COMPARING THE UTILITY OF AMBULATORY CARE AND EMERGENCY ROOM DATA FOR DISEASE OUTBREAK DETECTION

1
COMPARING THE UTILITY OF AMBULATORY CARE AND
EMERGENCY ROOM DATA FOR DISEASE OUTBREAK
DETECTION

• Marcelo A. Costa, Martin Kulldorff1, Ken
  Kleinman, W. Katherine Yih, Richard Platt
• Harvard Pilgrim Health Care and Harvard Medical
  School
• Richard Brand
• Kaiser Northern California
• John Hsu
• UCSF

2
Objective

• To compare the cluster-detection ability of
  ambulatory care and emergency department data
  sources, separately and combined
• Population 200,000 in eastern MA
• Syndromes studied
• Respiratory (RESP)
• Lower Gastro Intestinal (LGI)
• Upper Gastro Intestinal (UGI)
• Influenza Like Illness in under 5 (ILI5)
• Influenza Like Illness in 5 to 12 (ILI12)
• Influenza Like Illness in 13 and over (ILI13)

3
Parameters

• Space Time Permutation Analysis with day of the
  week Adjustment
• Settings ambulatory care and emergency
  department
• Period of surveillance 2005 and 2006
• Historical period to use 1 year (2004)
• Episodes (i.e. repeat visits excluded)
• Max temporal scanning window 14 days
• Max. circle size 50 of the population
• Day of the week use as a covariate
• No. Monte Carlo reps 9,999
• RI threshold 365 days

4
Datasets AC, ER and ACER
Note Repeat visits were excluded from ACER
dataset Therefore, the cases are usually less
than AC ER.
5
Proposed Analysis using SaTScan
Symbols
Data Streams
Ambulatory Care (AC)

Emergency Department (ER)

Repeated visits excluded (ACER)
Cluster in AT LEAST one data set (multiv0)
Cluster in BOTH data set SIMULTANEOUSLY (multiv1)
Kulldorff M, Mostashari F, Duczmal L, Yih K,
Kleinman K, Platt R. Multivariate spatial scan
statistics for disease surveillance. Statistics
in Medicine, 2006, in press
6
Respiratory
7
Respiratory Analysis
Synd.
zip
end
type
nzips
clu_cases
rr
llr
RI
3826
12/8/2005
acer
39
12
6.28
11.97
588
NEW
3110
12/8/2005
ac
22
5
22.67
10.83
357
2054
12/8/2005
er
3
2
58.73
6.18
6
3826
12/8/2005
multiv0
39
10.99
67
3826
12/8/2005
multiv1
39
10.99
77
3060
2/24/2006
acer
11
6
17.26
11.44
714
3060
2/24/2006
ac
11
6
19.67
12.18
909
1906
2/24/2006
er
3
2
26.4
4.62
1
RESP
3060
2/24/2006
multiv0
11
12.18
333
3060
2/24/2006
multiv1
11
12.16
500
3060
2/26/2006
acer
11
7
14.54
12.23
1250
3060
2/26/2006
ac
11
6
17.89
11.65
714
2537
2/26/2006
er
16
3
25.92
6.88
11
3060
2/26/2006
multiv0
11
12.59
333
3060
2/26/2006
multiv1
11
12.59
333
8
LGI
9
UGI
10
ILI5
11
ILI12
12
ILI13
13
Summary
RESP
NEW
LGI
NEW
UGI
NEW
ILI5
ILI12
ILI13
NEW
NEW
14
Discussion

• There is evidence that ER signals with RI 365
  are associated with weaker ACER signals. This
  might be due to dilution of the signal by the
  more numerous AC cases.
• AC signals with RI 365 are usually associated
  with stronger ACER signals, although one weaker
  ACER signal was observed for ILI12.
• Most of the detected Multivariate signals overlap
  with AC, ER or ACER.
• Only one new signal was found using the
  Multivariate analysis for LGI.
• Running ACER data analysis is much faster than
  Multivariate analysis.