of Categories Epidemiologic Studies
Epidemiologic Studies Observational Descriptive Analytic Experimental Randomized controlled trials Community trials
(Case reports) (Case series)
Ecologic Cross-sectional
Prospective cohort Retrospective cohort Case-control
Descriptive Observational Studies
Most commonly used to look for patterns of disease Measure occurrence of disease and/or risk factors for disease (exposures) in a population Answer the questions: Who? What? Where? When? May suggest a direction to explore Why?, but cannot provide an answer due to lack of an explicit comparison group
Analytic Observational Studies
In contrast to descriptive studies, analytic studies incorporate a formal comparison group Test specific etiologic hypotheses or Generate new hypotheses or Suggest mechanisms of causation As knowledge about likely etiologic factors builds, preventive hypotheses can be tested
The Epidemiologic Study
Uncontrolled Assignment Controlled Assignment
OBSERVATIONAL STUDIES
EXPERIMENTAL STUDIES
Sampling with regard to exposure
Sampling with regard to disease/outcome
Randomized person assignment
Community assignment
COHORT STUDIES
CASECONTROL STUDIES
RANDOMIZED CONTROLLED TRIALS
COMMUNITY TRIALS
Cumulative Incidence
Cumulative Incidence = Risk or Probability
Number of new cases of disease during a time period Number of subjects followed for the time period
"The only way to interpret risk is to know the length of the time period over which the risk applies"
Person-Time Incidence Rate
Number of new events in a specified time period Total person time at risk x Rate Base
Deals with uneven follow-up time by only considering time at risk contributed by each person in the study Person time: i.e. 100 person years = 100 persons x 1 year or 50 persons x 2 years or 25 persons x 4 years, etc.... Rate base: 100, 1000, 100,000 person-years
In published incidence rates, often the person-years are implied. Example: cancer risk of 8.5 per 100,000
Point Prevalence Rate (point prevalence proportion)
Numerator: Number of existing cases at a specific point in time Denominator: Total number in a defined population at the same point in time Reports of prevalence are usually referring to point prevalence (as opposed to period prevalence)
Note: Prevalence is a proportion (not a real rate)
Relationship between Incidence and Prevalence
Prevalence depends on the duration of the disease and the incidence of the disease:
Prevalence = Incidence X Average Duration of Disease
Age Adjustment
Age and sex have special status as confounders Often, when rates of disease are reported and compared, they are sex-specific and are adjusted for age Adjusted rates are most often used in descriptive epidemiologic studies
Measures of Association
Measures of disease frequency (proportions or rates) are compared between those with and those without the exposure of interest Comparison can be summarized using differences or ratios:
Rateexposed Rateunexposed Rateexposed / Rateunexposed
Risk Ratios
Example from the Honolulu Heart Program
Diabetes
Total incidence? IRc(exposed)? IRc(unexposed)? Is BMI related to risk of diabetes?
BMI High Low/ Norm
Yes 140 160 300
No 1860 3840 5700 2000 4000
Risk Ratio =
Rate Ratios
Example from the Honolulu Heart Program
Diabetes Incidence rate overall? IRP-T(exposed)? High IRP-T(unexposed)? Low/ Norm 140 160 300 BMI Yes PersonTime 9000 18000 27000
Is BMI related to risk of diabetes?
Rate Ratio =
Odds Ratios
Example from the Honolulu Heart Program
Diabetes
Odds(exposed)? Odds(unexposed)?
BMI High Low/ Norm
Yes 140 160 300
No 1860 3840 5700 2000 4000
Odds Ratio =
Defining Target Populations and Study Populations (Samples)
Target Population: the population to which we would like to generalize our findings Inference Study Population: a sample of members of the target population
Effect estimate, point estimate, study parameter
Sampling Error
The random error that can occur when we use sample statistics to represent population parameters (Oleckno p. 152) Type I (sampling) error:
The probability of finding an association in the study sample none when exists in the target population = P-value =
Type II (sampling) error:
The probability of not finding an association when one exists in the target population = 1-Power =
Confidence Intervals
Measures the precision of a point estimate
A narrow confidence interval indicates a precise point estimate A wide confidence interval indicates an imprecise point estimate
Because p-values are based on an arbitrary cut-off, most epidemiologists prefer to consider confidence intervals instead
Consider the SIDS trends for PEI and for British Columbia:
1985-1989 SIDS Deaths PEI 8 SIDS Rate (95% CI) 0.82 (0.35-1.62) 1.50 (1.34-1.67) 1994-1998 SIDS Deaths 4 SIDS Rate (95% CI) 0.49 (0.13-1.25) 0.41 (0.33-0.50)
Relative Risk 1994-8 / 1985-9 (95% CI)
0.59 (0.18-1.97) 0.27 (0.22-0.35)
BC
318
93
Association
A statistical relationship A measure of dependence between two variables The degree to which the rate of a disease or outcome is higher (or lower) among those with an exposure as compared to those without an exposure
Association
We've determined if an association exists between an exposure and outcome (relative risk, risk ratio, odds ratio or other correlation) But is the association real? Types of association: 1. Spurious 2. Non-Causal 3. Causal
Spurious Associations: are false associations arising from sampling error or bias
Sampling error occurs when the association we see is actually due to chance
p=0.05 means we accept that there is a 5% chance our association is a result of random sampling error
Bias is systematic (nonrandom) error in the design, conduct or analysis of an epidemiologic study
Non-Causal Associations
1. May be due to confounding factors 2. May be due to reverse causation
Important to establish the order of the exposure and outcome Exposure must not be a result of the outcome
Causal Associations
Changes in exposure result in changes in the outcome Cannot "prove causality"
Sampling error Selection bias or measurement error Unknown causal or confounding factors
Determining cause-effect relationship involves careful reasoning and judgment of literature
Conjecture and Refutation
Conjecture: a proposition (as in mathematics) before it has been proved or disproved Refute: to prove wrong by argument or evidence Develop Hypothesis Consider New Knowledge Evaluate Results Test Hypothesis
Source: Webb, 2006
Conjecture and Refutation
Conjecture: a proposition (as in mathematics) before it has been We can use Hill's criteria proved or disproved to help guide our Refute: to prove wrong by argument or evaluation of the evidence evidence Develop Hypothesis Consider New Knowledge Evaluate Results Test Hypothesis
Source: Webb, 2006
Attributable Risks and Attributable Rates
Risk differences have a specific meaning when the risk factor is causal Number of cases in the exposed that can be attributed to the exposure itself In the absence of the exposure, people who would have been exposed still experience the background risk of disease represented by the unexposed rate
Attributable Risk = IRc(exposed) - IRc(unexposed)
Population Attributable Risk
The incidence of a disease in the whole population that is due to a given exposure Population Attributable Risk = IRc(population) - IRc(unexposed)
Causes of Death and PAR
Population attributable risk has tangible public health significance. Nine factors account for about 50% of deaths in the United States (McGinnis and Foege see Oleckno page 133) - tobacco, diet and activity, alcohol, microbial agents, toxic agents, firearms, sexual behaviour, motor vehicles, and illicit use of drugs
All are modifiable through primary prevention
Study Designs:
Cross-sectional, prospective cohort, casecontrol and randomized controlled trials Which effect measure goes with which design and how to compute these measures How to conduct these studies Pros and cons of each design Biases that each is subject to
Study Aids
Learning outcomes list in the syllabus Previous exams online In-class exercises and homework Group project and individual assignment Drop-in review session with TAs (Dates and times TBD)