Wk 4 - Banishing Bias

Question 1

RANDOMISED CONTROLLED TRIALS

i) what is it?
ii) what is compared at the end?
iii) what type of test is a RCT?
iv) give three indications for a RCT

Answer 1

i) study in which participants are allocated randomly between an intervention (treatment) and a control group (no tx)
ii) compare outcomes
iii) analytic experimental - researcher manipulates circumstances to test hypotheses about exposure and outcomes
iv) ascertain safe dose of drug, monitor adverse effects of a new drug, demonstrate efficacy (does a drug work), show a drug is superior to another

Question 2

RCTS AND CONFOUNDING

i) name something that may confound results in an RCT?
ii) how is confounding protected against? what does this mean for any differences observed between groups?

Answer 2

i) if participants/investigators choose which group they are assigned to

ii) protect by randomly assigning people to groups so confounders are equally distributed amongst the groups
- means any diffs are due to intervention

Question 3

HIERARCHY OF EVIDENCE

i) which is the highest in the hierarchy of sci evidence - cohort study, case control, cross sectional study
ii) which type of analysis sits at the top?
iii) name a type of evidence that sits at the bottom

Answer 3

i) cohort then case control then cross sectional
ii) meta analyses and systematic reviews sit at the top
iii) case report, opinion paper and letters sit at the bottom

Question 4

RANDOMISATION AND EQUIPOISE

i) what is clinical equipoise? what does it provide?
ii) is it ethically acceptable to randomise patients to a treatment known to be inferior?

Answer 4

i) genuine uncertainty in the expert medical comunity over whether one treatment will be more beneficial than another
- provides ethical basis for medical research that involves giving patients different treatments

ii) yes

Question 5

RCTS

i) what is internal validity?
ii) what is external validity?

Answer 5

i) internal 0 is the exposure causing the outcome in this study?
ii) external - to what extent can these findings be generalised to other people, situation and times? (how likely do the results translate to the world in general)

Question 6

BIAS

i) what type of error being introduced to RCTs can cause bias?
ii) what is selection bias?
iii) what is performance bias?
iv) what is attrition bias?
v) what is observer/detection/information bias?
vi) what can all these biases impact on? (2)

Answer 6

i) introduction of systematic error - by selecting or encouraging one outcome over the other
ii) selection - not representative of the sample to the wider population, systematic diffs between baseline characs of group that are compared
iii) performance - systematic diffs between groups in the care provided or in exposure to factors other than the intervention of interest (eg follow up in diff locations)
iv) attrition - systematic diffs between groups in withdrawals from a study (people lost eg more control participants drop out because they are unwell)
v) observer - outcome measure does not adquately capture the outcome of interest or there are systematic diffs between groups in how outcomes are determined (eg doing ablood test in one group and a questionnaire in another)
vi) internal and external validity

Question 7

SOLUTIONS TO BIAS

i) how can selection bias be avoided? (2)
ii) how can performance, attrition, observer/detection/info bias be avoided?

Answer 7

i) inclusion/exclusion criteria, size randomisation, allocation procedures
ii) blinding or masking

Question 8

BLINDING/MASKING

i) what can people be blinded to? (3)
ii) what type of trial is it easy to blind? what ype of trial is it harder to blind?
iii) who is blinded in triple blinded trial?

Answer 8

i) exposure, disease or hypothesis (intervention/group allocation)

ii) easy to blind drug trials
- harder to blind surgery, psych intervention

iii) triple blind - patients, treating dr and study investigator

Question 9

ANALYTIC CONSIDERATION FOR REDUCING BIAS

i) what is intention to treat analysis?
ii) what is per protocol analysis?
iii) which one is a better reflection of real life?
iv) which is better to test the interventions actual effectiveness

Answer 9

i) ITT is anal of outcomes for everyone who is randomised, irrespective or whether they have or adhere to the intervention
ii) PP is to analyse outcomes of thos who recieved the intervention as specified in the protocol
iii) ITT is better for reflection of real life
iv) PP is better for actual effectiveness

Question 10

HYPOTHESIS IN AN RCT

i) what three things need to be included?
ii) what is the problem will having too few participants? (2)
iii) what si the problem with too many participants? (2) what does this increase the risk of?

Answer 10

i) testable prediction - phrased using participant intervention comparator outcomes
- alternative (experimental) hypoth eg there will be a diff,
- null hypoth - there wont be a diff between groups

ii) too few - if result is null you dont know if there is actually evidence of an effect

iii) too many - unethical, expensive
- larger samples inflate a chance of finding signif diff that arent necessarily clinically meaningful

Question 11

ERROR AND POWER

i) what is a type I error?
ii) what is a type II error?
iii) what is the significance level?
iv) what is power?
v) what rate of type II errors are we usually comfortable making?

Answer 11

i) when you observe a difference but there isnt really one
ii) when you dont see a difference but there actually is one

iii) the rate at which you say you are comfortable at making a type 1 error (type I error rate)
- 5%/0.05 significance level means you are willing to accept a type I error is made 5% of the time

iv) power - is the opposite of type II error rate eg the probability that a test will not miss an effect when it truly exists
- power tends to be set at 1-0.2 = 0.8 or 80%

v) 20% or 0.2

Question 12

i) what type of error when there is no observed diff but there is a true diff?
ii) what type of error when there is an observed diff but no true diff exists?

Answer 12

i) type II error
ii) type I error

Question 13

P VALUES

i) what is a P value?
ii) what type of error do we make sure we are not making with a P value?
iii) what does a P value of 0.12 reflect in relation to error?
iv) what level of P value is expected if there are few observations with high variability?
v) what level of P value is expected for many observations but low variability?

Answer 13

i) probability that the difference observed could have occured by chance if the groups compared were really alike (chance of getting the result if the null hypoth was true)
ii) P value - not making a type I error
iii) 0.12 - the probability of observing this result if there the null hypothesis was true
iv) high p value
v) low P value

Question 14

SAMPLE SIZE

i) what three things need to be taken into account when deciding sample size?

Answer 14

i) what is the clinically important effect/minimally clinical important difference
- enough people to detect the size of an effect

how much does the outcome score vary in the general population? (understand spread of scores eg standard dev)

how much error are we willing to allow? (signif level and power)

Question 15

give two strengths and two weaknesses of an RCT

Answer 15

strength - establish safety/efficacy of new intervetion, best single study evidence for causal association (estab temp precedence)

weakness - expensive, issues with internal and external validity

Question 16

SUMMARY

i) what is the hierarchy of evidence?
ii) what is a RCT good for? what does randomisation mean? what is equipoise?
iii) what are the four diferent types of bias and explain each
iv) wha is a type I and II error?
v) what does a P value show?
vi) adv and disadv of RCTs

Answer 16

i) HOE is the risk of bias assoc with different study designs

ii) RCT is good for single study evidence of protecting against confounding and bias
- randomisation - distrubute known and unknown confounders equally between arms of an RCT
- equipoise - genuine uncertainty about efficacy/effectiveness of an intervention

iii) selection - participants not repres/systematic diffs at baseline
observer/det/info - foesnt capture outcome of interest or captures otucomes differently between groups
performance - diff in care provided to each group
attrition - systematic diffs in groups in withdrawals from study

iv) type I - false evidence of effect
type II - false evidence of no effect

v) P val shows prob of obtaining the result if the null hypoth was true
vi) adv - tests safety and effective of new intervention, best causal evidence

disadv - expensive, tiem consuming, can lack int and ext validity