RCTs and comparing groups Flashcards
What is a RCT?
A study in which participants are allocated randomly between an intervention (e..g treatment) and a control group (e.g. no treatment or standard treatment)
What are the safety reasons for conducting trials?
- Ascertain the safe dose of a new drug
- Demonstrate safety and tolerability of a new compound
- Monitor adverse events of a new drug (against existing/placebo)
What are the efficacy/effectiveness reasons for conducting trials?
- Demonstrate efficacy of new drug- does it work?
- Show that treatment T is superior or equivalent to treatment X
- Demonstrate effectiveness, and cost-effectiveness, A vs b
Why randomise?
When looking at cause-effect relationships, randomisation allows all random factors (confounders) apart from the proposed cause to be held constant between groups
Field and hole (2003)- the only way to infer causality is through comparison of 2 controlled and identical situations
What is a confounder?
A confounder is a variable that influence both the dependent variable and independent variable causing a spurious (fake) association
How do you defend a study against the influence of confounders?
RANDOMISATION with a sufficiently large sample
Randomisation ensures all potential confounding variables will be distributed by chance across all groups
What is equipose?
Provides the ethical basis for medical research that involves assigning patients to different treatment arms of a clinical trial
Means there is genuine uncertainty over which treatment is more beneficial
Unethical to randomise patients to an arm of a trial which is known to be inferior.
What is internal validity?
What is external validity?
Is the independent variable causing the dependent variable in this study?
To what extent can these findings be generalised to other people, situations and times
How does bias occur in RCTs?
When systematics error is introduced into sampling or testing by selecting or encouraging one outcome or answer over others
May not be resolved by randomisation
What is the relationship between bias, sample size and statistical significance
How does this compare to random error?
Bias is independent of both sample size and statistical significance
Random error results from sampling variability and which decreases as sample size increases
Describe the following types of bias
a) selection
b) performance
a) Systematic differences between baseline characteristics of groups that are compared. Not representative of wider population. Can occur when participants are asked to volunteer or setting of recruitement (participant of diabetes trial taken from inpatients)
b) Systematic differences between groups in the care that is provided, or in exposure to factors other than intervention of interest
Describe the following types of bias
c) Attrition bias
d) Observer/Detection bias
c) Systematic differences between groups in withdrawals from a study e.g. one drug has more side effects
d) Systematic differences between groups in how outcomes are determined and information collected from groups. Outcome measure does not adequately capture outcome of interest
Describe
a) intention to treat analysis
b) on-treatment analysis/per- protocol analysis
a) Method for analysing results in a RCT where all participants are included in the statistical analysis and analyzed according to the group they were originally assigned, regardless of what treatment (if any) they received.
B) Only participants who finish treatment according to the study protocol are analysed. It can introduce attrition bias
Why blind trials?
Can reduce or eliminate experimental biases that arise from a participants’ expectations, observer’s effect on the participants, observer bias
What can you blind in a trial?
Group allocation/intervention
- Exposure
- Disease status
- Hypothesis
What types of trials can be blinded and what cant?
Drug trials are easy to blind
Surgery vs non-surgery, psychological intervention are harder
- could blind the hypothesis (hidden superiority trial)
Who is blinded in a
a) open RCT
b) single blinded RCT
c) double blinded RCT
d) triple blinded RCT
a) Everyone involved in trial knows
b) Blinding of patients
c) Blinding of patients and treating physicians
d) Blinding of patients, treating physicians and study investigators
What is a p-value?
How does it relate to observations?
- the probability that the null hypothesis is true/ observing the result you got
- the smaller the p-value becomes, the more likely that the null hypothesis is disproven.
- if p <0.05 (threshold) reject the null hypothesis
- few observations with high variability -> high p value
- many observations with low variability –> low p value
What you need to calculate sample size?
- Estimate the clinically important effect: the smallest difference in outcome, on average which will demonstrate a clear advanatge of one tretament over another
- This size of effect is determined from previous lliterature/exposure
- Understand spread of scores (standard deviations)
- Decide how much error can be allowed
Why is it important to calculate sample size?
- If too few participants, if you get a null result you cant tell if this is evidence of no effect or if there is no evidence of an effect
- If too many, its unethical and expensive
How does standard error change with sample size?
As sample size increases, error decreases
You should consider data properties when comparing them
Give an example of categorical and scare variable data
CATEGORICAL
- Binary: yes/no
- Categories with no order- smoking (neverm former, current)
- Ordered category- socio-economic status, likert scales
SCALE variables
- age, no. of symptoms, birth weight, BP
What assumptions do parametric tests make?
That the data is normally distributed
- Y axis = no. of participants with each score (frequency)
- X axis= each possible score
Plotted on histogram
What are the problems with skew?
Fix?
- the mean is different from the median (middle value)
- the mean is highly influenced by the tail of the distribution
- the median is therefore a better representation of the data
- give IQR instead of sd
What descriptive analysis would you use for
a) scale data
b) categorial data
a) mean and sd. If skewed, median and IQR
b) calculate frequencies and present as percentages
Which statistical tests would you do in the following circumstances?
a) Does one group get higher scores (scale) than another?
b) Does one group have a high proportion of an outcome than another (categories)?
a) T test, or ANOVA
b) Chi squared test
When can a t test be used to compare continuous variables between groups?
If the following assumptions are met:
- RANDOMLY SAMPLED (each data point in population has an equal chance of being included in sample)
- INDEPENDENT OBSERVATIONS (the value of one observation does not influence other observations)
- outcome data for each group are NORMALLY DISTRIBUTED
- EQUAL VARIANCES for each group
- LARGE enough sample
What are the results of a t test?
You are testing for a SIGNIFICANT DIFFERENCE
T-statistic and accompanying p value
When do you use an ANOVA?
- You do an analysis of variance when you are comparing more than 2 groups
When do you do a chi squared test?
What assumptions are made?
Applied to establish whether there is a significant difference between two groups of categorical data. It works by comparing the observed data with the data that could be expected if the null hypothesis was true.
Assumptions:
- randomly sampled
- independent observations
- large sample
What are the results of chi squared test?
You are testing for an ASSOCIATION
chi-squared (χ2) statistic and accompanying p-value
What are non parametric tests
Adv vs d.adv
“Distribution free tests”-
Fewer assumptions (but less power)
What are the non parametric tests which replace these parametric tests where the data is not normally distributed
a) t Test
b) ANOVA
c) Correlation coefficient
a) Mann- Whitney U Test /Wilcoxon Rank Sum
b) Kruskal Wallis
c) Spearman’s Rho
Describe the Mann-Whitney U test
What is it? When is it used? Assumptions? What is tested? Test output?
- Alternative for t test
- Used to compare non-normally distributed continuous variables between groups
- Randomly sampled and independent observations
- Tests differences between variables in groups
- The p value
