statistics Flashcards
What does a case series do?
tracks subjects with a known exposure
What does a cross-sectional study do?
uses data fro a population at a specific point in time
What does a case-control study do?
2 existing groups differing in outcome are identified and compared based off causal attribute
What do cohort studies do?
you map a cohort and perform cross-sectional itnervals over time
What is the hierarchy of evidence?
systematic reviews critically-appraised topics critically-appraise individual articles RCTs Cohort studies Case-controlled studies background information / expert opinion
what are confounders?
another variable associated with the outcome of interest and independent variable.
What is reverse causality?
because we are recruiting from the point of outcome, , the outcome may be affecting the variable
(oesophageal cancer may cause people to drink hot tea)
What do RCTs do?
similar people are randomly assigned to 2 (or more) groups to rest an intervention
what do systematic reviews do?
go through all the literature to identify every published (and some unpublished ones too) to answer the question we are posing
What are the 2 types of quantitative data?
continuous
discrete
What are the types of categorical data?
2 categories - binary
>2 categories - nominal / ordinal
What would you use to describe information if there are no outliers?
mean
standard deviation
What would you use to describe information if there are outliers?
Mean
interquartile range
What would a small SD suggest about a mean?
it is more useful
What sort of study design would be used to measure treatment effect?
RCT
What sort of study design would be used to measure exposure effect?
observational studies
In 10 000 control patients, 1 200 had a subsequent vascular event within 1 month…
What is the probability of a vascular event?
1 200 / 10 000 =
0.12
In 10 000 control patients, 1 200 had a subsequent vascular event within 1 month…
What is the percentage of those with a vascular event?
12%
In 10 000 control patients, 1 200 had a subsequent vascular event within 1 month…
What is the risk of a vascular event?
12 per 100
In 10 000 control patients, 1 200 had a subsequent vascular event within 1 month…
What are the odds of a vascular event happening?
1200 / 8800 =
0.14
What is relative risk?
probability of event on treatment / probability of event on control
What does a risk ratio of 1 suggest?
risk equal in intervention and control arm
What does a risk ratio > 1 suggest?
risk of outcome greater in treatment arm
What does a risk ratio < 1 suggest?
risk of outcome less in treatment arm
If you were looking at an exposure, what might you use instead of ‘treatment’ or ‘control?
risk in the exposed group
risk in the unexposed group
Name 3 ways you might interpret the relative risk
as a percentage increase or decrease in risk
as percentage of the risk in the other arm
risk in the intervention arm is RR times the risk in the control arm
What is the number needed to treat?
The number of patients, who on average need to be treated to prevent one event that would otherwise occur
NNT is 1/absolute risk difference
How do relative and absolute effects compare?
- relative measures are commonly reported
- relative measures can look large where an event is rare
- absolute measures are less susceptible to misinterpretation
NNT is an effective way to communicate to lay population
What are odds?
number with event / number without event
What is an odds ratio?
the odds of event on treatment / the odds of event on control
how does odds ratio compare to relative risk?
it’s a bit less intuitive
What is a mean difference?
mean in group 1 - mean in group 2
What might you use to measure treatment / exposure effects in continuous outcomes?
mean difference
What might you use to measure treatment / exposure effects for a binary outcome?
relative measures (risk ratio / relative risk, odds ratio)
absolute measures (risk difference)
What is sample variability?
the difference between truth (parameter) of the population and sample (statistic)
What is a parameter?
a value referring to the population that we cannot know
What is the standard error?
- describes the variability in the means
- tells us how accurate the mean of any particular sample is compared to the true population mean
the mean of 95% of the samples is within 2 SEs of each other
What does a large standard error suggest?
the mean from each sample are likely to differ a lot and so could be an inaccurate representation of true population mean
What does standard deviation do?
describes the variability in a sample.
essentially, how far, on average, a measurement is from the mean
about 95% of the sample have an observation within SDs of the sample mean
How does SE suggest how certain we are about estimates?
Small study - large SE - uncertain
large study - small SE - more certain
Which of SD or SE is usually bigger?
SD
What does a confidence interval do?
suggests how accurate our estimate is likely to be
expresses a range of values which we are pretty sure the population parameter lies in (centre being the sample mean)
What is the size of the confidence interval affected by?
1 - variation within the population
2 - sample size
3 - how confident do we want to be?
What conditions might make the confidence interval really wide?
greater variation in population
small sample
we want to be really accurate
What does a ‘95% confidence interval’ suggest?
95% of the time, it would contain the true mean
How would you calculate the lower value of a confidence interval?
2 SEs below the mean/exposure effect
How would you calculate the upper value of a confidence interval?
2 SEs above the mean/exposure effect
What does a 95% CI for a RD above 0 (null value) suggest?
95% certain the risk in the treatment arm is greater than in the control arm
What does a 95% CI for a RD below 0 (null value) suggest?
95% certain that the risk in the treatment arm is less that the control arm
What does a 95% CI for a RD containing 0 (null value) suggest?
there is not enough evidence to say that the risk is different in the treatment arm compared to the control arm
Name the 5 steps required to construct a hypothesis test
specify null and alternative hypothesis assume the null hypothesis is true and calculate test statistic convert to a p-value assess the evidence interpret the result
What does a null hypothesis suggest?
(H0) there is no difference
What would an alternative hypothesis suggest?
(H1) there is a difference
What is a p value?
the probability that the data could have arisen if the null hypothesis H0 were true
What does a high p value tell us?
high chance of seeing the difference we’ve seen if the null hypothesis were true
What does a smaller p value suggest?
greater evidence that the null hypothesis is not true
How is statistical significance assessed?
p-value
How is clinical importance assessed?
estimates
confidence intervals
Is a 95% CI contains the null value, then what is the p value likely to be?
> 0.05
If a 95% CI does not contain the null value, what is the p value likely to be?
<0.05
If a 95% CI ends at the null value, what is the p value likely to be?
0.05
What are the 3 categories used when comparing 2 CIs?
significant difference
unclear if significantly
different if boundaries overlap but means don’t
not significantly different if means overlap
What is a forest plot?
a way of summarising results of a systematic review
In a forest plot, what sort of studies are given a larger weight?
larger studies