go back (01aug2003/02jul04)
DrTim homepage link

CH2 Study Designs

Intro to Clinical Statistics by Timothy Bilash MD, MS
August 2003

based on:
Review of Basic & Clinical Biostatistics by
Beth Dawson, Robert Trapp (2001) CH2

  1. CLASSIFICATION of study designs: (from Fig 2-1 Dawson & Trapp p8)

    1. Observational Studies
      1. Case-series (group)
        1. describes group
      2. Cross-sectional studies, surveys (prevalence)
        1. describe disease and disease process
        2. diagnose and stage disease
      3. Case-control studies (retrospective)
        1. identification of possible risk factors working back from outcomes
        2. cause and incidence of disease
      4. Cohort studies (prospective or historical retrospective)
        1. identification of risk factors for outcome
        2. strong evidence for cause and incidence of disease outcome
        3. natural history, prognosis

    2. Experimental Studies/ Clinical Trials
      1. uncontrolled trials
        1. reports outcome, unknown whether better or worse than anything else
      2. controlled trials
        1. randomized (concurrent controls)
        2. not randomized (concurrent controls)
        3. self-controlled (sequential controls)
        4. crossover (sequential controls)

    3. Meta-Analysis and Reviews
      1. review
        1. compares published findings
      2. meta-analysis
        1. combines published results from observational studies or experiments to reach overall quantitative and summary conclusions
        2. may be done on observational studies or experiments
        3. most valid if use randomized and controlled experimental studies only
        4. should not combine observational and experimental results, report separately
        5. does not take the place of well-designed clinical trials

  2. OBSERVATIONAL STUDIES: (listed in order of increasing reliability)

    well-designed observational studies can provide useful insights on disease causation, even though they do not constitute proof of causes.

    1. Time-Limited Studies

      1. Case-series ("group") study
        1. group = series of patients, described and characterized in a report
        2. descriptive, weakest observational study
        3. usually over brief period of time, usually conceived and written after the fact
        4. simplest design is observation of a small number of patients
        5. subject to selection, observation as well as other biases
        6. useful to generate hypotheses and insight for further research

      2. Cross-sectional ("prevalence") study
        1. data collected at one point in time (snapshot in time)
        2. surveys (can also be in cohort study) and polls
        3. many surveys done from databases
        4. not the same group if repeated, so infer disease progression with caution
        5. purpose is to describe "what is happening right now", to obtain current opinion or practice
        6. useful for evaluating diagnostic procedure and current status of a disease or condition
        7. subject to drop-out bias over time (death of patients at upper limit of disease changes risk distribution)
        8. conclusions usually based on subset who agree to participate, may not represent entire population
        9. may lead to misleading conclusions if focuses on a time-dependent process

    2. Longitudinal ("Time") Studies

      1. "time study" of hypotheses over an extended period of time (ongoing)
        1. each has a tense (past or future, when events occur relative to the start of the study)
        2. each has a direction of inquiry (forward/to or backward/from the starting point in time)
        3. variables are risk/exposure and outcome
        4. functions used are grouping and statistical summarizing
        5. each subject known to have a risk of exposure (exposure is statistically defined)

      2. Case-control Retrospective (backward looking) study
        1. in the past/retrospective
        2. starts with a consequence and looks for precursors or risk factors
        3. usually conceived and written after exposure and outcomes (like case-series)
        4. backward-looking, now to past, retrospective backward timeline
        5. identify risk factors from outcome
        6. outcome has happened (outcome defines as a case)
        7. purpose is "what risk factor was present given outcome" (what caused the outcomes)
          1. outcome (case) group matched to a nooutcome (control) group for factors other than the outcomes
          2. risk factors associated (from) outcomes
          3. "was risk/exposure present more often in group that had outcome as compared to group that had nooutcome"
          4. statistics done on the risk/exposures for each outcome group
        8. can be completed in a shorter time and less expensive
        9. can provide useful insights if well designed
          1. useful for rare conditions or that take years to manifest
          2. valuable for testing a new premise or preliminary information
          3. if indicates a risk factor, then follow-up cohort study needed to demonstrate forward causality
        10. case-control has the greatest number of biasing factors of all time studies if used to establish cause
          1. risk factor present in outcome group is consistent with causation, but does not prove
            1. additional causative factor may be common to the risk and outcome groups
              1. example colon cancer vs fiber rather than caloric intake
              2. cancer vs coffee intake rather than smoking
          2. risk factors may be sufficient but not necessary as well may miss other risk factors
          3. depends on accurate records, since unable to verify risk factors (measure) after the fact
          4. backward probability not same as forward causal probability (as in cohort or experimental study) unless know there is causation
            1. if has causation, then backward and forward are the same
          5. big issue is selection of appropriate control group
            1. patients with completely unrelated problem useful as control
            2. recommend 2 control groups:
              1. one similar group to treatment group
              2. another dissimilar (healthy) group

      3. Cohort (forward looking) studies

        1. Outcome Assessment study
          1. quality of care
          2. population based questionnaires or statistically defined outcome measures
          3. patient satisfaction not necessarily reflects quality of care

        2. Historical Cohort Retrospective study
          1. forward-looking, past to now, retrospective forward timeline
          2. in the past/retrospective
          3. risk factors associated (to) outcomes
          4. both risk factors and outcomes have happened prior to the study
          5. purpose to describe "what outcome happened given risk"
            1. group has something in common, remain part of the group over time
            2. group by risk/exposure that occurred and was documented some time prior to the outcomes, then examine (statistic) outcomes for each risk/exposure group
            3. "was outcome present more often in group that had risk/exposure as compared to group that had no risk factor/exposure"
            4. past risk factors to present outcome (forward timeline)

        3. Cohort Prospective study
          1. forward-looking, now to future, prospective forward timeline
          2. in the future/prospective
          3. starts with a risk factor or exposure and looks at consequences
          4. study starts before the exposure or outcome
          5. risk factors associated (to) outcomes
          6. risk factor has happened (risk defines cohort), not outcome prior to the study
          7. purpose to describe "what outcome will happen given risk"
            1. group by risk/exposure, then examine (statistic) outcomes for each risk/exposure group collected at some time future to the risk
            2. group has something in common, remain part of the group over time
            3. "is outcome present more often in group that has risk/exposure as compared to group that has no risk factor/exposure"
            4. current risk factor to future outcome (forward timeline)
          8. good for exploring the causes of a condition. results from a well-designed cohort study carry more weight in establishing cause than case-control study, because of fewer biasing factors. causation generally cannot be proved with cohort studies because they are observational and not interventional. however, because they follow a cohort of patients forward through time, they possess the correct time sequence to provide strong evidence for causes and effects, as in the smoking and lung cancer controversy. in addition, prospective cohort studies, as opposed to historical control studies, control many sources of bias related to patient selection and recorded measurements. (p19) they are sensitive but not specific.


    1. Uncontrolled
      1. treatment without a comparison group
      2. often used for a procedure
      3. unable to know if some other factor common to the risk and outcome groups is causative, so doesn't provide causation
      4. doesn't compare to any other treatments (one group only)

    2. Controlled
      1. comparison to another cohort group (control)
      2. controlled studies have greater validity because separate experimental factors from other factors
      3. each subject known to be exposed or not exposed
      4. types of controls
        1. concurrent controls (comparison group)
        2. self-controlled
          1. Hawthorne Effect: patients change their behavior because of the extra attention rather than because of treatment
          2. similar to cohort studies
        3. crossover
          1. cohort study when switch treatment and control groups
        4. external/historical controls
          1. compare to group not in the study
          2. used for diseases which Ont have cure yet
          3. historical controls good for preliminary studies
          4. unsure that treatment group has same characteristics as control
          5. "biases in patient selection may irretrievably weight the outcome of historical controlled trials in favor of new therapies". beware of trials with historical controls that report in favor of the experimental treatment.
            1. Historical controlled trials showed therapy was better in 79% of historical trials, but only 20% of concurrent randomized controlled trials (comparing 50 randomized clinical trials to 56 trials with historical controls)
              1. patients in the historical control groups generally reported to have worse outcomes
          6. inappropriate use of historical controls has led to serious errors in medicine
            1. freezing gastric ulcers
            2. low-protein diets in hepatic failure
            3. concurrent controlled studies revealed the fallacies
          7. bias decreased when
            1. state hypothesis prior to trial
            2. same investigators treat the historical controls

    3. Randomized
      1. randomization
        1. provides the strongest evidence for concluding causation (if also controlled???)
          1. randomization limits selection bias for treatment vs control group
      2. non-randomization
        1. considered much weaker because no way to prevent bias in patient assignment
        2. whenever patients are assigned to treatments within big blocks of time, there is always the possibility that an important event occurred between the two times

    4. Blinded
      1. double-blind trials reduce actions based on the knowledge which affect the outcome by patients and researchers (prevents feedback)
      2. single-blind trial

  4. Clinical trials/Experimental vs Observational cohort studies compared

    1. Causality: Assignment, Exposure and Outcome Steps are required in any study
      1. assignment to group
        1. random (= not known but equal partitioning)
        2. controlled (= known)
      2. exposure or treatment of individual subject
        1. statistical link to actual exposure (risk of exposure)
        2. direct link to actual exposure
      3. reactions of the patient or researchers that alter the assignment or exposure (feedback)
        1. compliance
        2. blinding
      4. outcome of individual subject
        1. statistical link to actual outcome (risk of outcome)
        2. direct link to actual outcome
      5. bias
        1. factors that alter the measurement/reporting/assignment or exposure at any step
        2. diminishes evidence of causation

    2. study types implicitly use probability (0-100%) or have bias at each step in the analysis when determining causality
      1. the more certain the knowledge about result and the bias at each linking step, the more certain the causal probability
      2. reverseing the order or direction of steps changes the probability measure of causality
        1. order sometimes matters and not interchangeable (a-priori, post-priori)
        2. an example is assignment before (controlled) or after (historical cohort) outcome
      3. experiment has direct link to both exposure and outcome (known)
        1. experiment has direct causal links at each step
          1. in particular, exposure/intervention is direct in experiment
      4. observation has statistical link to some exposure factors and/or outcomes (probability)
        1. cohort has statistical links at each steps
        2. obeys Bayes theorems of post-priori probabilities
          1. probabilities multiple
      5. equivalent if the statistics accurately reflect the individual

    3. cohort studies give strong evidence for causation, but cannot prove because don't control other factors (observational and not interventional) (p19)
      1. treatment may be correlated with other unrecognized causative factor in cohort study
      2. a prospectively designed cohort study can control many sources of bias related to patient selection and recorded measurements as opposed to historical cohort
      3. cohort provides strong evidence for possible cause and effect of a condition because follows the correct time sequence for causality
      4. useful for natural progression of disease or identifying risk factors
      5. cohort studies are frequently weakened by patient attrition

    4. randomized controlled clinical trials provide the strongest evidence for causation because they are experiments subject to the least number of biases
      1. assignment is random
        1. randomization limits bias from exposure to factors being unequal in groups
      2. assignment is controlled
        1. control limits bias from mixing up treatment groups
      3. exposure is directly known in a given subject
        1. treatment is independent of other factors in experiment because imposed by researcher, not conditions of the group
      4. outcome is directly measured, in same subject as exposure
      5. measurements are unbiased

    5. an extended time period for a study makes it more difficult to argue causality, because other events in the intervening period may affect the outcome
      1. long period of time between exposure and effect
      2. vulnerable to patient attrition or migration that affects complete follow-up.


goto www.DrTimDelivers.com

page views since Sept2007