Define positive predictive value
probability a patient with a positive test actually has a disease
PPV = TP / (TP + FP)
an estimate of the variability of each individual data point from the mean
define effect size
magnitude of difference in means between two groups
case control studies (retrospective) typically generate what statistical measure?
Define Type I error (alpha)
null hypothesis is rejected even though it is true
negative likelihood ratio: define and formula
how the likelihood of a disease is changed by a negative test result
what is a type II error?
detecting no difference when there is one
accepting a null hypothesis wrongly
typically set at 0.8
Describe Post-test odds of disease
post-test probability = (pretest probabililty) X (likelihood ratio)
- likelihood ratio = sensitivity / (1 - specificity)
- pre-test odds = pre-test probability / (1 - pre-test probability)
post-test probability = post-test odds / (post-test odds + 1)
What does positive predictive value depend on?
prevalence of a disease
Define effect size
magnitude of the difference in the means of the control and experimental groups in a study with respect to the pooled standard deviation
Effect sizes are normally used for continuous variables in contrast to relative risk reduction which is used for dichotomous variables
NNT formula and definition
number needed to treat to get one additional favourable outcome
Define negative predictive value
probability a patient with a negative test actually has no disease
= TN / (FN + TN)
a highly ____ test with a (neg/pos) result can rule (in/out) the outcome of interest
SENSITIVE tests with NEGATIVE results rule OUT the outcome
SPECIFIC tests with POSITIVE results rule IN the outcome
number of newly reported cases of a disease during a given time period
What test do you use to compare 2 independent means from numeric data?
numeric data = t-test
How do you determine absolute risk reduction?
NNT = 1 / ARR
ARR = 1 / NNT
ARR = (risk in control group - risk in experimental group)
*this was on a previous exam
test for comparing means of 3 or more continuous dependent variables
(each with categorical independent variables)
this is essentially a t-test for 3 or more groups
ANOVA on 2 groups will give the same result as a t-test
Define Type II error
a false negative difference that can occur by:
- detecting no difference when there is a difference
- accepting a null hypothesis when it is false and should be rejected
used to compare means of continuous data that is normally distributed
What does a relative risk > 1 mean?
incidence of the outcome is greater in the exposed/treated group
test to see if two continuous variables are related or not
e.g. comparing age to BP
Kaplan-Meier equation (in laymans terms)
The number of failures / total number still being followed
statistical tool in meta-analyses to detect publication bias
describe negative likelihood ratio
describe how the likelihood of a disease is changed by a negative test result
negative likelihood ratio = (1 - sensitivity) / specificity
define relative risk and give formula
- risk of developing disease with exposure compaerd to risk of developing disease without exposure
- risk of disease with exposure=(have exposure and disease)/(all with exposure)
- risk of disease without exposure=(no exposure but have disease)/(all without exposure)
- RR=(risk of disease with exposure)/(risk of disease without exposure)
Mann-Whitney or Wilcoxon rank sum tests
comparing means of non-continuous data
Describe Relative risk
risk of developing disease for people with known exposure
risk of developing disease for people without known exposure
number of new cases in a given time period
How do you compare categorical data?
What is the fisher exact test used for?
Comparing proportional or categorical data when sample sizes are small or number of occurences in a group is low
positive likelihood ratio; formula and definition
how likelihood of a disease is changed by a positive test result
What does blinding do?
minimizes observer bias
Define positive likelihood ratio
describes how the likelihood of a disease is changed by a positive test result
positive likelihood ratio = sensitivity / (1-specificity)
define likelihood ratio
likelihood a result is expected in someone with the disease compared to likelihood the result is expected in a normal person
An underpowered study is prone to what type of statistical error?
Type II error (beta)
Define Number needed to treat
number of patients that must be treated in order to achieve one additional favourable outcome
NNT = 1 / absolute risk reduction
What does randomization do?
Decreases selection bias
What test do you use to compare 2 means with non-parametric (non-normal) data?
mann-whitney or wilcoxon sum rank test
Name 2 ways of minimizing the effects of chance in study design
Having an adequate sample size based on power calculations
The use of appropriate levels of significance in hypothesis testing
What is an independent risk factor for postop MI post total joint arthroplasty?
What can you do to prevent it?
Administration of beta blockers for 7 days decreases risk of cardiac ischaemic events and in-hopsital deaths
What is a funnel plot?
Is a simple scatter plot of the intervention effect estimates from individual studies against some measure of each study’s size or precision
Is used to detect publication bias in meta-analyses
probability that test results will be positive in patients with the disease
= TP / (TP + FN)
= people with the disease who tested positive / everyone with the disease
What test do you use to compare 2 means with normal (parametric) data
Define false negative
Patients with the disease but have a negative result
False negative rate = FN / (TP + FN)
= FN / everyone who is positive
Describe Odds ratio
probability of having a risk factor if one has a disease
obtained from case control studies (retrospective studies)
OR = (odds of developing disease in exposed patients) / (odds of developing disease in unexposed patients)
test to show how a change in a variable (categorical or continuous) affects another variable (categorical, dichotomous)
e.g. seeing if body weight is related to presence or absence of lung cancer
define power (wrt stats) and give the formula
estimate of the probability a study can detect a true effect of the intervention
number of cases of something at a single point in time for a given population
probability that the test result will be negative in patients without the disease
= TN / (FP + TN)
= people without disease who tested negative / everyone who doesn't have the disease
What does PPV depend on?
prevalence of a disease
In a screening test, you want it to be highly _______________
What kind of studies do you get odds ratio's from?
retrospective studies (case-control)
b/c it is the proability of having a risk factor given a person has the disease
Define False positive
patients without the disease who have a positive test result
false positive rate = false positives / (FP + TN)
= FP / everyone who is negative
Where do you get a relative risk ratio from?
b/c you need incidence to calculate it
test for comparing means of 2 continuous dependent variables
(each with categorical independent variables)
e.g. comparing average BP (the continuous dependent variable) in a group of smokers vs. nonsmokers
(smoking is the categorial independent variable)
compare means of 3 or more independnet groups in normally distributed data
test to compare 2 or more categorial dependent variables
(each with categorial independent variables)
e.g. comparing # of ppl with ecoli who ate burgers with #ppl with ecoli who didnt eat burgers
(# of ppl is categorical)
i.e. if you can fit your data into a 2x2 table, use a chi-square test
(ate burger, didnt eat burger vs. ecoli, no ecoli)
In a Kaplan-Meier analysis, what do you do with the patients that are lost to followup
Exclude them from the study
They are assumed to be similar to the patients still in the study
define odds ratio and give formula
probabilyt of having a risk factor given you already have the disease
OR=(odds of getting disease when exposed)/(odds of getting disease if unexposed)
What is a bonferroni correction?
post-hoc statistical correction made to P values when several dependent or independent statistical tests are being performe dsimultaneously on a single data set
What do cross-sectional studies aim to achieve?
Identify the prevalence of a condition
what is a type I error?
rejecting null hypothesis when it's true
we typically set this at 0.05
Describe a confidence interval
The Interval that will include a specific parameter of interest, if the experiement is repeated
Define alpha level?
probability of a type I error occuring (reject null when its true)
typically set at 0.05
What studies are generally reported as an odds-ratio?
mann-whitney sum rank tests
chi quare test
fischer exact test
are all types of what?
used to test specific hypotheses about associations or differences among groups of subjects/sample data
test to compare two categorical variables with small sample sizes
fisher exact test (similar idea to chi-square but with small sample sizes)
for samples <5 or total of all cells <50 (in your 2x2 table) - see chi-square card
What does NPV depend on?
prevalence of disease
Define Power (stats)
an estimate of the probability a study will be able to detect a true effect of the intervention
power = 1 - (probability of a type II, or beta, error)
total number of cases of a disease present in a location at any time point
what is a bonferroni correction?
post-hoc statistical correction to P-values when several tests are performed simultaneously on a single data set
What studies are generally reported as relative risk?
define Likelihood ratio
likelihood that a given test result would be epxected in a patient with the target disease
likelihood that the same result would be expected in a patient without the target disease
What does matching in a study design do?