Liselotte van Asten, Cees van den Wijngaard, Carel Harmsen, Frederika Dijkstra, Wilfrid van Pelt, Wi

1
Liselotte van Asten, Cees van den Wijngaard,
Carel Harmsen, Frederika Dijkstra, Wilfrid van
Pelt, Wim van der Hoek, Marianne van der Sande,
Marion Koopmans

• Syndromic surveillance with death data
• a pilot study in the Netherlands

2
Introduction

• MORTALITY SURVEILLANCE INFECTIOUS DISEASE
• Increase capacity to deal with
• Pandemic influenza (preparedness)
• Other unexpected disease events
• Mortality surveillance could offer information
  for
• PH action / decision
• Estimating the impact - tracking stage / severity
  / burden of an outbreak
• Rumour reduction
• Possibly a robust data source in time of crisis
• Currently For infectious disease surveillance
  there is no prospective mortality surveillance
  system in place

3
Introduction

• OBJECTIVE OF PILOT
• To evaluate the potential use of mortality data
  in the Netherlands for real-time surveillance of
  infectious disease events
• RESEARCH QUESTIONS
• Can we set up a working relation with Statistics
  Netherlands?
• Can we receive retrospective prospective data?
• Can we set up a baseline and prediction limits?
• Can death data detect infectious disease
  outbreaks?
• Do peaks in death coincide with peaks of inf.
  diseases?
• Do death data reflect trends in infectious
  disease?
• What can we learn from retrospective data? What
  are the correlations with other inf. Dis.
  indicators?

4
Methods

• CRUDE MORTALITY
• Nr. of deaths in a given time period
  irrespective of death cause.
• DUTCH SYSTEM OF DEATH NOTIFICATION
• Physicians report deaths to local municipality
• Municipality transmits data electronically to
  Statistics Netherlands (daily)
• Statistics Netherlands publishes monthly death
  counts on their website
• After 4 days gt50 notified.
• After 1 week (7 day delay) 93 is notified to
  Stat. NL.

5
Methods

• Together with Statistics Netherlands
• Agreement to start a prospective pilot (feb
  2008)
• Monthly excel file by email
• Daily nr of reported deaths (by date of death)
• Stratified by age and region
• 5 agegroups (0-4, 5-19, 20-64, 65-79, 80 )
• 4 regions N, E, W, S
• Receipt of retrospective data 2000-2006
• With information on registration delay
• Used for determination of baseline mortality
  levels and prediction intervals (training data)

6
Methods

• ILI data
• Weekly counts (received weekly in week 40-20
  prospectively)
• Sentinel system of general practitioners
• Appr. Coverage 2
• Historical data available

7
Methods

• BASELINE AND THRESHOLD
• Serfling regression model on the historical data
• 7 years of training data (2000-2006)
• 2007 unfortunately not yet included
• Then compare baseline and threshold with
  prospective data (start feb 2008)
• Model
• of deaths intercept linear trend
  seasonal trend
• Threshold 95 CI
• Modelled daily weekly deaths
• Large historical epidemics raise your baseline
  and threshold.
• Need to adjust?
• Option explored leaving out 10 of most extreme
  values

8
Results

• Population size 16 million
• Approximately (in winter seasons)
• 400 daily deaths
• 3000 weekly deaths

9

• I
• Historical data
• Clear peaks in winter mainly coinciding with
  respiratory seasons
• Slight decreasing trend in the number of deaths

10

• Historical data serfling model (baseline and
  threshold)
• Sign. Decreasing linear trend, sign. seasonality

11

• I
• Historical prospective data
• 2008 So far hardly any peaks in mortality (data
  until mid nov)

12

• I
• ILI surveillance (weekly incidence)
• Respiratory wise 2008 milder year than usual

13

• I
• 2008
• Baseline in black. Threshold in red.
• 4 threshold exceedences in (june, july, oct)

14

• I
• Deaths by age

15

• I
• Deaths by region

16

• I
• Aggregated by week mild year

17

• I
• Narrowing the threshold level
• (Modeling after leaving out 10 most extreme
  historical values in the training data)

18
Results

• Retrospective data
• Correlations and time lag between ILI and
  mortality trends
• Lags nr of weeks that mortality is lagging
  behind ILI
• High overall correlation
• Mortality lagging 1-2 weeks behind ILI
• However timelag differs by year

19
Conclusion

• Pilot shows us
• Can we set up a working relation with Statistics
  Netherlands? - yes
• Can we receive retrospective prospective data?
  - yes
• Can we set up a baseline and prediction limits? -
  yes
• Can death data detect infectious disease
  outbreaks?
• Probably, but 2008 was a very mild year.
• Can trends in infectious disease help explain
  mortality trends?
• Probably yes good correlation with ILI trends
  (but deaths lagging behind at different time lags
  by year)

20
Discussion

• A PROSPECTIVE SYSTEM STILL VERY MUCH UNDER
  CONSTRUCTION
• A lot of research questions still ongoing
• Should we intensify and automate frequency of
  receipt of mortality data / analyses?
• Or should we make a system with data requested
  only when something occurs?
• Should we look at daily or weekly nr of deaths?
• Narrower tresholds?
• Include 2007 data into the training data (the
  calculation of a baseline)
• Quantify the number of detected excess deaths
  (observed expected) depending on how high you
  set your baseline.
• What other data sources to compare mortality with
  (prospectively)?
• ILI, laboratory pathogen counts, webbased
  selfreport of ILI, meteorological data