Basic Principles of research design

Document Sample
Basic Principles of research design Powered By Docstoc
					Basic Principles of
 research design

   Research Design
      Dent 313
Research fields in medicine
   Biological sciences
    •   Biology of disease
   Clinical sciences
    •   Information to care for individual patients
    •   Clinical Epidemiology
   Population sciences
    •   Epidemiology
         • Study of disease occurring in human population
   Health services
    •   Study of how non-biological factors affect the patient’s
Clinical epidemiology
   The science of making predictions about
    individual patients
   By counting clinical events of similar
   And using strong scientific methods
   To ensure that the predictions are
Purpose of clinical epidemiology
   Develop and apply methods of clinical
    observation that will lead to valid
    conclusions by avoiding being mislead
    by systematic error and the play of
   Obtaining the kind of information
    clinicians need to make good decision in
    the care of patients
Clinical epidemiology
   It is clinical
    • it answers clinical questions
    • It guides clinical decision making
Evidence-based medicine
   Application of clinical epidemiology to
    the care of the patient
Basic principles
   Clinical question
   Variables
    • Things that vary and can be measured
       • Dependent vs. independent variables
   Health outcomes
   Numbers and probability
    • Quantitative vs. qualitative
Clinical question
   Is the patient sick or well (abnormality)
   How accurate are tests used to diagnose disease (diagnosis)
   How often does a disease occur (frequency)
   What factors are associated with an increased risk of disease
   What are the consequences of having a disease (prognosis)
   How does treatment change the course of disease (treatment)
   Does an intervention on well people keep disease from arising
   What lead to disease (cause)
Health outcomes
   Events that can be studied directly in intact
    humans only
   Include the five Ds
    •   Disease
    •   Discomfort
    •   Dissatisfaction
    •   Disability
    •   Death
Numbers and probability
   Clinical science depends on quantitative measures
   Impressions, instincts and beliefs are only important
    when added to a solid grounds of numerical
    •   This allows for better confirmation
    •   And estimation of error
   Prediction of treatment outcomes or disease
    •   Better be expressed as a percentage
   (Probability) needs to be expressed
    •   Estimated by referring to past experience with groups of
        similar patients
Population and samples
 Populations
 All people in a defined setting with
  certain defined characteristics
    • Examples:
      • The general population
      • A hospitalized population
      • A population of patients with a specific
Population and samples
A sample
 Is a subset of people in the defined population

 Selected from that population

 It is not practical to test all the population

 Clinical research is carried out on samples

 A sample makes inference about the
Population and samples
   Two important points in sampling
       • Are the conclusions of the research
        correct for the people in the sample?

       • If so, does the sample represent fairly
        the population of interest?
A sample is representative
   Depends on how a sample was selected
   Equal chance for all members vs.
   Computerized programs for selection of
   Definition:
    • “A process at any stage of inference tending
        to produce results that depart systematically
        from the true values”
    •   “Any trend in the collection, analysis,
        interpretation, publication, or review of the
        data that can lead to conclusions that are
        systematically different from the truth
Categories of bias
   Selection bias
   Measurement bias
   Confounding bias
Selection bias
   Occurs when comparisons are made
    between groups of patients that differ in
    ways other than the main factors under
   Example:
    •   Examine dental caries among different age
    •   Examine perio condition without adjustment
        for smoking
Measurement bias
   Occurs when the methods of
    measurement are not similar among
    different groups of patients
   Examples
    •   Examine dental caries visually vs.
    •   Examine the WL of Roots using different
Confounding bias
       Occurs when two factors or processes are
        associated or "travel together " and the effect
        of one is confused with or distorted by the
        effect of the other
       Example:
    •      TG and cholesterol levels are associated with risk
           for coronary heart disease
    •      Education and/or income with good health
    •      Folic acid vs. lower rates of colon cancer
          • People taking multivitamins are health conscious
               about diet and exercise
Confounding bias
   A variable is not confounded if it is
    directly along the path from cause to
   A confounding variable is not necessarily
    a cause itself
    • May be related to the suspected cause and
      the effect in an instance but not related in
   Selection bias is an issue in patients
    selection for observation, and so it is
    important in the design of a study
   Confounding bias is an issue in
    analysis of the data, once the
    observations have been made
   Often in the same study more than one
    bias operates
   A distinction must be made between
    the potential for bias and the actual
    presence of bias in a particular study
Dealing with bias
   Identification of bias
   Measuring the potential effect of bias
    • Modifying the research design when the
        potential effect on the result is big
    •   Changing the conclusions in a clinically
        meaningful way when the effect is not big
   Unbiased samples may misrepresent the
    population because of chance
   Chance is the divergence of an
    observation on a sample from the true
    population value
   is called also “random variation”
    • Example: Tossing a coin 100 times
    • The larger the sample size the less the
Chance vs. bias
   Bias distorts the situation in one direction
    or another
   Chance / random variation results in an
    observation above the true value as
    likely as one below it.
    • The mean of many unbiased observations of
        a sample approximates the true observation
        of the population
    •   In small samples this may not be close to the
        true observation of the population
             Bias v. chance
   Bias can be prevented by proper
    conduction of clinical investigations
   Bias can be corrected through proper
    data analysis

   Chance cannot be eliminated
   Its influence can be reduced by
    proper research design
 Statistics can be used to estimate the
  probability of chance or random
 Chance can’t be eliminated, but its
  influence can be reduced by proper
  design of research
Relationship between bias & chance

           True BP               BP measurement

     Intra-arterial canula      Sphygmomanometer




           80                        90
   Truth
   Validity is correspondence to the true value
    measured or searched for
   For an observation to be valid, it must be
    neither biased nor incorrect due to chance
   Types
    • Internal validity
    • External validity
Internal validity
   Is the degree to which the results of a study
    are correct for the patients being studied
   Internal
    •   Applies to the conditions of the particular group of
        patients being observed and not to others
   Is determined by how well the design, data
    collection and analyses are conducted and
    threatened by biases and random variation
   Necessary but not sufficient by itself
External validity
   Is the degree to which the results of an
    observation hold true in other settings
   The answer of:
    •   “Assuming that the results are true in other settings,
        do they apply to my patients as well?”
   Generalizability assumes that patients in a
    study are similar to other patients
   A study with high internal validity may be
    misleading if its results are generalized to the
    wrong patients
All patients with condition of interest              Internal validity


       patients                                    A Selection       B

                                                Measurement & confounding
   population                    ?

                                 ?                       chance
            External validity
            Generalizability                          Conclusion

Shared By: