Download Lab 8: Types of and Study Designs

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Evidence-based toxicology wikipedia , lookup

Triclocarban wikipedia , lookup

Transcript
Lab 8: Types of Studies and
Study Designs
Lab Workbook (pp. 37 – 40)
A fish or being taught to fish?
• Lab based on study by Jolson et al. (1992)
• Concepts and techniques remain valid for
– all disciplines
– all populations
– all designs
Background
• Population = patients undergoing bone
marrow ablation
• Exposure = generic drug
– Group 1 = exposed (N1 = 25)
– Group 0 = nonexposed (N0 = 34)
• Disease (outcome) = cerebellar toxicity
• Hypothesis – generic drug presents greater
risk of toxicity
Question 1 (p. 37)
• Read the Patients and Methods of the
article. Is this study experimental or
nonexperimental?
• The investigators studied the exposure
without intervention.
• Thus: nonexperimental (“observational”)
Question 2
• Suppose you could redesign the study as a
trial. Describe a scheme for randomizing
the exposure.
• Options:
– Flip of coin
– Tokens in a hat (half 1, half 0)
– Use www.randomization.com
Question 3
• What is the primary benefit of
randomization?
• Randomization balances measured and
unmeasured cofactors (potential
confounders)
• Hence, difference found at end of study
attribute to exposure and not confounding
Question 4
• The study is a cohort study ... Suppose it we
had conducted it ecologically ... difficulties
with ecological design . . . ?
• Greater opportunity for confounding
(discuss)
• Opportunity for the aggregation bias /
ecological fallacy (discuss)
Question 5
• Results
 risk1 = 11 / 25 = 44%
 risk0 = 3 / 34 = 9%
• What is random error in this context? …discuss…
• How it was dealt:
– one-way ANOVA tests of means
– chi-square and Fisher’s tests of proportions
– 95% confidence intervals for risk ratios
Question 6
• Confounding derives from inherent differences at
baseline . . . How did investigators address
potential for confounding
• Table 1 -- no large differences by age, sex, type of
leukemia, stage of disease, kidney function, etc.
• Also adjustment of RRs [Mantel-Haenszel]
• Concluded: potential for confounding was small
Question 7
• Misclassification / (information) diagnostic
suspicion bias?
• Yes, greater level of scrutiny in patients
taking the generic drug!
Question 8
• Study population was identified because of
the problem. Selection bias?
• Yes, this might be a 1 in a 1000 chanceoccurrence
– What does the p value mean in the context?
– Is this like shooting the broad side of a barn and
drawing the bull’s-eye afterwards?
Question 9
• Is the relation between the exposure and
outcome causal?
• Causal inference consider other factors
– e.g., Hill’s criteria (studied in epi)
– Understanding causal mechanism is key
Question 10
• Should drug be pulled from market?
• Factors that contribute to the decision
–
–
–
–
Scientific evidence
Finance (profitability)
Medico-legal (law suits)
Politics
• Use of scientific results for political and
economic purposes are always suspect!