mapping programs. Our Tool Requirements

61. Aggregation Tool C14 B23 A21 RegionCases Original Block Data † Regions SAS Tool † Simulated data 6 8 15 1 0 4 3 11 10 CasesBlock 103202/2002 103202/2001 014500/3010 014500/3009 014500/3008 014500/3007 014500/3005 122300/2005 122300/2004 6 8 15 1 0 4 3 11 10 CasesRegionBlock C103202/2002 C103202/2001 B014500/3010 B014500/3009 B014500/3008 B014500/3007 B014500/3005 A122300/2005 A122300/2004

62. Aggregation Process Populated blocks with the fewest cases are merged first. If there is a tie the program starts with the block with the fewest neighbors. Selected block then is merged with the closest neighbor in the same census block group. After merging the first block the list of neighbors is updated. Process repeats until all regions have a minimum number of cases –program can also merge to user specified population

63. Special Situations Tool tries to avoid merging blocks in different census areas: –Census block groups –Census tracts ( homogeneous population characteristics). –Counties Tool tries to avoid merging blocks across major water bodies e.g. Finger lakes, Hudson River, Atlantic Ocean

64. Water

65. 9 Cases 98Population † Simulated data

66. 9 Cases 98Population † Simulated data

67. 9 Cases 98Population † Simulated data

68. 9 Cases 98Population † Simulated data

69. 9 Cases 98Population † Simulated data

70. 9 Cases 98Population † Simulated data

71. 9 Cases 98Population † Simulated data

72. 9 Cases 98Population † Simulated data

73. 9 Cases 98Population † Simulated data

74. 9 Cases 98Population † Simulated data

75. 9 Cases 98Population † Simulated data

76. 9 Cases 98Population † Simulated data

77. 9 Cases 98Population † Simulated data

78. 9 Cases 98Population † Simulated data

79. 9 Cases 98Population † Simulated data

80. 9 Cases 98Population † Simulated data

81. 9 Cases 98Population † Simulated data

82. New York State Descriptive Statistics Year 2000 populated census blocks 14741 Median Census blocks 3820101 Median cases 1,66788247784 Average Population 11,38121,52539,748225,167 Number 24 cases12 cases6 cases Original Census Blocks Statistic (calculated using populated regions only)‏ New Regions: Level of Aggregation NY number of cases 470,000 NY population 18,976,457

83. Performance Measures Compactness Homogeneity with respect to demographic factors (measured as index of dissimilarity) Similar population sizes. Number of aggregated areas. Aggregated zones are completely contained within larger areas. –e.g. blocks aggregation areas contained within tracts

85. Index of dissimilarity the percentage of one group that would have to move to a different area in order to have a even distribution b i = the minority population of the ith area, e.g. census tract B = the total minority population of the large geographic entity for which the index of dissimilarity is being calculated. w i = the non-minority population of the ith area W = the total non-minority population of the large geographic entity for which the index of dissimilarity is being calculated. Wikipedia

86. The End