Overview of Epidemiology Concepts Used in Assessment

1
Overview of Epidemiology Concepts Used in
Assessment

• Christie Spice, MPH
• Washington State Department of Health
• Introduction to Community Health Assessment
• September 19, 2006

2
Learning Objectives

• Explain the purpose of descriptive epidemiology
  and how it is used for assessment
• Describe why rates are important in doing
  assessment
• Name three kinds of rates
• Explain how to use confidence intervals to
  compare rate
• Describe the problem posed by small numbers and
  list one solution

3
Epidemiology

• Study of the distribution and determinants of
  diseases and injuries in human populations

4
Analytic Epidemiology

• Identify the determinants of health outcomes
• Research cause-effect relationships, testing
  hypotheses
• Case-control studies
• Cohort studies
• Randomized clinical trials

5
Descriptive Epidemiology

• Monitor known health concerns and identify
  emerging problems
• Prioritize public health problems
• Identify population groups and places at greatest
  risk
• Target resources and interventions
• Inform policy and program development
• Evaluate programs and practices

6

• Patterns of health events relative to
• Person,
• Place,
• and Time

7
Patterns Relative to Person

• Who has the greatest risk of getting disease or
  other health outcome?

• Age
• Education
• Race
• Gender

• Occupation
• Income
• Ethnicity
• Behavior

8
Motor Vehicle Injury Hospitalization Rates by Age
King County, 1989-1993
65
65
45-64
45-64
25-44
25-44
15-24
15-24
1-14
1-14
lt1
lt1
0.0
50.0
100.0
150.0
200.0
250.0
Rate per 100,000
9
Stratifying by Race/Ethnicity

• Race is a marker for social, economic, and
  political factors that influence health
• Analyzing data by race is essential for
  identifying disparities in health outcomes
• But be cautious
• Racial categories may not be reflective of
  individuals self-identification
• May suggest biological differences and/or
  reinforce negative stereotypes
• http//www.doh.wa.gov/Data/Guidelines/
  Raceguide1.htm

10
Patterns Relative to Place

• Where is the health outcome occurring?
• State-wide
• County level
• Sub-county level

11
Asthma Hospitalizations for Ages 1-17 in King
CountyThree Year Average, 1994-1996
Rate per 100,000
50.5 to 116.6116.7 to 252.9253.0 to 565.9566.0
to 567.6
12
(No Transcript)
13
(No Transcript)
14
Patterns Relative to Time

• Is the frequency of the health outcome changing
  over time?
• Annual trends
• Seasonal occurrence
• Daily or hourly changes

15
Accidental Poisoning Mortality RatesKing County,
1980-1996
16
Measuring Frequency

• Counts
• And
• Rates

17
Counts

• Absolute frequency (number) of health events
• Used for
• Health services planning describe the magnitude
  of the problem
• Sentinal events health events that are so rare
  or serious that any occurrence raises concern

18
Hepatitis A Cases
Where is the risk of Hepatitis A higher? Is it
hard to tell? What other information do you need?
19
Anatomy of a Rate
K A standard unit of the population (per 100,
1000, or 100,000)
20
Hepatitis A Cases (1993-1995)
The risk of hepatitis A is higher in Klickitat
County When comparing rates in different
places use the same time period for each place
21
Rates

• An expression of events relative to the size of
  the population in which they occurred.
• Allow us to compare the risk of health events
  across different groups of people, places, and
  time periods.

22
Type of Rates

• Crude rates
• Category-specific rates (e.g., age, gender, race)
• Age-adjusted rates

23
Crude Rates

• Total number of events divided by total
  population
• For mortality, crude rates can either be deaths
  from all causes or cause-specific (e.g., deaths
  from coronary heart disease)

24
Crude Rates CHD Mortality
25
Category-Specific Rates

• Rates for categories of the population defined on
  the basis of particular characteristics.
• Numerator and denominator are restricted to the
  group of interest

26
Age-Specific Rates CHD
27

• Age-Adjusted Rates
• Standardization

28
Unadjusted (Crude) Rates
29
The Problem with Crude Rates

• They dont account for
• underlying demographic differences
• between communities
• (or between time periods)
• that can affect rates.

30
MERCER ISLAND
POPULATION PYRAMID, BY AGE AND GENDER

1997
AGE
GROUP
85
80-84
75-79
TOTAL MALES 9,691
TOTAL FEMALES 10,601
70-74
65-69
60-64
55-59
50-54
45-49
40-44
35-39
30-34
25-29
20-24
15-19
10-14
5-9
lt5
-6
-4
-2
0
2
4
6
PERCENT OF TOTAL POPULATION




MALES
FEMALES
31
SNOQUALMIE VALLEY
POPULATION PYRAMID, BY AGE AND GENDER

1997
AGE
GROUP
85
80-84
TOTAL MALES 16,985
75-79
TOTAL FEMALES 16,713
70-74
65-69
60-64
55-59
50-54
45-49
40-44
35-39
30-34
25-29
20-24
15-19
10-14
5-9
lt5
-6
-4
-2
0
2
4
6




MALES
FEMALES
PERCENT OF TOTAL POPULATION
32
Two Solutions to the Problem

• Compare only age-specific rates

33
Age-Specific Rates CHD
34
Two Solutions to the Problem

• Compare only age-specific rates cumbersome!
• Construct age-adjusted rates summary measures
  that accounts for differences in the underlying
  age distributions of populations.

35
Age-Adjustment

• Weight the observed age-specific rates by a
  standard population
• 2000 projected US population
• Represents the hypothetical rate that would have
  been observed if the population of interest had
  the same age distribution as the standard
  population.

36
Age-Adjusted Rates CHD
Age-adjusted to projected US population in year
2000.
37
Age-Adjusted Rates
AA1 age-adjusted to the 1940 US population
standard AA2 age-adjusted to the 2000 US
population standard Choice of a population
standard makes a difference! Use the 2000 US
population standard as a rule.
38
Rates Summary

• Five questions to ask about a rate
• What kind of rate is it?
• What is it a rate of?
• To what population does it refer?
• How stable is it?
• How was the information obtained?
• Abramson, JH. Making Sense of Data. New York
  Oxford University Press. 1994.
• http//www.doh.wa.gov/Data/Guidelines/
  Rateguide.htm

39
Making Comparisons OverPersons, Places, and Time

• How stable are the rates?
• When is a difference or change
• statistically significant?

40
Suicide Rates

• 1982 1983
• Kent 5.3 18.8
• King County 13.1 13.1

41
Suicide Rates

• 1982 1983 1984
• Kent 5.3 18.8 6.6
• King County 13.1 13.1 14.0

42
Suicide Rates

• 1982 1983 1984 1988
• Kent 5.3 18.8 6.6 17.9
• King Co. 13.1 13.1 14.0 14.8

43
Unstable Rates

• When rates fluctuate widely, it is often because
  the size of the population and/or the frequency
  of the health event is small
• The addition or subtraction of a few cases
  dramatically affects the size of the rate

44
Suicide Rates, Kent Health Planning Area
1980-1995
20
18
16
14
12
10
Rate per 100,000
8
6
4
2
0
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
45
Unstable Rates

• Rates based on small numbers of events area
  affected by random variation or chance
• Addition or subtraction of a few cases
  dramatically affects the size of the rate
• Comparisons across years or between areas are
  hard to interpret when rates are unstable

46
Confidence Intervals (CI)

• The CI for a rate indicates the expected range of
  random variation due to chance
• CIs tell you about the stability of a rate
• Bigger random variability leads to wider CIs and
  more unstable rates

47
Age-Adjusted Suicide Rates 1995
N. Central Seattle
Highline/Burien
Upper limit
Lower limit
Rate
King County
Kent
Kirkland/Redmond
H.P. 2000
0.0
5.0
10.0
15.0
20.0
25.0
30.0
35.0
Rate per 100,000 (age-adjusted to U.S. 1940)
48
Using CIs to Assess Statistical Significance

• CIs not a test of statistical significance, but
  they are often used as a proxy
• Two observed rates
• Significant if CIs do not overlap
• An observed rate and a target
• Significant if CI does not contain the target
• http//www.doh.wa.gov/Data/Guidelines/ConfIntguide
  .htm

49
Interpreting Statistical Significance

• For 95 CI there is less than 5 probability
  that any observed difference in the rates is
  merely the result of chance or random variation
• If the rates are not significantly different,
  chance or random variation cannot be ruled out as
  a cause of the difference

50
Age-Adjusted Suicide Rates 1995
N. Central Seattle
Highline/Burien
King County
Kent
Kirkland/Redmond
H.P. 2000
0.0
5.0
10.0
15.0
20.0
25.0
30.0
35.0
Rate per 100,000 (age-adjusted to U.S. 1940)
51

• Testing for Significant
• Change Over Time

52
Testing for Trend in Rates

• Tests whether an increase or decrease in a series
  of rates is random variation or a true change
  occurring in the population.

53
P-values

• P-value the probability that an increase or
  decrease in rates is just random variation.
• The smaller the p-value, the more statistically
  significant the trend.
• Conventional definition of statistical
  significant is p-value is less than 0.05
• Less than 5 chance that the increase or decrease
  is due to random variation

54
Teen (Age 15-17) Birth Rates 1980-1995
SE Seattle, Highline/Burien, North County
70
SE Seattle
Highline/Burien
60
N. County
p0.012
50
40
plt0.001
30
20
p0.084
10
0
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
55
Teen (Age 15-17) Abortion Rates 1981-1995
Seattle and King County Outside Seattle
90
80
Seattle
King Co-Seattle
70
p0.962
60
50
Rate per 1,000 Females Age 15-17
40
30
plt0.001
20
10
0
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
56
Teen (Age 15-17) Abortion Rates 1981-1995
Seattle and King County Outside Seattle
90
Seattle
80
King Co-Seattle
Linear (Seattle)
70
p0.962
60
50
Rate per 1,000 Females Age 15-17
40
30
plt0.001
20
10
0
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
57
Teen (Age 15-17) Abortion Rates 1981-1995
Seattle and King County Outside Seattle
90
80
plt0.001
plt0.001
70
60
50
Rate per 1,000 Females Age 15-17
40
30
plt0.001
20
Seattle
King Co-Seattle
10
0
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
58
Testing for Trend in Rates

• Trend testing in VistaPHw
• Do not use Vistas trend test for hospitalization
  data
• For other data in VistaPHw
• If plt0.01, the trend is likely significant
• If pgt0.05, the trend is not significant
• If p-value is between 0.01 and 0.05, use
  JoinPoint to analyze trend

59
Testing for Trend in Rates

• Objective approach to evaluating difference or
  change
• Look at the data when testing for trends
  observe changes in the rates
• Remember that the practical significance of a
  trend may differ from the statistical significance

60

• Dealing with Small Numbers and Unstable Rates

61
(No Transcript)
62
Aggregating to Stabilize Rates

• Grouping is used to increase the size of the
  numerator and denominator and stabilize the
  rates
• Combine across multiple years of data
• Combine across geographic areas (e.g., counties)
• Combine across groups of people (e.g., aggregate
  5-9 and 10-14 age groups into a 5-14 age group)

63
Aggregation

• Numerator sum of all events occurring during
  the period
• (events in year 1 events in year 2 events
  in year 3)
• Denominator sum of total population at risk
  during the period
• (population in year 1 pop. in year 2 pop. in
  year 3)

64
(No Transcript)
65
Infant Mortality Rates, 1995-1997
Island
Grant
State Total
Thurston
Whatcom
0
2
4
6
8
10
12
14
16
18
20
Rate per 1,000 Live Births
66
Infant Mortality Rates, 1993-1997
Island
Grant
State Total
Thurston
Whatcom
0
2
4
6
8
10
12
14
16
18
20
Rate per 1,000 Live Births
67
Infant Mortality Rates, 1988-1997
Island
Grant
State Total
Thurston
Whatcom
0
2
4
6
8
10
12
14
16
18
20
Rate per 1,000 Live Births
68
Small Numbers Summary

• Aggregation
• Key limitation to aggregation loss of
  information
• Dilemma how many years to aggregate?
• Often a tradeoff between stability of rates and
  specificity of the information. Consider the
  purpose of the analysis.
• http//www.doh.wa.gov/Data/Guidelines/SmallNumbers
  .htm

69
Rolling Averages

• Used for graphical presentation to smooth trend
  lines.
• Overlapping aggregate rates.
• Not appropriate for testing statistical
  significance of trends in rates.

70
Suicide Rates Kent Health Planning Area
1980-1995
20.0
18.0
16.0
14.0
12.0
10.0
Rate per 100,000
8.0
6.0
4.0
Single year rates
2.0
Five year rolling averages
0.0
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995