ECBM Lecture 1--stats review Flashcards
scientific method
prove or disprove hypothesis
precision
Target: clustered in same spot
accuracy
Target: hit bullseye
necessary characteristics of research
Objectivity precision verification economic reasonable
Research: Observational
Exploratory –> often when thing under observation would be unethical
RCT
randomized control trial
Comparitive analysis
research on drug A vs placibo
research on drug B vs placibo
comparison
Strongest evidence in clinical research?
Randomized control trial RCT
Observational study divisions
Cohort– groups with/ without exposure –>outcome
Case Control– grps exposure with/without disease–> outcome
N
population–compare pt to population
n
sample–characterized by statistics
n statistics
draw conclusions and broadly apply
variable being manipulated (the intervention)
Independent variable
any variable being measured (the outcome)
dependent variable
whole units of data Qualitative
discrete –
data with range
continuous
ranked data (1st, 2nd, 3rd)
ordinal scale
ordered data w/out a meaningful zero (water temp)
interval scale–0 temp is an actual temp
bar graph with no spaces–for trend
histogram
most frequent #
mode
value in middle
median
standard deviation
68% of data to either side of mean on line graph = confidence interval
confidence interval most often used
95% confidence – related to P needing to be less than 0.05
trustworthy graph
tall and narrow–wide curve not precise
when asked “what is the confidence interval?”
compare upper and lower limits–not 95%
empirical rule
when you have a normalized bell curve–two standard deviations from mean will give 95% confidence interval
null hypothesis is rejected when true
Type I error–ex. assume you have enough $ for groceries, but you don’t
the null hypothesis is kept when it is fake
Type II error– you assume you don’t have enough $, but you don’t
Better error to make in medicine?
Type II error–underestimate drugs efficacy
Regression
predict one factor from existing graph/data
how benefitial is drug
treatment effect
which numbers for 2X2 square?
think of scenario
read abstract
2X2 – the scenario
always deal with PEOPLE–may need to add/subtract groups
2X2 – the abstract
avoid scores–look for PEOPLE–may be in percentages
For 2X2– what is the % of people who something happened to
Event Rate (EER)–incidence of some thing–may be in abstract
First step
calculate
EER (experimental event rate) and
CER (control event rate)
second step
compare EER and CER (risk ratio, relative risk)
see ratio think…
divide RR = EER/CER
100 minus risk ratio– 100-X= Y
relative risk reduction RRR
how many people would I need to treat with drugX to save one life
number needed to treat NNT
NNT
100%/ARR (in %)
apply NNT math to negative outcomes
number needed to harm NNH
odds for treatment/ odds for control
odds ratio (OR) – Odds vs risk
odds
“yes” column divided by “no”
positive test has diseas
true positive TP
negative test has disease
False negative FN
positive test doesn’t have disease
False positive FP
negative test doesn’t have disease
true negative
sensitivity=TP/(TP+FN)
FN will impact sensitivity most–only on one side of equation
true positive rate (% positive test results in pt who have the disease
SnNOUT
snesitivity–“SeNsitivity means Negative rules ““it”” OUT”
SpPin
“Specificity means Positive rules ““it”” in”
SpPin
low false positivity–
true negative/true negative + false positive
SnNout
low false negativity=
true positive/true positive + false negative
sensitivity and specificity equation key
look for the data point on only one side of equation–this factor will play in more to results
sensitivity and specificity combined
Likelihood Ratios
TPR + FNR =
TNR + FPR=
100% – always
default hypothesis that A and B have no correlation i.e. drug has no effect
null hypothesis
