Epidemic Intelligence: Signals from surveillance systems

1
Epidemic IntelligenceSignals from surveillance
systems

• EpiTrain III Jurmala, August 2006
• Anne Mazick, Statens Serum Institut, Denmark

2
Epidemic intelligence

• All the activities related to
• early identification of potential health threats
• their verification, assessment and investigation
• in order to recommend public health measures to
  control them.

3
(No Transcript)
4
Components core functions
Early warning
Response
5
Indicator vs. Event-based surveillance

• Indicator-based surveillance
• computation of indicators upon which unusual
  disease patterns to investigate are detected
  (number of cases, rates, proportion of strains)
• Event-based surveillance
• the detection of public health events based on
  the capture of ad-hoc unstructured reports issued
  by formal or informal sources.

6
Scope of this presentation

• What surveillance signals are required for EI
• Current communicable disease surveillance
• Additional more sensitive surveillance for new,
  unusual or epidemic disease occurence
• Basic requirements for signal detection
• Use of early warning surveillance systems
• 3 examples

7
Indicator-based early warning systemsObjectives

• to early identify potential health threats -
  alone or in concert with other sources of EI
• in order to recommend public health measures to
  control them
• For new, emerging diseases
• For unusual or epidemic occurence of known
  diseases

8
Indicator-based surveillance

• Identified risks
• Mandatory notifications
• Laboratory surveillance
• Emerging risks
• Syndromic surveillance
• Mortality monitoring
• Health care activity monitoring
• Prescription monitoring
• Non-health care based
• Poison centers
• Behavioural surveillance
• Environmental surveillance
• Veterinary surveillance
• Food safety/Water supply
• Drug post-licensing monitoring

9
Current surveillance systems for communicable
diseases
specificity

• Main attributes
• Representativity
• Completeness
• Predictive positive value

sensitivity
10
From infection to detectionProportion of
infections detected
specificity
50 Shigella notifications (5)
1000 Shigella infections (100)
sensitivity
11
From infection to detectionTimeliness
Analyse Interpret
Signal
time
12
From infection to detectionTimeliness
Urge doctors to report timely
Frequency of reporting Immediately, daily, weekly
Analyse Interpret
Signal
time
13
From infection to detectionTimeliness
Analyse Interpret
Signal
time
14
From infection to detectionTimeliness
Signal
Automated analysis, thresholds
time
15
Potential sources of early signals
Sensitive systems for new, unusual or epidemic
diseases

• Laboratory test volume
• Emergency primary care total patient volume,
  syndromes
• Ambulance dispatches
• Over-the-counter medication sales
• Health care hotline
• School absenteeism

time
16
To detect all events as early as possible

• More sensitive case definitions
• Cave sensitivity ? false alerts ?
• costs of response
• Social and political distress
• Combining information from other sources of
  epidemic intelligence
• Frequency of reporting
• Automated analysis
• Low alert thresholds

17
Current surveillance systems for communicable
diseases

• Important source for EI, but
• Additional systems needed to fulfil all EI
  objectives
• Timeliness
• Sensitivity
• For rapid detection of new, unusual or epidemic
  diseases

18
Principle of signal detection

• To detect excess over the normally expected
• Observed expected system alert
• What are we measuring? Indicators
• What is expected? Need historical data
• Which statistics to use? Depends on disease
• Where to set threshold? Depends on desired
  sensitivity

19
Early warning indicators

• Early warning indicators
• Count
• Rates
• Number of cases/population at risk/time
• Proportional morbidity
• of ILI consultations among all consultations
• Percentage of specific cases
• case fatality ratio
• children under 1 years among measles cases
• of cases with certain strain

20
Statistical methods for early warning

• Depends on the epidemiology of the disease under
  surveillance

21
Thresholds

• Choice of threshold affected by
• Objectives, epidemiology, interventions
• Absolute value
• Count 1 case of AFP
• Rate gt 2 meningo. meningitis/100,000/52 weeks
• Relative increase
• 2 fold increase over 3 weeks
• Statistical cut-off
• gt 90th percentile of historical data
• gt 1.64 standard deviations from historical mean
• Time series analysis

22
Clinical meningitis, Kara Region, Togo 1997
23
Weekly Notification of Food Borne
Illness,National EWARN System, France,1994-1998
24
(No Transcript)
25
(No Transcript)
26
Use of statistics computer tools

• For systematic review of data on a regular basis
• to extract significant changes drowned in routine
  tables of weekly data
• They do not on its own detect and confirm
  outbreaks!
• Epidemiological verification, interpretation and
  assessment ALWAYS required!

27
Tools do not make early warning systems, but
early warning systems need appropriate tools
28
System alert interpretation
Every system alert
Other sources of epidemic intelligence
Media reports Rumours Clinician
concern Laboratories Food agencies Meteorological
data Drug sales/prescription International
networks EWRS
Validate analyse
Signal
Interpret Public health significance?
Alert
No Alert
29
Danish laboratory surveillance systemof enteric
bacterial pathogens

• To detect outbreaks and to analyse long-term
  trends
• Administered by Statens Serum Insitute (SSI)
• Danish reference laboratory
• Receives all salmonella isolates for further
  typing
• Also gets many other strains, including E. coli.,
  for further typing

30
National register of enteric pathogens

• At SSI
• Includes everybody who test positive for a
  bacterial GI infection in Dk.
• Person, county, agent, date of lab receiving
  specimen, travel, no clinical information
• First-positives only
• Mandatory weekly notifications from all 13
  clinical laboratories

31
Outbreak algorithm

• Computer program, which calculates if the current
  number of patients exceeds what we saw at the
  same time of year in the 5 previous years
• Time variable date of lab receiving specimen
• Calculation made each week for specimens received
  in the week before last
• Calculation made by county and nationally
• Adjustment for season, long-term trends and past
  outbreaks
• Uses poisson regression, principle developed by
  Farrington and friends

32
Current week 35 past weeks
2004

• Present counts are compared to the counts in 7
  weeks in each of the past 5 years

33
Output

• Each week the output is assessed by an
  epidemiologist
• Alerts thought to represent real outbreaks are
  analysed further
• Website www.mave-tarm.dk

34
(No Transcript)
35
Point source outbreak
36
Point source outbreak
37
Usefulness Widespread outbreak
38
S. Oranienburg outbreak

• Hypothesis generating interviews (7 cases)
• All had eaten a particular chokolade from a
  german retail store
• Outbreak in Germany (400 cases)
• Case-control study pointed to chokolade
• But the particular chokolade was very popular in
  Germany (not in Denmark)
• Same DNA-profil

Werber et al. BMC Infectious Diseases 5 7 (2005)
39
What is the most useful?

• Systematic weekly analysis
• Defines expected levels
• Good to detect widespread outbreaks with
  scattered cases
• Good use of advanced lab typing method

40
Early warning signals from mortality
surveillance

• Excess deaths
• due to known disease under surveillance
• Increased incidence
• Increased virulence
• due to disease/threats not under surveillance
• Known diseases
• New, emerging threats
• Environmental threats
• Deliberate release

41
Would mortality surveillance been of use in
2003/04to assess the impact of Fujian influenza
on children in Denmark?

• Absence of signal
• Reassurance of public

42
All-cause deaths and influenza like illness (ILI)
consultation rate, 1998-2004, Denmark
Period of model fitting
Forecast
43
Observed and expected all-cause
deaths,1998-2004, Denmark,
Excess mortality
44
Model testing, season 2003/2004
45
Model testing, season 2003/2004
46
Model testing, season 2003/2004
Signal
disease surveillance (flu, meningitis
etc) meteorological office -
Media reports Community concern Rumours Clinician
concern
47
Model testing, season 2003/2004
Signal
48
Observed and expected number of death among
children (1-15y), Denmark, 1998-2004
49
Model testing, season 2003/2004
50
Evaluation of early warning and response systems

• Important
• usefulness has not been established
• investigating false alarms is costly
• CDC tool for evaluation of surveillance systems
  for early detection of outbreaks

51
Early warning system in Serbia

• ALERT implemented 2002
• To strenghten early detection of outbreaks of
  epidemic prone and emerging infectious diseases
• 11 syndromes to detect priority communicable
  diseases
• All primary health facilities report weekly
  aggregated data
• Complements routine surveillance of individual
  confirmed cases

52
Evaluation of ALERT 2003

• ALERT detected outbreaks more timely than the
  routine systems but ALERT did not detect all
  outbreaks
• Missed clusters of brucellosis and tularaemia
• ALERT procedures response not regulated by law
• Investigation and verification process that
  follows system alerts and signals not fully
  understood
• Recommendations
• Add data source (eg emergency wards) to increase
  sensitivity
• Better integration with routine system
• Change in surveillance perspective requires
  TRAINING!

Valenciano et al, Euro surv 2004 9(5)1-2
53
Useful links

• CDC. Framework for evaluating public health
  surveillance systems for early detection of
  outbreaks. http//www.cdc.gov/mmwr/preview/mmwrhtm
  l/rr5305a1.htm
• Annotated Bibliography for Syndromic Surveillance
  http//www.cdc.gov/EPO/dphsi/syndromic/index.htm
• The RODS Open Source Project, Open Source
  Outbreak and Disease Surveillance Software
  http//openrods.sourceforge.net/