Clinical Trials in CVD Flashcards
What is the power of clinical trials?
ability to control bias and confounding
What are clinical trials?
- longitudinal
- designed to assess if an intervention (removal of exposure) changes the incidence of an outcome
- most are expected to decrease incidence
- most involve a control group for comparison
- prospective follow-up to capture outcomes
What does evidence from clinical trials provide?
- the ‘gold standard’ for causality by:
- actively changing the exposure status to see if incidence of disease or outcome of interest is reduced relative to a comparator
- in a tightly controled study environment to minimize confounding and bias
- they provide most of the evidence for EBP
What relative measures of intervention effect are used in CTs?
- relative risk
- comparison of the incidence measures between the intervention and control groups
- hazard ratios
- interpreted similarly to relative risk
What absolute measures of intervention effect are used in CTs?
- absolute risk/rate reduction
- number needed to treat
What are the key outcomes of CTs?
- relative:
- relative risk
- hazard ratio
- absolute:
- absolute risk/rate reduction
- number needed to treat
- survival analysis (explicit consideration of time-to-event)
Randomization deals with
confounding
Confounding is
a mechanism through which distortion of truth can occur
Masking or blinding deals with
information bias
How is information bias dealt with?
- blinding/masking subjects and/or investigators
- blinded/masked committee who decides if an outcome has occured based on pre-described, objective criteria to reduce subjectivity
What is intention-to-treat analysis?
- deals with selection bias
- those who drop out are almost always systematically different from those who don’t
- ignores drop-out and crossover when analyzing subject data
- ie treating subject data as they were intended to be treated, not the group they crossed over to
- underestimates treatment effect (conservative estimate) because the groups have become more similar
What is survival analysis?
- during follow-up, explicit capture of outcomes and their time of occurence
- measured as time to event
- plotted on a Kaplan-Meier curve, a plot of hazard or survival (1-hazard) over time
- survival = avoidance of event (not necessarily death)
- can note from KM curves whether incidence of a disease is different in treatment vs control, and when that difference occurs for differen tdiseases (circles)
What is hazard ratio?
- Hintervention:Hcontrol
- conceptually similar to relative risk, interpreted similarly
- but relative risk applies only at a specific time period
- outcome of survival analysis
-
applies to the whole period of follow up
- a weighted average of the whole period of follow up
- example:
- if HR = 0.5, then at any given point in time within the period of follow-up, probability of an outcome in the intervention group is half that of the control group
What is the interpretation of the HR for CHD and stroke?
- HR for CHD is 1.29
- HRT compared to placebo leads to a 29% increased risk (1.29x the probability of CHD over the period of follow up)
- HR for stroke is 1.41
- HRT compared to placebo leads to a 41% increased risk (1.41x the probability of stroke over the period of follow up)
How is relative risk calculated?
- rate of outcome (/py) in intervention divided by rate of outcome in control e.g.:
- if control is 10/100py and intervention is 7/100py
- RR = 7/10 = 0.7
- this equates to a 30% relative reduction in likelihood of the outcome conferred by the intervention
- if control is 10/100py and intervention is 7/100py
How is absolute risk/rate reduction calculated?
rate of outcome in control arm e.g. 10/100py minus rate of outcome in intervention arm e.g. 7/100py
ARR = 10/100py - 7/100py = 3/100py
How is number needed to treat calculated?
NNT = 1/ARR (absolute risk/rate reduction)
e.g. if ARR = 3/100py, NNT = 100/3 = 33.3 people per year
must be reported with time reference
What influences the number needed to treat?
- relative effect (often constant)
- underlying likelihood of the outcome
- e.g. if likelihood of outcome is 10% less, NNT needs to be 10x more:
- ARR = 3/100py, NNT = 33.3
- ARR = 3/1000py, NNT = 333
- e.g. if likelihood of outcome is 10% less, NNT needs to be 10x more:
What is the number needed to harm?
- measure of what interventions increase risk/rate of outcome
- number of people who would need to undergo an intervention for one of them to be harmed from it
- e.g. if rate of outcome in control is 10/100py, and rate of outcome in intervention is 14/100py (ie higher):
- RR = 14/100py / 10/100py = 1.4 (same as 0.7 benefit)
- ARR = 10/100py - 14/100py = -4/100py
- tf NNT = -25; negative implies NNH
- tf NNH = 25
