measures of impact 18 th epiet/euphem introductory course september-october 2012 lazareto, menorca,...

38
Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

Upload: maria-wilkinson

Post on 27-Mar-2015

216 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

Measures of Impact

18th EPIET/EUPHEM Introductory Course

September-October 2012Lazareto, Menorca, Spain

Ioannis Karagiannis

Page 2: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

2

Objectives

• To define measures of impact

• To calculate the attributable risk- among the exposed

- in the population

• Eventually, make sense of stuff

Page 3: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

3

Scenario

• You are in charge of health promotion “Preventing automobile-related deaths”

• Limited budget best reduction of deaths

• Evidence: retrospective cohort study: “causes of automobile-related deaths”

Page 4: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

4

Relative Risks

• Best reduction of deaths?

• Prevent drink & drive?

• Prevent speeding?

Relative Risk

Driving too fast 5

Driving while drunk 11

Page 5: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

5

Relative Risks

0.0000050.000001

0.50.1

Risk (exposed) Risk (unexposed)

RR = 5.0

Page 6: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

6

Measures of Impact

• Provide information about the public health impact of an exposure

• Contribution of an exposure to the frequency of disease

• Several concepts- Attributable risk (AR)

- Attributable risk among exposed (AR%)

- Attributable risk in the population (PAR)

- Preventable fraction among exposed (PF)

Page 7: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

7

Attributable Risk (AR)(synonyms: Risk Difference)

• Quantifies disease burden in exposed group attributable to exposure in absolute terms

• AR = Re - Ru

• Answers:- what is the risk attributed to the exposure?

- what is the excess risk due to the exposure?

• Only use if causality “exposure outcome”

Page 8: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

8

Attributable Risk (AR)

• AR = Re - Ru

Outcome

a

c d

yes no

exposed

not exposed

b

Attributable Risk =

a

a+b

c

c+d

a+b

c+d

a

a+b

c

c+d-

Attributable Risk = Re – background risk

= Re

= Ru

Page 9: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

9

Attributable Risk (AR)

Risk

0.01

0.05

Risk of death by speeding

Risk of death by driving below the speed limit

How high is the added risk of dying caused by the exposure “speeding“?

Added risk ?

exposure: speeding0.00

Page 10: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

10

AR Speeding

AR (speeding) = 0.05 - 0.01 = 0.04 “speeding increases the risk of dying by 0.04. Four out of 100 speeding drivers will die in addition to normal (=background) because they drove too fast“.

Page 11: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

11

AR Drunk driving

AR (drunk driving) = 0.15 - 0.01 = 0.14

“drunk driving increases the risk of dying by 0.14. Fourteen out of 100 drunk drivers die in addition to normal (background) death by driving because they were drunk while driving."

Page 12: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

12

Summary so far

Measure Speeding Drunk driving

Relative Risk 5 11

Attributable Risk 0.04 0.14

Page 13: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

13

Attributable Risk Percent (AR%)(synonyms: Attributable Fraction)

• Attributable risk expressed as a percentage of risk in the exposed population

• Proportion of disease among the exposed which:

- can be attributed to the exposure

- could be prevented by eliminating the exposure

• AR% looks at exposed population,not the total population

Page 14: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

14

Attributable Risk Percent (AR%)

• Example speeding: What proportion of all speeding deaths (denominator) died because they drove too fast (numerator)?

deaths caused by speeding

deaths of all who drove too fastAR% = x 100

Page 15: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

15

Attributable Risk Percent (AR%)

Risk (exposed) - Risk (unexposed)

Risk (exposed)x 100

RR > 1

AR% =

Risk (exposed) Risk (unexposed)

Risk (exposed) Risk (exposed)= - x 100

1

Relative Risk= 1 - x 100

RR - 1

RR= x 100

Page 16: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

16

AR% Speeding drivers

AR% (speeding) = 80%“80% of all people who died while driving too fast, died because they drove too fast“.

Page 17: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

17

AR% Drunk drivers

AR% (drunk driving) = 93%“93% of all people who died while being drunk, died because they were drunk“.

Page 18: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

18

Summary so far

Measure Speeding Drunk driving

Relative Risk 5 11

Attributable Risk 0.04 0.14

Attributable Risk% 80% 93%

Page 19: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

19

AR & AR% in Case-Control Studies

• No direct risk estimates in case-control study- AR (risk difference) and AR% calculation

IMPOSSIBLE!

Relative Risk - 1

Relative RiskAR% = x 100

Page 20: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

20

AR & AR% in Case-Control Studies

• No direct risk estimates in case-control study- AR (risk difference) and AR% calculation

IMPOSSIBLE?

• If odds ratio approximates relative risk, then

Relative Risk - 1

Relative RiskAR% = x 100

Odds Ratio - 1

Odds RatioAR% = x 100

Page 21: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

21

Population Attributable Risk (PAR%)

• Proportion of cases in the total population attributable to the exposure

• Proportion of disease in the total population that could be prevented if we could eliminate the risk factor

• Determines exposures relevant to public health in community

• Only use if causality “exposure outcome”

Page 22: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

22

Population Attributable Risk (PAR%)

• Example speeding: What proportion of all people who died (denominator) died because they drove too fast (numerator)?

deaths caused by speeding

total deaths in the populationPAR% = x 100

Page 23: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

23

Population Attributable Risk (PAR%)

Risk (total pop) - Risk (unexposed)

Risk (total pop)x 100PAR% =

p (RR - 1)

p (RR - 1) +1x 100PAR% =

p = proportion of population exposed

PAR% = p(cases) x AR%

p(cases) = proportion of cases exposed

Page 24: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

PAR(%) according to the relative riskfor various level of exposure frequency

among cases

0.0%

10.0%

20.0%

30.0%

40.0%

50.0%

60.0%

70.0%

80.0%

90.0%

100.0%

1 2 3 4 5 6 7 8 9 10

Relative risks

Pop

ulat

ion

attr

ibut

able

fra

ctio

n

Pe 10%Pe 25%Pe 50%Pe 75%Pe 100% (AFe)

Page 25: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

25

PAR% Speeding

Risk (total) - Risk (not exposed)

Risk(total)PAR% = = = 0.44

0.018 - 0.01

0.018

= 44%

risk in total population

risk in unexposed

Page 26: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

26

PAR% Speeding

100

80 7920

2000

8000

10000

dead alive

180 9820

speeding

notspeeding

1900

Risk

100/2000 = 0.05

80/8000 = 0.01

Attributable Risk (AR) = 0.05 - 0.01 = 0.04

AR

Risk(exposed)AR% = x 100 = (0.04/0.05) x 100 = 80%

p(cases) = % cases exposed = 100/180 = 0.55

PAR% = pc x AR% = 0.55 x 80% = 44%

Page 27: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

27

PAR% Drunk driving

Risk (total) - Risk (unexposed)

Risk(total)PAR% = = = 0.22

0.018 - 0.014

0.018

= 22%

risk in total population

risk in unexposed

Page 28: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

28

Summary

Measure Speeding Drunk driving

Relative Risk 5 11

Attributable Risk 0.04 0.14

Attributable Risk% 80% 93%Pop. attributable risk% 44% 22%% drivers with risk factor in population

20% 3%

• Best reduction of deaths?

• Prevent drinking or speeding?

Page 29: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

29

PAR% in Case-Control Studies

• proportion of controls exposed ≈ proportion of population exposed

PAR% =Pcontrols – (OR – 1)

x 100Pcontrols (OR – 1) + 1Pcontrols = Proportion of controls exposed

PAR% = Pcases ( OR – 1 ) x 100OR

Where Pcases = proportion cases exposed

Page 30: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

30

Summary

Measure Meaning Question answered

RR, OR Strength of association (between exposure and outcome)

Is the exposure associated with the risk of getting ill/ the outcome?

AR Excess risk of exposed (in absolute terms)

What is the difference in risk between exposed and not exposed?

AR% Proportion of risk of exposed attributed to exposure, potential prevention of exposed

What proportion of those who are exposed and ill is likely due to the exposure?

PAR% Proportion of risk of population attributed to exposure, potential prevention of population,

Public Health relevance

What proportion of those who are ill in the population is likely due to the exposure?

Page 31: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

Take-home message

• There is more death and disability from frequent exposure to lower risks than to rare exposures to higher risks

• Examples: More people die from marginally elevated blood

pressure (common) than from seriously elevated blood pressure (uncommon)

More people acquire HCV from unsafe injection (common exposure, lower risk) than from unsafe blood products (rare exposure, high risk)

Page 32: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

32

Preventable fraction (PF)

• Exposure associated with decreased risk

• Where RR < 1, exposure is protective

• Proportion of cases that would have occurred if exposure hadn’t been present

Page 33: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

33

• RR < 1 protective exposure (protective factor)

• Proportion of cases that were prevented because of the exposure

Risk (unexposed) - Risk (exposed)

Risk (unexposed)

Preventable fraction (PF)

PF =

Risk (unexposed) Risk (exposed)

Risk (unexposed) Risk (unexposed)

PF = -

PF = 1 - Relative Risk

Page 34: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

34

Preventable Fraction (PF)Vaccine efficacy

  Pop. CasesCases

/100,000

Vaccinated 200,000 100 50

Unvaccinated 300,000 600 200

Total 500,000

Risk (unexposed) - Risk (exposed)

Risk (unexposed)PF =

PF = 600/300,000 - 100/200,000

600/300,000= 0.75

Page 35: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

35

Preventable Fraction (PF)Vaccine efficacy

• How many people would have been ill without the vaccine?

• 200/100,000 cases of unvaccinated

• In population of 200,000 we expect 400 cases

• Only 100 cases occurred; 300 cases were prevented (by vaccine)

• 300/400 = 75% of hypothetical cases were prevented

Page 36: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

True or false?

• The relative risk of lung cancer and smoking is 9

• Therefore, if nobody smoked, the incidence of lung cancer would be nine times lower than it currently is

FalseMeasures of association are not measures of impact.The prevalence of smoking in the population also

matters!

Page 37: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

True or false?

• 90% of patients with lung cancer are smokers

• Therefore, if nobody smoked, the incidence of lung cancer would be reduced by 90%

False The proportion of a disease that may be explained by a specific exposure does not depend on the proportion of cases exposed. It also depends on the strength of the association (90% of patients with lung cancer also eat fresh salad for lunch every day)

Page 38: Measures of Impact 18 th EPIET/EUPHEM Introductory Course September-October 2012 Lazareto, Menorca, Spain Ioannis Karagiannis

Thank you