[JC SERIES] Episode 2: Therapy
All About Therapy
Therapy foregrounds seek to answer:
- Does a given intervention have a meaningful effect on patient outcomes?
Often answered by RCTs or noninferiority trials
Randomized Control Trials:
- Randomize study participants into TREATMENT (can be more than one) and CONTROL groups
- Treatment: receives the intervention being studied
- Control: receives placebo, nothing or some other agent
- Then outcomes are compared
- Why randomize?
- Equalize baseline characteristics of each group
- Reduce selection bias
- More likely that treatment effect is actually due to the intervention, not differences between groups
Bias
- Not good in studies!
- Goal of critical appraisal is to uncover any and all sources of bias
BEEM Appraisal Tools
We’ll be using these throughout the series – you can find all of them at the Skeptics Guide to Emergency Medicine
Quality checklist Appraisal Tool (for RCTs)
- Selection
- Who are we studying?
- Does this population make sense/generalize to your patient population?
- Recruitment
- How were the patients selected?
- Single or multi-center?
- Offered reward or compensation?
- What were the inclusion/exclusion criteria?
- Randomization
- Unblinded – both participants and researchers know assignment
- Single-blinded – participants don’t know assignment
- Double-blinded – neither participants nor researchers know assignment
- Make sure groups were treated equally outside of study variable
- Outcomes & Follow-up
- How long were patients followed? Who was lost to follow up and why?
- Attrition bias – similar patients may drop out or be lost to follow-up, a possible source of bias in the results
- Generally accepted attrition rate ~20% (so follow-up rate 80%)
- Patient important outcomes
SISSOL
Stats In 60 Seconds or Less
Want to read more about these statistical topics? Check out the User’s Guide to the Medical Literature via JAMA
Risk
Risk is a statistical measure of an event occurring as a result of exposure to some variable. In an RCT, risk answers questions like: “How likely is it for our treatment group to have outcome X as compared to our control group?”
Risk is assessed via a 2x2 table
- If RR = 1 − Risk in exposed = Risk in non-exposed − No association
- If RR > 1 − Risk in exposed > Risk in non-exposed − Positive association; ? causal
- If RR < 1 − Risk in exposed < Risk in non-exposed − Negative association; ? protective
The Number Needed to Treat
Derived from the Risk Difference – a measure of the actual difference in risk between experimental and control groups.
So in the above example, the risk difference is 0.053-0.013 = 0.04. The number needed to treat is the inverse of the risk difference (1/0.04 = 25). This means we would need to perform 25 appendectomies to prevent one wound infection (this is a fairly high NNT – we can assume that appendectomies are probably not protective against wound infections).
The number needed to harm is the opposite of the NNT how many people would we have to treat before we harm one.
Confidence Intervals
A range of values that is likely to encompass the true value. We typically use a confidence level of 95%; this means there is a 95% chance the calculated value lies within the confidence interval (but a 5% chance it doesn’t). The confidence interval is graphically depicted by a standard bell curve, with the center being the 95% confidence interval and the two outlier sections being the 5% of possible ranges outside the confidence interval.
The precision of a given confidence interval is directly related to sample size.
Want to read more about these statistical topics? Check out the User’s Guide to the Medical Literature via JAMA