Spatial and Syndromic Surveillance for Public Health (eBook)

Andrew Lawson, Department of Epidemiology and Biostatistics, University of South Carolina, USA Andrew has published many papers in leading journals, and a number of books on spatial statistics, including four for Wiley. Ken Kleinman, Department of Ambulatory Care and Prevention, Harvard Medical School, Boston, USA Ken is an epidemiologist who specializes in disease surveillance, and has recently worked on projects modeling the spread of anthrax following a potential terrorist attack.

Spatial and Syndromic Surveillance for Public Health 3
Contents 7
Preface 13
List of Contributors 15
1 Introduction: Spatial and syndromic surveillance for public health 17
1.1 What is public health surveillance? 17
1.1.1 Spatial surveillance 17
1.1.2 Syndromic surveillance 18
1.2 The increased importance of public health surveillance 18
1.3 Geographic information, cluster detection and spatial surveillance 19
1.4 Surveillance and screening 20
1.5 Overview of process control and mapping 21
1.5.1 Process control methodology 21
1.5.2 The analysis of maps and surveillance 22
1.6 The purpose of this book 23
1.6.1 Statistical surveillance and methodological development in a public health context 23
1.6.2 The statistician’s role in surveillance 23
1.7 The contents of this book 24
Part I Introduction to Temporal Surveillance 27
2 Overview of temporal surveillance 29
2.1 Introduction 29
2.1.1 Surveillance systems 29
2.1.2 Surveillance attributes 30
2.1.3 Early detection of unusual health events 31
2.2 Statistical methods 32
2.2.1 Historical limits method 32
2.2.2 Process control charts 35
2.2.3 Time-series analysis 38
2.3 Conclusion 44
3 Optimal surveillance 47
3.1 Introduction 47
3.2 Optimality for a fixed sample and for on-line surveillance 49
3.3 Specification of the statistical surveillance problem 50
3.4 Evaluations of systems for surveillance 51
3.4.1 Measures for a fixed sample situation adopted for surveillance 52
3.4.2 False alarms 53
3.4.3 Delay of the alarm 53
3.4.4 Predictive value 55
3.5 Optimality criteria 55
3.5.1 Minimal expected delay 55
3.5.2 Minimax optimality 56
3.5.3 Average run length 56
3.6 Optimality of some standard methods 57
3.6.1 The likelihood ratio method 57
3.6.2 The Shewhart method 59
3.6.3 The CUSUM method 60
3.6.4 Moving average and window-based methods 62
3.6.5 Exponentially weighted moving average methods 63
3.7 Special aspects of optimality for surveillance of public health 64
3.7.1 Gradual changes during outbreaks of diseases 64
3.7.2 Change between unknown incidences 65
3.7.3 Spatial and other multivariate surveillance 66
3.8 Concluding remarks 67
Acknowledgment 68
Part II Basic Methods for Spatial and Syndromic Surveillance 69
4 Spatial and spatio-temporal disease analysis 71
4.1 Introduction 71
4.2 Disease mapping and map reconstruction 72
4.3 Disease map restoration 73
4.3.1 Simple statistical representations 73
4.3.2 Basic models 78
4.3.3 A simple overdispersion model 82
4.3.4 Advanced Bayesian models 83
4.4 Residuals and goodness of fit 84
4.5 Spatio-temporal analysis 87
4.6 Surveillance issues 91
5 Generalized linear models and generalized linear mixed models for small-area surveillance 93
5.1 Introduction 93
5.2 Surveillance using small-area modeling 94
5.2.1 Example 94
5.2.2 Using the model results 95
5.3 Alternate model formulations 96
5.3.1 Fixed effects logistic regression 96
5.3.2 Poisson regression models 97
5.4 Practical variations 98
5.5 Data 99
5.5.1 Developing and defining syndromes 100
5.6 Evaluation 101
5.6.1 Fixed and random effects monthly models 101
5.6.2 Daily versus monthly modeling 108
5.7 Conclusion 109
6 Spatial surveillance and cumulative sum methods 111
6.1 Introduction 111
6.2 Statistical process control 112
6.2.1 Shewhart charts 112
6.2.2 Cumulative sum charts 112
6.3 Cumulative sum methods for spatial surveillance 121
6.3.1 Maintaining a cumulative sum chart for each region 121
6.3.2 Maintaining cumulative sum charts for local neighborhoods around each region 122
6.3.3 Cumulative sum charts for global spatial statistics 126
6.3.4 Multivariate cumulative sum methods 127
6.4 Summary and discussion 128
Acknowledgments 129
Appendix 129
7 Scan statistics for geographical disease surveillance: an overview 131
7.1 Introduction 131
7.1.1 Geographical disease surveillance 131
7.1.2 Tests for spatial randomness 133
7.1.3 Scan statistics 133
7.2 Scan statistics for geographical disease surveillance 135
7.2.1 Probability models 135
7.2.2 Likelihood ratio test 136
7.2.3 Scanning window 137
7.2.4 Adjustments 138
7.3 Secondary clusters 139
7.4 Null and alternative hypotheses 140
7.4.1 The null hypothesis 140
7.4.2 Spatial autocorrelation 140
7.4.3 The alternative hypothesis 141
7.5 Power 142
7.6 Visualizing the detected clusters 142
7.7 A Sample of applications 143
7.7.1 Cancer surveillance 143
7.7.2 Infectious diseases 145
7.7.3 Other human diseases 145
7.7.4 Veterinary medicine 146
7.7.5 Plant diseases 147
7.8 Software 147
Acknowledgment 147
8 Distance-based methods for spatial and spatio-temporal surveillance 149
8.1 Introduction 149
8.2 Motivation 150
8.3 Distance-based statistics for surveillance 152
8.3.1 MEET statistic 152
8.3.2 The interpoint distribution function and the M statistic 153
8.4 Spatio-temporal surveillance: an example 157
8.4.1 Temporal component 158
8.4.2 Bivariate test statistic 160
8.4.3 Power calculations 161
8.5 Locating clusters 163
8.6 Conclusion 167
Acknowledgments 168
9 Multivariate surveillance 169
9.1 Introduction 169
9.2 Specifications 170
9.3 Approaches to multivariate surveillance 171
9.3.1 Reduction of dimensionality 171
9.3.2 Reduction to one scalar statistic for each time 172
9.3.3 Parallel surveillance 173
9.3.4 Vector accumulation methods 176
9.3.5 Simultaneous solution 178
9.4 Evaluation of the properties of multivariate surveillance methods 178
9.5 Concluding discussion 180
Part III Database Mining and Bayesian Methods 183
10 Bayesian network approaches to detection 185
10.1 Introduction 185
10.2 Association rules 186
10.3 WSARE 188
10.3.1 Creating the baseline distribution 188
10.3.2 Finding the best one-component rule 190
10.3.3 Two-component rules 190
10.3.4 Obtaining the p-value for each rule 192
10.4 Evaluation 193
10.4.1 The simulator 193
10.4.2 Algorithms 195
10.5 Results 197
10.6 Conclusion 202
11 Efficient scan statistic computations 205
11.1 Introduction 205
11.1.1 The spatial scan statistic 207
11.1.2 Randomization testing 207
11.1.3 The naive approach 208
11.2 Overlap-multiresolution partitioning 209
11.2.1 Score bounds 212
11.3 Results 213
11.3.1 Comparison to SaTScan 216
11.4 Conclusions and future work 218
12 Bayesian data mining for health surveillance 219
12.1 Introduction 219
12.2 Probabilistic graphical models 220
12.3 Hidden Markov models for surveillance: illustrative examples 222
12.4 Hidden Markov models for surveillance: further exploration 226
12.4.1 Beyond normality 226
12.4.2 How many hidden states? 228
12.4.3 Label switching 228
12.4.4 Multivariate extensions 228
12.5 Random observation time hidden Markov models 230
12.6 Interpretation of hidden Markov models for surveillance 236
12.7 Discussion 237
Acknowledgments 237
13 Advanced modeling for surveillance: clustering of relative risk changes 239
13.1 Introduction 239
13.2 Cluster concepts 239
13.3 Cluster modeling 240
13.3.1 Spatial modeling of case event data 240
13.3.2 Spatial modeling of count data 246
13.3.3 Spatio-temporal modeling of case and count data 247
13.4 Syndromic cluster assessment 251
13.4.1 The Bayesian posterior distribution 252
13.5 Bayesian version of the optimal surveillance alarm function 255
13.5.1 Clustering and ƒ(x(u)|µ) 255
13.5.2 A simple real-time biohazard model 256
13.6 Computational issues 258
13.7 Conclusions and future directions 259
References 261
Index 283

"... gives a sound introduction to a range of methods which are suitable for health surveillance purposes." (JRSSA: 169;3)