Stats Flashcards
General structure for critical appraisal
Question + relevance
Population - characteristics, how many, how were they recruited
Intervention
Control - what is the comparison - usual care? placebo?
Outcome - primary vs secondary outcomes?
Validity - internal and external
Ethics - approval, Helsinki, four pillars?
Funding - who funded it?
Conclude - question, strengths, weaknesses, key outcomes, implications
Answer to does this change practice?
Likely no - this is one single study, thinking about hierarchy of evidence you need systematic review/meta analysis
What’s research equipoise?
Genuine uncertainty about the therapeutic merits of each arm
What is the p value
Probability that an outcome could have happened by chance - p<0.05 is significant, p<0.01 is highly significant
Confidence interval definition?
There is an XX% chance that true value lies within the interval
Types of bias
Selection bias - how patients were chosen
Performance bias - if patients’ performance was influenced/could influence results
Observational bias - if researcher’s observation was influence/could influence results
Attrittion bias - patients leaving the trial unequally
Confounding - other factors
Ways to mitigate bias
Randomisation, multi-centre
Blinding
Intention to treat vs per protocol analysis
Confounding - randomisation or matching for equal distribution of confounders, stratify by confounders, multivariate analysis
Intention to treat vs per protocol
ITT - maintains effect of randomisation, reduces risk of selection bias, more representative of real life
Per protocol - shows whether the intervention was effective in those who fully adhered
How to calculate 95% CI
= sample mean +- 1.96 x standard error
incidence vs prevalence
Incidence - new cases in specified time period, prevalence - proportion of population who have illness
Absolute risk
Number of events in group/number of people in that group
Relative risk
Incidence in treatment / incidence in control
Absolute risk reduction
Risk in control - risk in treat
Number needed to treat
1/absolute risk reduction
Odds ratio (used in case control)
Odds of the exposure amongst cases / odds of the exposure amongst controls
Hazard ratio
Risk of outcome in exposed group/ risk of outcome in non-exposed group
Type 1 error vs Type 2 error
Type 1 - false positive ; type 2 - false negative
How to avoid a type 1 error
Increase sample size. Reduce p value to 0.01 and significance level (CI)
How to avoid a type 2 error
Increase sample size. Increase p value and significance level
What does a bigger box on a forest plot represent
Study with more weight in the meta analysis - could be because more statistical power, larger sample size
what does a diamond represent on a forest plot
the combined results of the trial
Sensitivity
How many of those who actually have the disease will test positive
Specificity
How many of those who don’t have the disease will test negative
T-test
Comparing differences between 2 groups
Chi squared
Measure of the difference between observed and expected frequencies
How many observations will be within 1 SD
68%
How many observations will be within 2 SDs
95%
Mann Whitney
Like t-test - testing for differences between groups but when we can’t assume normal distribution
Positive predictive value
probability that following a positive test result, that individual will truly have that specific disease.
Negative predictive value
probability that following a negative test result, that individual will truly not have that specific disease.
what does spearman’s rank tell you
Measure of correlation between 2 variables +ve or -ve
What does a Kaplan-Meier show
Cumulative survival probabilities - steeper slope=higher death rate ie. worse survival prognosis
