biostats & epidemiology

MEDD course

____ the probability of detecting a difference, given the alternate hypothesis is true.

Power is the likelihood of correctly rejecting the null hypothesis when the alternative hypothesis is indeed true.

Logistic Regression

iv: continuous or categorical/binary and confounding variables

dv: binary (can only be positive)

linear regression

iv: continuous or binary/categorical and confounding variables

dv: continuous

correlation

iv= contionus

dv= continous

ANOVA or f-test

iv: more than 2 samples

dv: continous

Chi-squared

iv: binary

dv: binary

2 sample t-test

iv: binary

dv: continuous

paired t-test

iv: 2 paired observations (before/after meds)

dv: continuous

risk

number of events / population at risk

incidence rate

new cases/ person-time at risk

person-time

average number of people in at risk pool * time frame

attributable risk

AR/ R(experimental)

AR= Risk experimental - Risk control

Relative Risk Increase

Re- Rc/ Rc

or

RR-1

odds

disease/ no disease or risk/1-risk

kappa statistic

compares agreement between observers with likelihood of agreement by chance

1= perfect agreement, no chance

0= no agreement, or agreement due to random chance

negative= agreement is worse than chance

sensitivity

the ability to detect a condition among individuals who truly have a condition

true positive/ true positive + false negative

good RULE OUT, low false negative rate

false negative rate= 1 - seNsitivity

specificity

identify the absence of a condition among individuals who truly do not have the condition

true negative/ true negative + false positive

good Rule In, low false positive rate

False positive= 1 -sPecificity

positive predictive value

true positive/ true positive + false positive

if you have a positive test, what is the probability that it is true

based on prevalence

negative predictive value

if you have a negative test, whats the porbability it is true

true negative/ true negative + false negative

how do rare diseases affect PPV and NPV

PPV is largely affected, specificity may not be a good rule in

NPV is not as affected, sensitivity will still be a goof rule out

SPIN SNOUT

positive likelihood ratio

how much more/less likely a positive test result is for someone with a condition compared to someone without a condition

True positive/ false positive

sensitivity/1-specificity

negative likelihood ratio

How much more/less likely is a negative test in someone WITH the disease compared with someone WITHOUT the disease?

How much does a NEGATIVE test change the odds that someone has the disease?

false negative/true negative

1-sensitivity/specificity

specificity always in the denominator

describe a receiver-operating characteristic curve

higher the curve the better

changing cutoff creates different sensitivity and specificity trade offs

are under the curve for ROC graph

ranges from 0.5-1

tells you how well a test with continuous or numerical data points can distinguish sick people from healthy people across all possible cutoffs

0.5 means no discrimination

how are clinical prediction rules derived

1. development phase-derivation

   1. observational study, derives a rule/scale/scoring system

2. validation phase

   1. rule tested in separate population usually in observational population

3. impact analysis

   1. rule tested in clinical setting to see if it improves patient outcomes

4. integration into practice

observational studies

lowest tier

animal/laboratory studies

case reports

cross-sectional studies

case control studies

cohort studies

experimental studies

2nd tier

randomized control trials

synthesized evidence

1st tier

systematic reviews

meta analysis

appraisal

structured critical evaluation of a study’s methodology, results, and relevance to decide whether its evidence should influence clinical decisions

steps to appraisal

1. abstract overview

   1. confirm type of question and type of trial

2. methods survey

   1. identify PICO

   2. For RCTs identify process of randomization and allocation concealment

3. enrollment/flow diagram

   1. assess selection biases

4. baseline characteristics

   1. variables inherent to patient population, randomization will balance all of this, reducing confounding variables

5. risk of bias

   1. internal validity- degree to which the study results reflect reality for the types of patients included in the study

6. key results

   1. these are your risk ratios, relative risk, attributable risk and stuff like that

7. applicability

   1. degree to which study results can be applied to real-world populations outside the trial

equipoise

the medical ethical state when there exists genuine uncertainty about the benefits/harms of treatments

no equipoise means its not ethical to continuous