Flashcards in Biostats Deck (32):
Case controlled study
(observational and retrospective)
compares cases of people with disease to people without disease, looking for Prior Exposure or Risk Factors
Data is analyzed by the Odds Ratio (tells what odds a person with a particular exposure will get the disease. e.g. COPD pts. have higher odds of a history of smoking than people without COPD)
Can happen if study is done in one place only
- happens when any factor in the study leads to subjects not being a representative sample of the general population
Absolute Risk Reduction (ARR)
Reduction in risk associated with a treatment compared to a placebo
ARR = Risk of control - Risk of treatment
e.g. if 8% of people who receive placebo get the flu v.s. 2% of people who receive flu vaccine, then ARR = 8% - 2% = 6%
(observational and prospective)
compares group WITH given exposure or risk factor to a group WITHOUT those exposures/factors
followed to see if disease develops.
Uses Relative Risk (e.g. smokers have a higher risk of developing COPD than nonsmokers)
"Who will develop the disease?"
SENSITIVITY (True + Rate)
probability that a person WITH dz. will have + test
Column 1: A/(A+C)
- SNOUT: want high sensitivity to rule out disease, especially if missing a disease is REALLY bad
e.g. ELISA for HIV
probability that a patient WITHOUT dz. will have (-) test
Column 2: D/(D+B)
- SPIN: want specificity to rule in disease, to confirm a likely diagnosis if a false + result may prove harmful
e.g. Western Blot
A/C / B/D --> AD/BC
estimate of relative risk that is used in case control studies
- odds that the group with the disease was exposed to a risk factor DIVIDED BY odds that the group without disease was exposed to a risk factor
RELATIVE RISK (Risk Ratio)
A/(A+B) / C/(C+D)
incidence of disease (exposed group) / incidence of disease (unexposed group)
how much more likely an exposed person is to get a disease compared to an unexposed person, used in Cohort studies
Positive Predictive Value (PPV)
portion of + test results that are True + (probability that a pt. with a + test truly HAS the dz.)
The more specific a test, the higher its PPV.
The higher prevalence, the higher PPV.
1st row: A/(A+B)
Negative Predictive Value (NPV)
proportion of negative tests results that are True (-) probability that pt. with a - disease does NOT have dz.
The more sensitive a test, the higher its NPV.
The higher prevalence, the lower its NPV.
2nd row: D/(D+C)
+ Likelihood ratio formula
(diseased people with + result/non diseased with + result)
sensitivity/ (1 - specificity)
(-) Likelihood ratio formula
(diseased people with - result/non diseased with - result
(1 - sensitivity)/ specificity
Type I Error
concluding that there is a difference between groups when in fact there isn't
estimate of probability that differences could have happened by chance alone [
Type II Error
concluding that there i s NOT a difference between groups when in fact there IS
Confidence Interval (CI)
way of expressing statistical significance, reliability of an estimate
CI of 95% = p
- means that there is a 95% chance that the true Hazard Ratio is within the range of the confidence interval. So, if the CI contains 1.0, there is no difference between the two factors and the null hypothesis cannot be rejected.
Largest effect size, what do you choose?
Study with greatest power, what do you choose?
Best test, what do you choose?
Significant and powerful
Cross Sectional study
assess frequency of disease and related risk factors at a particular point in time
"What is happening?"
Compares monozygotic and dizygotic frequency
difference in risk between the exposed group and unexposed group
A/(A+B) - C/(C+D)
Number Needed to Treat
number of pts. who need to be treated for 1 pt. to benefit
1/ (Risk of control - Risk of treatment)
e.g. in the flu vaccine example of 8% -2% = ARR of 6% then number needed to treat is 1/.06 = 16
knowledge of presence of disorder alters subject recall
Nonrandom people in a study group
subjects not representative of general population
subjects in different groups not treated the same
a factor present is related to both exposure and outcome and distorts the effect of exposure on the outcome
group being studied changes its behavior due to knowledge that they are being studied
If HR for a scenario = 1, then both meds have the same likelihood of causing the symptom.
If HR > 1 then Drug A has greater likelihood than drug B, and opposite is true if HR
any study that includes blinding (e.g. double-blinding) prevents observational bias.