Home / Statistics / Randomised controlled trials and hypothesis testing
Statistics
Randomised controlled trials and hypothesis testing
What a hypothesis is, how to conduct an RCT and the level of evidence it provides.
23 questions3 source pages
Images appear with the first
question taken from each source page — tap a question to open it.
23 questions
Q1What is a hypothesis?▸
A proposition that serves as a starting point for further investigation
Q2What is a null hypothesis?▸
A primary assumption that any differences between groups occurred purely by chance
Q3What are type 1 and 2 errors?▸
Type 1 (alpha) error: rejecting the null hypothesis when it is true - no true difference but a difference is found
Type 2 (beta) error: failing to reject a false null hypothesis - a true difference exists but is not detected
Q4How do you conduct a randomised controlled trial?▸
Identify the problem, define the research question, set null and alternate hypotheses
Literature review using the PICO principle (patients, intervention, comparison, outcome) to identify gaps; ethics committee
Study design (PROSD): population (control and treatment groups, inclusion/exclusion criteria), methodology (randomisation/blinding/stratification for confounding), outcome measures, sample size by power analysis
Register and conduct the trial, recruit patients, collect and analyse data, interpret and publish
Q5What types of analysis can be used for deviation from the study protocol?▸
Intention to treat
Per protocol
As treated
Q6What factors determine sample size in a power analysis?▸
Power of study 0.8 and P value 0.05 (predetermined)
Variability of the result (standard deviation)
Chosen clinically important difference in the primary outcome
Q7Define power and power analysis.▸
Power = the ability of a study to detect a difference between 2 interventions if one in fact exists = 1 - beta (1 - type II error)
Power analysis = the process of determining the sample size needed to reject the null hypothesis
Conventional power 80% = 80% chance of finding a statistical difference if there is one (probability of a type II error is <20%)
Cut-off determined from: pilot study, literature research, minimum clinically important difference (MCID)
Q8What are the steps of a power analysis?▸
Set the smallest meaningful outcome difference and effect size
Set alpha and beta (0.05 / 80%)
Find the variance
Q9What is effect size?▸
The magnitude of the difference in the means of the control and experimental groups
Expressed with respect to the pooled standard deviation
Q10What is bias and how is it reduced?▸
A systematic error, conscious or unconscious, that leads to a false representation of the true state of affairs
Reduced by randomisation, masking/blinding and meticulous attention to the study protocol
Q11What is confounding and how does randomisation address it?▸
Confounding occurs when factors not under study affect the results
Randomisation reduces it by distributing independent variables equally among the treatment arms
Q12What types of randomisation are used and what are their drawbacks?▸
Simple (by computer): easy to implement but may result in unequal groups
Stratified: separate randomisation procedures within subgroups defined according to the predefined characteristics (e.g. smoking)
Block: addresses imbalance but the executer can predict the next assignment
Selection bias (randomisation), information bias (recall, workup, interview)
Windowing (analysis), publication bias
Q14Differentiate type 1 and type 2 errors and how they are reduced.▸
Type 1 (alpha): no true difference but a difference is found - falsely rejecting the null hypothesis; reduce by decreasing the p value (Bonferroni correction)
Type 2 (beta): a true difference exists but is not detected - falsely accepting the null hypothesis
Reduce type 2 error by increasing p value, sample size, effect size or outcome variability
Q15What is a confidence interval and how is it calculated?▸
The range in which the true effect lies on either side of the mean; refers to the uncertainty of the study
95% CI = mean +/- 1.96 SD
99% CI = mean +/- 2.58 SD
Q16What is a p value?▸
The probability of an observed difference occurring by chance; p < 0.05 is taken as statistically significant
Q17What is the Bonferroni correction and what is its drawback?▸
A post-hoc statistical correction made to P values when several dependent or independent statistical tests are performed simultaneously on a single data set
May increase a type 2 error
Q18What determines the level of study of a randomised controlled trial?▸
Depends on the quality of the RCT: confidence interval, percentage of follow-up, blinding
Can be level 1 or 2
Q19Outline the levels of evidence.▸
Level 1: RCT with narrow CI; systematic review of RCTs with homogeneous findings
Level 2: cohort study, RCT with <80% follow-up; systematic review of cohort studies with homogeneous findings
Level 3: case control study; systematic review of case control studies
Level 4: case series; Level 5: expert opinion
Q20What are the types of study designs?▸
Descriptive: cross sectional, case reports, correlational studies
Analytical: cohort, case control, RCT, survival analysis
Q21What is the difference between a cohort and a case control study?▸
Cohort study: observational design where patients are selected on the basis of an exposure variable
Case control study: observational design where patients are selected on the basis of an outcome variable
Q22What is the difference between a systematic review and a meta-analysis?▸
Systematic review: combines the information of several studies on the same topic
Meta-analysis: collects the results of several studies and analyses them again with a statistical technique
Meta-analysis is a type of systematic review
Q23What is heterogeneity and how is it measured?▸
Variation between the studies' design (clinical, methodology)
Chi-squared heterogeneity test (Q test): large value = heterogeneity