click below
click below
Normal Size Small Size show me how
WEEK 7:
Experimenting on people- RCTs:
| Question | Answer |
|---|---|
| randomised controlled trial | planned experiment involving patients and is designed to elucidate most appropriate method of treatment of future patients with a given medical condition |
| difference in outcome only due to | intervention (so can establish causation) |
| key design features of RCTs (4) | choosing a control treatment, randomisation to study group, allocation concealment, and blinding |
| control group includes (2) | standard treatment (is new treatment better than old) and placebo (does treatment work at all or better than placebo) |
| placebo | an inert substance which looks, tastes and is packaged identically to the comparison substance |
| placebo effect | A beneficial effect produced by a placebo drug or treatment, which cannot be attributed to the properties of the placebo itself, and must therefore be due to the patient's belief in that treatment |
| nocebo effect | reversed placebo effect - expectation of negative effects increases likelihood of unpleasant sensations |
| how do RCTs remove selection bias | equal chance for participants to get allocated to treatment/control group |
| how do RCTs minimise confounding variables including unknown confounders | big sample with both groups being comparable in all aspects (other than their treatment allocation) eg sex and age |
| allocation concealment | masks the randomisation process before randomisation (Study team and patients no know what group next participant will be randomised to) |
| allocation concealment prevents what (2) | selection bias and confounding |
| blinding | masks allocation of treatment AFTER randomisation (participnts and researchers should not know which treatment is being received) |
| single blinding | one party (participants or investigators) are unaware |
| double blinding | participants and study staff are unaware |
| open label | participants and study staff are aware of allocation/ treatment |
| blinding prevents | ascertainment bias |
| how does blinding prevents ascertainment bias | assessment of the outcome is distorted by the participants and/or researchers |
| effect of inadequate finding (according to a review) | not blinding makes treatment look more effective than it really is (Exaggerated benefits) |
| potential explanations for findings of observational studies (4) | Chance – minimised in an RCT by studying enough people Confounding – removed in an RCT by randomisation and allocation concealment Bias – removed in an RCT by allocation concealment and blinding Truth |
| well designed large RCT can remove | chance, confounding and bias so effect is seen due to intervention (Causation) |
| how is bias removed in RCT | by allocation concealment and blinding |
| how is confounding removed in RCT | removed in an RCT by randomisation and allocation concealment |
| how is chance removed in RCT | – minimised in an RCT by studying enough people |
| problem with ethical considerations in RCTs | - can only test potentially beneficial interventions e.g. can’t test whether smoking causes lung cancer with an RCT. |
| problem with exploring social factors and demographics in RCTs | cant randomise ethnicity, age, poverty, sexuality, educational backgroun |
| problem with expense and time and logistics in RCTs | expensive, sometimes logistically unfeasible and can take long time eg breast feeding and adult onset of diabetes |
| problem with generalisability in RCT | may not be generalisable - controlled nature can be artificial |
| problem with participants in RCT | situations where participants are unwilling to be randomised or are non compliant |
| problem with selective samples in RCT | RCTs often exclude children, elderly, pregnant women, vulnerable populations, people with co-morbidities). Historically, women and BAME people have been under-included |
| suitable outcome measures are (3) | defined in advance, focus on primary hypothesis and outcome, secondary outcome included |
| why are suitable outcome measures defined in advance | What, when and how will outcomes be measured? Prevent ‘data dredging’ (or ‘p-hacking’) Measurements to check for differences between groups at baseline, adverse effects during trial, final measurements to compare effect of treatments |
| why are suitable outcome measures focused on primary hypothesis and outcome | used in sample size calculation - how many study participants do you need to test your hypothesis |
| why are suitable outcome measures linked to secondary outcomes | often includes side effects or sub group analyses |
| what makes an ideal outcome | Appropriate and relevant to patient/clinician; valid, attributable and sensitive to effect of treatment; reliable and robust to different people doing the measuring; simple and cheap to measure. Bear in mind timeliness – may need proxy measures. |
| attrition bis | systematic error where losses to follow-up are related to the study group |
| how to reduce as much attrition bias as possible | Clear information regarding expectations given at recruitment Ensure follow-up schedule is practical and not too onerous |
| non compliance | patients do not follow treatment protocol |
| how to reduce non compliance as much as possible | Clear and simple instructions Checks on compliance e.g. tablet counts at end of study Active monitoring e.g. blood or urine tests |
| per protocol analysis | analyses those who participated in study according to protocol (removes from analysis participants who did not comply with study protocol) - identifies effect which would occur under optimal conditions (true efficacy) |
| problem with per-protocol analysis (explanatory) | no longer protected from confounding effects as non compliers may differ from compliers in non random ways |
| intention to treat analysis (pragmatic) | includes everyone you intended to treat including non compliers in analysis (pragmatic as this is how the world works- real world conditions) |
| so how should trials be done (explanatory or pragmatic) | pragmatic (intention to treat analysis) unless there is good reason to do otherwise |