# 5 Biostats

ambirnba's version from 2015-10-31 15:33

## Section 1

Effect size - concepthow big an effect did the intervention produce?
Effect size - equation risk of a given outcome (e.g. death) w/ intervention - risk of outcome w/o intervention
Linear regressionindicates how a dependent/outcome variable (y) changes in relation to changes of an independent/predictor variable (x), e.g. y = mx + b
Multiple linear regression indicates how dependent variable (y) changes in relation to changes in >1 independent variable (x1, x2, etc.)
rcorrelation coefficient; tells you how well your linear regression line fits the data Ranges between -1 and 1; close to |1| → very good correlation
r^2 coefficient of determination; tells you how much of the variation in y is accounted for by x
Selection biassubjects don’t accurately reflect population. e.g. b/c volunteers have their own agenda; very sick decline participation; no outreach to underrepresented groups
Recall biaspeople’s memories of past exposures influenced by various factors e.g. difficult to remember diet; people tend to search for explanations for unpleasant experiences, such as illnesses, & may over-emphasize importance of certain factors as a result
Response bias: people lie.. e.g. obese subjects tend to under-report actual weight
Misclassification bias: exposure or outcome may be misclassified. e.g. due to unclear records, mix-ups of people w/ similar names, improper calibration
Publication bias: negative/equivocal results tend not to be published
Bias from confounding: when 3rd variable associated w/ both exposure & outcome variables (adjust for confounder → becomes covariate)
Performance bias: when skill has an impact on exposure variable (e.g. surgical technique performed), must account for that in analysis (e.g. by randomizing which surgeon operates on which pt)
Hawthorne effect: process of being observed changes people’s behavior
Digit preference: people prefer to round #s to end in 0 or 5; may skew outcome data
Attrition bias: people may drop out of study at different rates (may be skewed towards best or worst outcomes)
what kind of study reduces risk of bias?Randomized control trial

## Section 2

Lead time bias: perceived increase in survival time of a disease due to screening methods that enable earlier diagnosis, without actually altering course of disease or improving survival
Length time bias: slower-growing cancers have greater probability of being found by screening, and are more likely to be curable, thus making screening appear to be more effective than it actually is; screening programs tend to miss the most aggressive cancers
External validityhow generalizable is this study? How well do the conditions of the study approximate the real-world situation we care about?
Internal validity how sure are we that the change in the dependent variable is due to the change in the independent variable? How well was the study designed (incl. eliminating confounders), and how well was it run?
scale: nominaldata in categories
scale: ordinalranking
scale: intervalconsistent difference between each
scale: ratiointerval that drops to zero
if _________, you can reject the null hypothesis and accept the alternate hypothesis p<0.05
Type I error: rejecting a true null hypothesis May be due to errors in study design (e.g. more healthy people receive the treatment and more sick people receive the placebo, so the treatment looks effective even though it’s not)
α is your risk of making a type __error - it is the probability that the results____. usually it is set at _____1; will show something that is not really present, that the results were just a matter of chance; 5%
Power = 1-β
β = the probability that _____; by convention, β is generally set at ______. β is the probability of making a type _____ errorthe sample size is too small to detect a difference; 0.20. ; II
You can increase power by increasing sample size
type II error is _____. you run a risk of making this error if....missing something that is actually there.... if your sample size is too small