Deriving Biological Inferences From Epidemiologic Studies

1
Deriving Biological Inferences From Epidemiologic
Studies
2

• Dr. Mostafa Arafa
• Associate Prof. of Family and
  Community medicine
• Faculty of Medicine Medical Sciences
• King Khaled University, S.A.
  
• mostafaarafa_at_hotmail.com

3
Learning Objectives

• To learn the inferences about a diseases
  etiology that can be derived from different
  epidemiologic studies.
• To learn the reasoning by which
  epidemiologists select the most plausible
  inference.

4

• The first step in the epidemiologic analysis is
  the demonstration of a statistical relationship
  between a disease and a biological
  characteristic. The second step is to ascertain
  the meaning of that relationship.
  

5

• Statistical associations can be explained as
• 1- Artificial
• 2- Due to association of interrelated
• but non-causal variables
• 3- Due to uncontrolled confounding
• 4- Causal or etiological

6

• Artificial Association
• Artificial association can result from biased
  methods of selecting cases and controls. It may
  also arise from biased methods of recording or
  obtaining information by interview. Errors in
  conduction or design of the study also may
  introduce spurious association.

7

• Non-causal association
• An association between many variables can be
  observed and still be non-causal, because many
  variables can occur together without being a part
  of causal chain. Examination of interrelated
  associations is useful as they may suggest ways
  to reduce exposure to causal variables.
  

8

• Confounding
• If any factor either increasing or decreasing
  the risk of a disease besides the exposure under
  study is unequally distributed in the groups that
  are compared with regards to the disease, this
  will give rise to difference in diseases
  frequency in the compared groups. Such
  distortion, termed confounding and variables are
  called confounder variables.
  

9
E Etiological Factor
D Diseases
CF Confounding factor
10
Cigarette smoking
Lung cancer
Alcohols
11

• If any study did not control adequately
  control for potential confounders, the inferences
  drawn from the results may not be well founded.
  Studies in which there was inadequate control of
  all known confounders, the results of which may
  be explained by an unequal distribution of
  extraneous variables in the study groups and not
  by the effect of exposure on disease.

12
Methods used for controlling of confounders

• During the design of the study
• Restriction to a specific group
• Matching
• During analysis
• Stratification multivariate analysis

13

• Causal association
• The logicians definition of cause is that a
  factor which must be necessary and sufficient for
  the occurrence of a disease.
• The concept necessary and sufficient implies
  there must be a one-to-one relationship between
  the factor and the disease.

14

• Example of the sufficient cause for development
  of a disease
• A1
• A2
• A3 B
  C
• A4
• A5

Disease
Cellular reaction
15

• Example for necessary cause for development of
  a disease
• A1 A2 A3 B C

Cellular reaction
Disease
16
Assessing Causality

• The following concepts are used in making a
  causal inference
• Strength of association
• Consistency of observed association
• Specificity of the association
• Temporal sequence of events
• Dose-response relationship
• Biological plausibility of association
• Experimental evidence

17

• Strength of association
• It is measured by Relative Risk, and Odds
  ratio. A strong association between exposure and
  outcome gives support to causal hypothesis. When
  a weak association is present, other information
  is needed to support causality.

18

• Consistency of observed association
• Confirmation by repeated findings of an
  association in different studies, in different
  population, and in different settings strengths
  the inference of causal association. It is
  equivalent to replication of results in
  laboratory experiments.

19

• Specificity of the association
• It has been postulated that one exposure
  should cause one disease and no other exposures
  should cause the disease. This has its roots in
  bacteriological models where one organism is
  associated with one disease. This could not be
  applied in chronic diseases as one exposure could
  lead to many adverse outcomes e.g. smoking and
  cancers, CVD, Birth-outcome.

20

• Temporal sequence of events
• It is very obvious that an exposure must
  precede the disease to cause an effect. An
  example for that is the prenatal exposure and
  malformation. In case control studies the problem
  of temporality is quite obvious, however a cohort
  design can resolve the issues of temporality.

21

• Dose response relationships
• The risk of development of a disease should be
  related to the degree of exposure of the causal
  factor e.g. duration of estrogen use and the risk
  of endometrial cancer, dosage of smoking and lung
  cancer staging.

22

• Biological plausibility
• A causal hypothesis must be viewed in the
  light of its biological plausibility. Statistical
  significant relationship should be understood in
  the view of biological significance. A good
  example is cigarette smoking and lung cancer
  relationship which was initially biologically
  implausible by some, but carcinogens in
  cigarettes were identified, which lent biological
  plausibility to the observed association.

23

• Experimental evidence
• Randomized clinical trials is a well run trial
  that may confirm a causal relationship between an
  exposure and outcome. However ethical issues may
  prevent the conduction of such trials.

24
Summary

• Epidemiologic inferences lead to action, to
  changes in clinical practice, public policy,
  legislation, health education or new health
  directions. Epidemiologic studies can provide
  very strong support for hypotheses of either a
  causal or indirect association. Inferences from
  such studies must take into account all relevant
  biological information. Epidemiologic and other
  evidence can accumulate to point where a causal
  hypothesis becomes highly probable.