1. Lecture 2: Ecologic Studies and Cross-Sectional Studies
Dr. Dick Menzies
June 7th, 2006

2. Ecologic studies Basic design: Study populations:
Exposures at group level
Diseases (outcomes) at group level
Association is between group characteristics, and disease occurrence in the group
Study populations:
Any level or type of grouping of subjects
Country, Province, city, neighbourhood, city block
KEY - group should be relatively homogeneous
At least for exposure of interest
Differences between groups should be more than differences within groups

3. Ecologic studies - exposures
Exposures are measured at group level
National/regional data - Per capita income, consumption of alcohol, cigarettes
Reported illnesses (if 1 disease leads to another)
Environmental data – climate, air pollution
Census data – income, education, housing
Down to city block
KEY – exposures should be uniform for all of group

4. Ecologic studies – comparisons:
Temporal – Disease changes over time
Adv – same population (genetics, lifestyles)
Disadv – Many other things can change
Spatial – Different incidence of disease in different places
Adv – same time so temporal changes less
Disadv – may be many other differences

5. Ecologic studies – analysis:
Correlation (can answer)
Is there a relationship between X and Y? (agreement)
Is it statistically significant?
What direction (positive or negative)?
Regression (can answer)
Is there a relationship between X and Y?
What is magnitude of effect?
Y = a + bX

6. R = +1.0 Perfect correlation (rare)

7. R = +0.6 Strong correlation, but not perfect

8. R = -0.8 Strong correlation, but negative

9. R = 0 No correlation at all

10. R = 0 Is the correlation zero for the same reason?

11. Correlation – tells us about agreement of tests
Regression - estimates change of X per unit change in Y
Figure Scatterplot of C1q-binding complexes and IgG-containing complexes (Adapted from N Engl J Med 1978; 298: 126.)

12. Ecologic study - example
Skin test sensitivity to coccidiomycosis and place of residence
What might be a source of error?
What does it tell us an individual’s skin test?

14. Ecologic studies - the observation
National mortality rates from Cardio-vascular disease

16. Ecologic study – CHD mortality
National mortality rates from Cardio-vascular disease differ widely
What could account for observed differences?
Some plausible hypotheses:
How to test these hypotheses?

18. Ecologic study – air pollution
Observation: Rates of lung cancer lower in rural areas than urban.
Why – what are plausible hypotheses:
Confounding - what else might be different?
How to control for this?

20. Ecologic studies and Tuberculosis

21. Decreasing coal use and decreasing TB USA: 1953 – 2003
Coal use ( ) and TB incidence ( ) .

22. Increasing coal use and increasing TB China: 1978 - 2004
Total coal combustion ( ) Notified cases of of TB ( ).

23. Ecologic studies – summary of problems
Exposures – Assumed uniform for all individuals, but this is rarely true
Smoking or alcohol – NOT at all
Income – Depends on size of group
Environmental – More uniform
Fluoridation of water - ?Uniform exposure?
KEY – heterogeneity within population should be less than between populations
Confounding – HUGE problem.
Temporal – many things change over time
Spatial – many other differences between places

24. Cross-Sectional or Prevalence Studies
Snap shot of disease and exposures at the same time in a population
Like a case control study –
measures disease that is present/has occurred
Measures exposures at same time.
Key difference - strategy to identify controls

25. Uses Define risk factors for a disease:
Personal – demographic, life-style
Medical (lipids, BP, other meds)
Environmental
Occupation
Define prevalence of disease and of exposures
Defines population impact of a given exposure.
Useful for Health policy, planning health services utilization, public health programmes

26. Limitations Not useful for
Etiologic research (cannot be sure of cause and effect)
Temporal trends – increasing prevalence may reflect greater incidence, or longer duration (better survival), or changes in population (aging, or selective in – or out-migration)
Problems of cross-sectional surveys in general
Higher prevalence may be associated with factors because:
Causes higher incidence of disease = BAD
Or, longer duration = GOOD

27. Diseases best studied
Diseases studied should be reasonably common, ie high prevalence
Otherwise study will involve too many controls without condition - this is inefficient
Chronic disease with long duration (Higher prevalence)
Or, acute disease with very high incidence

28. Study Population Without specific exposures General population samples
This is difficult and so not done commonly:
One method is random digit dialing (telephone list) or random household selection (using GIS)
Staged cluster sampling
Must have a complete list of persons, communities to select sample

29. Study Populations Proxy General Population School populations
Primary school will be more complete than high school
Requires % of children in school to be representative - but still ignores older, younger, child-less
Work-forces - eg Electricians, Nurses
Not as representative of total population (age, SES, education, healthy worker effect)
But can be used for non-occupational determinants - (as well as occupational determinants)

30. Study Population - Exposure based
Workforce Studies
Workforce studies for occupational exposures
eg Asbestos workers and Lung Ca or mesothelioma
Health care workers and TB
But healthy worker effect
And some characteristics might be quite specific to work force
Special Populations
Prisoners, military, mental institutions
Useful for studying selected exposures in these populations

31. Selecting the Study Population
Census survey - means survey all of the population
Feasible if - you are the government
or - you have a small group
Sample surveys – random selection
Individuals - from the entire population
Need a LIST of population
Cluster sampling – select population groups
Need a LIST of all groups
Take all persons in selected units
Eg., workers on certain wards in hospitals or residents of certain neighbourhoods

32. Study methods - Detecting the disease
Case definition:
Must be very clear,
Same as other published studies
Include mild or asymptomatic cases ?
Diagnostic method
Questionnaires - “Have you been diagnosed with …?”
Quicker, cheaper
Validity – has questionnaire been used before??
Direct - sero-prevalence, diabetes, lipids, TST
Will this be practical, feasible, acceptable
Will it be valid?

33. Measures of Disease Occurrence - Prevalence
Prevalence = number of persons with condition or disease at a given point in time
Prevalence is really a ratio
Numerator = number of persons with disease
Denominator = all persons in population
Prevalence can be expressed as:
At a given point in time - eg, January 1st, 2004
Or on entry to university or military service
Or can be for a period or time, eg., prevalence during medical school or a five year period of time

34. Specific definitions
Prevalence (P) = Persons with disease/Total population – this is a ratio (NOT A RATE)
Point Prevalence = number of persons with disease at a specific point in time
Period Prevalence = number of persons with disease during a specific period of time
Annual Prevalence = number of persons with disease over one year.
Sero-Prevalence = number of persons with serologic evidence of disease or infection or exposure

35. Measures of Disease Occurrence - Incidence
Incidence = number who develop Disease X in a population initially free of Disease X in YY time.
Numerator = persons with newly developed Disease X
Denominator = persons without Disease X at the beginning of the period of study
Time - YY weeks, months, years
Incidence = NDiseaseX in TimeYY / Ntotal
Births and deaths are a form of incidence
Birth rate, mortality rate

36. Relationship between Incidence and Prevalence
Prevalence = Incidence x Duration
This holds ONLY when:
Incidence is stable
Duration is also stable
These conditions are often not true
Eg., HIV – incidence is changing
Duration is also changing with new effective therapy

37. Prevalence and incidence - examples
In population in small town:
Prevalence of Asthma = 6%
Incidence of asthma (past 20 years) = 2/1,000/year
Duration = ??
Pub Health official needs estimate of ART needed for next 5 years
Incidence symptomatic HIV infection: 0.1% annually
Median survival with ART: 20 years
Prevalence now: 2%
Prevalence in 5 years?

38. Types of Incidence and Prevalence Measures
Rate
Type
Numerator
Denominator
Mortality rate
Incidence
Number of deaths from a disease (or all causes)
Person-years at risk in the population
Attack rate
Number of new cases of a disease
Total population at risk, for a limited period of observation
Period prevalence
Prevalence
Number of existing plus all new cases during given time period
Total population (at risk)

39. Types of Incidence and Prevalence Measures (cont’d)
Rate
Type
Numerator
Denominator
Point Prevalence
Prevalence
Number of cases with disease at one moment in time
Total population (Number of persons surveyed)
Neonatal mortality rate
Incidence
Number of deaths in infants under 28 days of age
N. live births in same period, usually per 1000 annually
Infant mortality rate
Number of deaths in a year of children less than 1 year of age
N. live births in the same period, usually per 1000 annually

40. Incidence or prevalence?
Number of homicides in Montreal in 2005
Number of homicide detectives in Montreal in 2005
Number of new homicide detectives hired by Mtl police force in 2005
Female homicide detectives in Montreal on Dec

41. Exposure Assessment - Misclassification
Often exposures measured retrospectively
Major weakness in cross-sectional studies
Especially diseases with long latency
Or prevalent disease with long duration
Inaccurate recall – random misclassification
Strategies – measure current exposures
Use objective measures: age, sex, BP, weight
Or easily remembered
Smoking history
Pregnancies and children
Occupation

42. Exposure Assessment – Recall bias
Cases with disease remember exposures better
Often prompted by their doctors
Example - Fetal malformations – both parents spend all their time remembering things….
So – almost all exposures significantly more common in diseased
Control by – measure disease directly at time of survey, and assess exposure before disease status known. (eg screening for HIV, TST)
If disease ascertained by questionnaire use more objective exposure measures

43. Misclassification Random – poor measurement
Questionnaire – memory problems
Direct measurements – poor tests
Reduces chance of finding real associations
Non-random – biased measurement
Questionnaire – recall bias
Direct – rarely a problem
Will produce biased estimates
No, or spurious, or even reverse associations

44. Measures of Disease Association Prevalence Odds Ratios
Summary measure of disease association in prevalence studies
General formula:
odds of exposure given disease
odds of exposure given no disease
Like case control studies, prevalence studies identify subjects on the basis of disease status.

45. Measures of disease association Prevalence Odds Ratio
In a prevalence survey, 60 individuals were found to have diabetes out of 1,000 surveyed
Prevalence of diabetes total = 6%
Prevalence of diabetes among obese persons = 27/200 = 13.5%
Prevalence of diabetes in non obese persons = 33/773 = 4.3%
Obesity
Not Obesity
Totals
Diabetes 27 33 60
No Diabetes
200
740
940

46. Prevalence Odds Ratio, cont’d
Obesity
Not Obesity
Totals
Diabetes 27 33 60
No Diabetes
200
740
940
Express the findings as prevalence odds
i.e., odds of exposure if disease
or, odds of obesity if diabetes = 27/33 = 0.81
Odds of obesity if not diabetes = 200/740 = 0.27
Prevalence odds ratio (POR) = 0.81/0.27 = 3.0
For cross-sectional or prevalence studies the prevalence odds ratio is the same as the ratio of the prevalence of disease in persons with and without the risk factor

47. POR: Effect of random mis-classification
CHILDHOOD obesity and adult diabetes
Child Obesity
Not obese as child
Total
Yes No
Diabetes
The truth
Answers 27 14 13 33 16 17 60
No diabetes
200
100
740
370
940

48. POR: Effect of random mis-classification
CHILDHOOD obesity and adult diabetes
Child obesity
Not obese as child
Totals
Diabetes 30 60
No diabetes
470
940
Odds of childhood obesity if diabetic:
30/30 = 1.0
Odds of childhood obesity if not diabetic
470/470 = 1.0
POR = 30/30 / 470/470 = 1.0

49. POR: Non-random mis-classification
Adult diabetics more likely to recall obesity in childhood
Child Obesity
Not obese as child
Total
Yes No
Diabetes
The truth
Answers 27 -- 33 15 18 60
No diabetes
200
740
940

50. POR: Non-random mis-classification
Recall bias - CHILDHOOD obesity and adult diabetes
Child obesity
Not obese as child
Totals
Diabetes 45 15 60
No diabetes
200
740
940
Odds of childhood obesity if diabetic:
45/15= 3.0
Odds of childhood obesity if not diabetic
200/740 = 0.27
POR = 3.0 / 0.27 = 11.4

51. “Longitudinal prevalence studies”
In some cross-sectional studies inferences can be made about incidence, as if a cohort design was used
When population has spectrum of years of exposure/age
Tuberculin or HIV sero-prevalence survey
Years of work as health professional
However, this design still has same problems of retrospective exposure assessment

52. Longitudinal prevalence study - example
Results of HIV sero-prevalence in males and females in different age groups:
Sex
Age groups
0-14
15-18
19-22
23-26
27-30
Males 1% 2% 4%
10%
18%
Females 8%
14%
16%
What can one say about incidence
By age?
By age and sex?

53. Longitudinal prevalence study - example
Estimated annual incidence in each age/sex category
Sex
Age groups
0-14
15-18
19-22
23-26
27-30
Males
0.06%
0.25%
0.5%
1.5% 2%
Females
0.12%
Incidence higher in younger females, but then males ‘catch up’