Ch 12: Biostatistics and Pharmacoeconomics Flashcards
spread
how similar or how varied values are. uses range and std deviation
gaussian distribution
normal dist, bell shaped curve of continuous values. symmetrical with 1/2 values on left and 1/2 on right. mean = median = mode. 68% of values are within 1 SD, 95% are in 2 SDs of mean. otherwise not gaussian –> skewed.
null vs alt hypothesis
null = no difference, trying to reject with statistical significance. alt = difference, trying to prove.
confidence interval
if narrow, high precision. if wide, low precision. % chance of no error with tx effect (range). if alpha is 0.05, CI is 95%. if alpha is 0.01, CI is 99%. if CI contains 0 - not significant.
alpha level
max permissible error margin (% chance of error), usually set at 5% or 0.05 as threshold for rejecting null hypothesis. smaller can be chosen but requires more data, pts, and money or larger tx effect
p value
if less than alpha - reject null. if equal or more than - accept null (failed to reject null, not statistically significant)
type I error
alpha. false pos, when null is rejected in error.
type 2 error
beta. false neg. null is accepted when should have been rejected.
study power
probability that a test will reject the null hypothesis correctly, power to avoid a type 2 error. as power inc, type 2 error dec. power = 1 - beta. power determined by number of outcome values, different in outcome rates between groups, and significance level.
relative risk (RR)
aka risk ratio. %risk tx / %risk control. 1 is no diff. >1 more risk of outcome. <1 less risk of outcome. calculated as decimal.
risk (R)
pts with event (regardless of tx or control) / #total in study
relative risk reduction (RRR)
how much risk is reduced in tx group compared to control group. (%risk control - %risk tx)/%risk control. OR 1-RR. calculated as decimal.
absolute risk reduction (ARR)
%risk control - %risk tx. includes reduction in risk and incidence rate. net effect beyond the effect obtained from placebo.
number needed to treat (or harm)
1/ARR. # patients that need to receive tx for 1 patient to be harmed or receive benefit of tx. always round up to next whole number for NNT and down for NNH.
odds ratio
estimate risks with tx in case-controls. calculate odds that outcome will occur with exposure compared to odds without exposure. (see flash where OR = AD/BC). = 1 no diff in events (CI does not cross 0). >1 tx has more events. <1 tx has less events.
hazard ratio
survival analysis uses hazard instead of risk - ex. cancer. rate of an unfavorable event occurring within short period of time. ratio between hazard rate tx and hazard rate control. = 1 no diff in events (CI does not cross 0). >1 tx has more events. <1 tx has less events.
discrete data
nominal or ordinal
nominal
categories like male and female
ordinal
has logical order like pain scale or NYHA class, steps on scale are not divisible (unlike ratio and interval data)
continuous data
interval and ratio. both continuously inc by same amt.
interval data
no meaningful zero (zero does not equal none). ex. fahrenheit temp scale, 0 is not “no temp” its cold.
ratio data
has meaningful zero (zero equals none). majority of med studies. ex. HR, 0 is death.