Quiz 3

10/31

LIMITATIONS OF EPIDEMIOLOGICAL DESIGN

Stages of Epidemiologic Study

  1. Observe possible association

  2. Develop a hypothesis about a cause/effect relationship

  3. Test hypothesis through formal study

  4. And next…?

Study Validity

  • Degree to which conclusions drawn from study are warranted

  • Takes into account

  1. Representativeness of study sample

  2. Study methods

  3. Extraneous (outside) factors that impact study results

  • 2 broad components

    • 1) Internal validity

→ proper selection of study group

→ lack of error in measurement

Appropriate measurement of

→ exposure

→ outcome

→ association between exposure and outcome

  • 2) External Validity

→ Ability to extrapolate results from a sample pop to a target pop

Internal Validity: extend to which differences in outcome between groups in a study can be attributed to the effects of exposure or intervention being investigated

→ has IV when

  • Random error

  • Bias

  • Confounding 

Have been ruled out


Error in Epidemiologic Research: 2 Broad Categories

  • Random Error (non systematic)

    • Leads to false association between exposure and disease that arises from chance that seems to have no assignable cause

    • Sampling error: (The most common form of random error) the possibility that the sample collected is not representative of the target population. Is relevant to all types of epi studies

      • A non representative sample may occur even in random sampling – bad luck

      • How might one reduce sampling error?

      • To Avoid: increase sample size

  • Systematic error: Key distinction from random error: this type of error is introduced by the study process.

    • 3 Types:

      • 1) Selection Bias

      • 2) Information/Observation Bias

      • 3) Confounding

    • A more serious problem to study validity

    • Selection & information bias result from an error committed by the investigator in the design or conduct of the study that leads to a false association between exposure and disease

    • Confounding stems from the natural mixing of effects between exposure and disease and a third factor called a confounder. Distorts the relationship between exposure and disease


Bias

  • Selection Bias

    • Relationship between exposure/disease is diff for those who participate in/eligible for study but do not participate

    • Relevant to both selection & retention of study subjects in study

  • Information/Observation Bias

Bias in epidemiology

  • Bias in epidemiology is different from common definition:

    • Not prejudice or inclination by investigator

    • Rather ignorance or poor decision making

Consequences of Bias

  • Important because it results in an incorrect or invalid estimate of the measure of association 

    • i.e. results of the study

Selection Bias in Case Control Studies

  • Self selection bias: Who is likely to respond to a health survey?

    • Iowa Women’s Health Study

    • Looking back, non respondents were more likely to die of smoking associated diseases

    • Participates or Refuses to participate based on disease or exposure status 

      • To reduce: increase response rate > 80%

      • Assess for self-selection bias: check characteristics of participants

  • Control selection bias

    • In a case control study of smoking (risk factor)  and chronic lung disease (disease outcome) , controls are selected from a hospital population

      • Already sick

      • Can you predict what impact this may have on the reported association between exposure (smoking) and disease (chronic lung disease) ?

      • Smoking is associated with many diseases which may result in hospitalization

    • Will this control selection bias Strengthen or Weaken the association between smoking and chronic lung disease??

    • Weaken the association

    • The issue: hospital controls in this case do not represent the prevalence of exposure (smoking) in the community from which cases of chronic lung disease arise

      • Distortion of the exposure-disease association by this control group selection

    • Correction

      • Select controls from population/community, from which cases arose

      • High participation rates >80%

  • Differential surveillance, diagnosis or referral based on exposure history

    • Different referral patterns of patients based on exposure history

      • Example: risk of venous thromboembolism among oral contraceptive users

    • Based on exposure history

      • Prior to the study:

Several case reports of the association between oral contraceptives and venous thrombosis (DVT) and an increased risk of pulmonary emboli (PE) had already been published before

  • Further investigation

    • Because of past reports of DVT and PE risk with oral contraceptives,

    • doctors more likely to hospitalize women on oral contraceptives (OCs) with symtoms of DVT compared with women with similar symptoms of DVT who were not on OCs

    • Example: selection bias

  • The problem here is:

    • At the start of this study, the cases have greater exposure to oral contraceptives compared with controls

  • Loss to follow up

    • Reduce the power of the study

    • May bias the results

      • Over or under estimate the magnitude of the relationship between exposure and outcome

      • Produce an association where none exists

      • Hide an association where one exists

  • Selection Bia