31-Analytical Tools in Surveillance_2011

Analysis & graphical display of surveillance data

EPIET Introductory Course

October 2010

Surveillance system

Health Care Services

Indicator

Public Health Authority

Reporting

Collection

Data

Intervention

Collation,

Analysis,

Interpretation

& Presentation

Information

Dissemination

2

Purpose of surveillance

 Monitor trends (PPT)

 Monitor control programmes

 Detect unusual events

Objectives

 Define steps in surveillance analysis

 Perform descriptive analysis

 Use surveillance data for alert

 Understand mechanisms of more complex analysis

Knowledge of surveillance system / surveillance data

– Changes over time

– Multiple sources of information

– Data entry and validation

– Problem of quality and completeness

Evaluation of the system

Surveillance indicators & Denominator issues

Choice of indicator according to availability of denominator :

• no denominator available:

 crude number of cases

 proportional morbidity

• denominator available:

 calculation of rates

 standardisation

Denominators

Albanian and refugee populations,

Albania, week 16/1999

Surveillance indicators

Distribution of attendance at health facility by diagnosis

Albania, Week 19 (10-16/05/1999)

Diarrhoea by age groups, weeks 15-19, 1999, Albania

1400

Number of notified cases

1200

< 5 years

5 years and +

1000

800

600

400

200

0

15 16 17

Week

18 19

Notifying healh-care centres and notified out-patients

Albania, weeks 15-19, 1999

140

120

100

80

60

40

20

0

15

NGO

Number of health centres

MoH

16 17

Week

18 19

Number of out-patients

30000

25000

20000

15000

10000

5000

0

15 16 17

Week

18 19

800

600

400

Number of notified cases

1400

1200

< 5 years

5 years and +

Diarrhoea by age groups, weeks 15-19, 1999, Albania

30

Proportional morbidity

< 5 years

5 years and +

20

1000 10

0

2

15 16 17 18

Diarrhoea / cardiovasc.

19

1

200

0

15 16 17

Week

18 19

0

15 16 17 18 19

Surveillance indicators

& denominator issues

Choice of indicator according to availability of denominator :

• no denominator available:

 crude number of cases

 proportional morbidity

• denominator available:

 calculation of rates

 standardisation

Incidence Rate of Disease =

XX.X / 1000

Incidence Rate of Disease =

YY.Y / 1000

Compare XX.X to YY.Y

Disease frequency varies according to age

Age structure of the 2 populations is different

Standardisation

Descriptive Analysis of Person Characteristics

 Frequency distributions

– Tables

– Histograms for quantitative classification: age…

– Bars for ordinal or nominal qualitative classification: uneven age-groups, suspectconfirmed…

– Pie for nominal qualitative classification: sex, strain, region

14

Descriptive analysis

- Place -

 Mapping

– Dot/Spot map

– Chloropleth map

 Choice of

– Colors

– Scale

Asthma cases in Barcelona by district

January 21, 1986 l l l l l l l 7

4 l

5 l l l l

3 l l l l l l l l l l l l l l l l l

2 l l l l l l l l l l l l l l l l l l l l l l l l

1 l l l l l l l l l l l l l l l l l

6 l l l l l l l l l l l l l l l l l l

9 l

10

Mediterranean sea

8 l

H1N1 Map by number of confirmed cases

Wikipedia May 9, 2009

50 000+ confirmed cases

5 000+ confirmed cases

500+ confirmed cases

50+ confirmed cases

5+ confirmed cases

1+ confirmed cases

17

Combination?

18

• Equal amplitude scale : (120 / 4)

INCIDENCE | Freq Percent Cum

--------------+---------------------

1 – 30 | 10 33.3% 33.3%

31 – 60 | 16 53.3% 86.6%

61 – 90 | 2 6.7% 93.3%

91 – 120 | 2 6.7% 100.0%

--------------+---------------------

Total | 30 100.0%

• Equal frequency scale : (30 / 3)

INCIDENCE | Freq Percent Cum

--------------+---------------------

1 – 30 | 10 33.3% 33.3%

31 – 39 | 10 33.3% 66.6%

40 – 120 | 10 33.3% 100.0%

--------------+---------------------

Total | 30 100.0%

Convenience scale

INCIDENCE | Freq Percent Cum

--------------+---------------------

< 100 | 28 93.3% 93.3%

>= 100 | 2 6.7% 100.0%

--------------+---------------------

Total | 30 100.0%

Seasonal influenza: incidence rate (%) by region

France, January-March 2003

Incidence Rate (%)

0,00 – 2,85

2,86 – 5,69

5,70 – 8,54

Equal amplitude scale

Incidence Rate (%)

0,00 – 1,02

1,03 – 2,52

2,53 – 8,54

Equal frequency scale

Diarrhoea, week 40, 2008

Estimated number of cases / 100,000

Descriptive analysis

- Time -

Descriptive analysis of time

 Graphical analysis

 Requires aggregation on appropriate time unit

 Choice of time variable

Date of onset

Date of notification

 Use rates when denominator changes over time

 Describe trend, seasonality

25

20

15

10

5

0

37

Descriptive analysis – Time –

Graphical analysis

Number of cases

Foodborne Intoxications Clusters

19

17 17

12

11

12

9

6

5

66

16

15

12 12

9

8 8

6

8 8

7

3

2

5

44

55

44

2

3

4

2

44

3

1

3

2

4

6

12

11

7

14

13

12

11

6

0

14 14

7

6

7 7

4

5

6

7

8

12

10

1

10

2

77

6

333

2

5

44

5

1

4

3

2

3

4

9

8

3

4

5

6

12

10 10

11

12

11

12

10

11

10

6

7

6

8

7

3

9

50 11 24 11 24 37 37

Weeks

50

24

Descriptive analysis – Time –

Graphical analysis

15

10

5

25

Number of cases

20

0

37 50 11 24 37

Weeks

50 11 24 37

100

80

60

40

20

0

50

25

0

50

25

-25

0

50

25

0

Descriptive Analysis of Time

Components of Surveillance Data

Signal

Trend

Seasonality

Residuals

Smoothing techniques:

Moving average

Notifications

Moving average of 12 weeks

Moving average of 52 weeks

30 -

Number of notified cases

25 2005 2006 2007 2008

20 -

15 -

10 -

5 -

0 -

26 39 52 13 26 39 52 13 26 39 52 13 26 39 52 13

Weeks

Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

Nov

Dec

Descriptive Analysis of Time

Smoothing Techniques

822

654

546

728

872

890

692

465

869

726

834

945

1000

500

3622/5=724,4

3690/5=738.0

3728/5=745.6

0

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Effect of the Moving Average Window Size

Weekly Notifications of Salmonellosis, Georgia, 1993-1994

3 weeks

5 weeks

7 weeks

10 weeks

29

Steps in Surveillance Analysis

 Prerequisite to Surveillance Analysis

– Knowledge of surveillance system (evaluation)

• Nature of surveillance data

• Data quality

– Choice of indicator & Denominator issue

 Analysis

– Descriptive (PPT)

– Detection of unusual variations / test hypothesis

30

Methodological considerations when testing for time hypothesis

 Surveillance data

– Does not result from sampling cases

– Can be viewed as a sample of time units

“ Ecological analysis ”

– Time units are not independent

“ Correlated over time ”

– Specific testing methods need to be applied

Testing for time hypothesis

 Convert to rates (if needed)

 Remove time dependency

– Trend and seasons

– By restriction or modelling

 Test for detection of outbreaks

– More cases than expected?

 Test for changes in trend

– Departure from historical trend?

Accounting for Time Dependency

400

300

200

100

700

600

500

Is the red dot consistent with the data?

0

1 10 19 28 37 46 55 64 73 82 91 100 109 118 127 136

Tests not accounting for time dependency

Mean + 1.96 Standard Deviations

700

600

500

400

300

200

100

-10

0

10 30

Yes

110 130

95% CI

Mean

150 50 70 90

Randomly ordered data

Tests accounting for time dependency

95% CI

0,0

-0,1

-0,2

-0,3

-0,4

0,3

0,2

0,1

400

300

200

100

700

600

500

0

1

1

10

12

19 28

23

37 46

No

Chronologically ordered data

55 64 73 82 91 100 109 118 127 136

34 45 56 67

Mois

78 89

Residuals, after removing trend and seasonality

100 111 122

95% CI

Mean

Statistical tests for time series

 For time series with no trend and seasonality: random series

– Tests not accounting for time dependency (TD)

– Chi square, Poisson

 For time series with seasonality and no trend

– Tests accounting for TD by restriction

– Similar historical period mean/median

 For all time series

– Tests accounting for TD by modeling

Olympic Games Surveillance, Athens 2004

Septic Shocks, Syndromic Surveillance

 Poisson test

– Count of cases/average previous 7 days ( l

)

P-value between 1-4% <1%

37

Restriction approach:

Historical mean

2009 X 0

2008 X 1

2007 X 4

X 2

X 5

2006 X 7 X 8

2005 X 10 X 11

2004 X 13 X 14

X

X

3

6

X 9

X 12

X 15

12-15 16-19 20-23

Mean and standard deviation

X = X i

/ 15

Std(X) = (X i

X)² / n

Test

X 0 >X + 1.96*Std(X)

Only applicable if data does not present a significant trend

Conclusion

Know the system / the data

– role of artefacts, errors, …

 First step = graphic description

– PPT

 More complex analysis

– Statistical testing

– Chance, bias, truth?

 Data analysis by epidemiologist

– Added value +++

 Hypothesis must be validated

– Specific investigation / study

Analysis of surveillance data

= Translating data into information

 Provides the basis for public health action

 Requires sound analysis and interpretation

 Extracts meaningful, actionable findings

 Requires clear presentation of complex issues

41

To know more about surveillance data analysis taking into account time dependency

EPIET TSA Module!

42