Biostats-7 Flashcards
Observational Studies
- what are they?
- Association or Causal?
- Ex.
- Natural course of events is just observed and recorded
- ONLY Association NOT causation
Examples:
- Case Series
- Cross-Sectional
- Ecological
- Case-control
- Cohort
Experimental/Interventional
- what are they?
- Association or Causal?
- Ex.
- Investigator allocates exposure
- Can determine Causation
Examples:
- Randomized Controlled Trials
- Animal Trials
What is the Gold Standard or Study Design?
Randomized, Double-Blind, Placebo Controlled Trial
What is randomization?
- what is its purpose?
- If there are 2 treatments the patient should have an equal chance at either
- Ensures that treatment groups are equal on baseline (known) characteristics and unknown characteristics
- Eliminates Selection Bias
- GUARANTEES that the Statistical Significance Tests are valid
What is a double blind trial?
Both investigator and subject are blinded to the exposure so that they do not influence the ascertainment of the outcome
What is the purpose of trials being placebo controlled?
- Controls for Placebo Effect
- Helps to assess side affects
What do you do in Intention to Treat Analysis?
- what if you don’t do this?
- Groups are analyzed according to original, randomized group regardless of whether they received the treatment or not
- By not doing this you introduce selection bias (this is important for accounting for unknown factors that may exist among groups)
What is Type II error?
- when do people end up with Type II error?
- Null is not rejected when it should be
- Inadequate power leads to type II error
What is power?
- what is the threshold?
- What determines power?
- Ability to detect an association or difference if it really exists (higher power is better chances of finding a treatment benefit)
- Power for a study should be over 80%
- Determined by SAMPLE SIZE (small sample = low power)
- Treatment effect
- Increasing the Alpha
How do you calculate power?
Power = 1 - Beta error
What is relative risk?
- when do you use it?
RR = Incidence in exposed group / Incidence in nonexposed group
- Measure of Association used in Clinical Trials
How do you calculate incidence of disease in an exposed group?
- Non-exposed?
Incidence exposed = a/(a+b)
Incidence non-exposed = c/(c+d)
Therefore Relative Risk:
RR = [a/(a+b)]/[c/(c+d)]
What is the risk difference?
- what kind of study is it used in?
RD = Incidence (exposed) - Incidence (unexposed)
How do you calculate Number needed to treat?
- what does this tell you?
= 1/(risk difference)
= 1/ (incidence in placebo - incidence in treated)
**The number of patients you would need to treat with the drug before you saw a successful treatment
What is validity?
Degree to which a measurement or study reaches a correct conclusion
Differentiate between internal and external validity.
Internal Valdity:
- Extent to which the results of the study reflect the truth
External Validty:
- Extent to which the results of a study are applicable to other populations (generalizability)
What is the difference between bias and confounding?
Bias:
- Systematic error that leads to a distortion of the results
Confouding:
- Mixing of the effect of an extraneous variable with the effects of the exposure and disease of interest
What is type I error?
Error that occurs when the null hypothesis is true but it is incorrectly rejected
E.g. if significance level is determined to be 0.05 then 5% of the time null will be rejected when it shouldn’t be
What determines the possibility of you having type I error in your experiment?
- alpha = level of signficance
**This is pre-determined by the investigator - how much error they are willing to accept
What factors does internal validity rely on?
Chance
Bias
Confounding
What does the P-value tell you?
Probability that the results are due to chance
**typically 0.05 or LESS
What does a 95% confidence interval tell you?
Says that if you repeated the study 100 times then 95 of the values obtained will fall within that range
What are some common confounders in studies?
- Age is common because it is associated with both disease and exposure
- Cigarette Smoking
- Alcohol Use
- Physical Activity
- BMI
What are some ways to deal with confounding?
- Stratification by age, gender, smoking, etc.
- Statistical adjustment of the model
- RANDOMIZATION
- matching?
