Presenting the Evidence Flashcards
Define point estimate, and confidence interval.
Point estimate (usually the mean) indicates the magnitude of the effects of the experimental intervention compared to the control intervention.
A confidence interval describes the uncertainty of this estimate, i.e. range of values within which we can be reasonably sure that the true effect actually lies.
What is a 95% confidence interval?
We are 95% certain that the true value of the mean lies within the confidence interval.
What does a wide vs a narrow confidence interval tell you ?
A narrow confidence interval implies a higher level of confidence in the results and vice versa.
-May expect a narrow confidence interval in experiments where bias or uncertainty is unlikely (e.g. measurments of heights) whereas more subjective studies (e.g. studying number of tumour cells) can have a wider confidence interval.
Distinguish between the conclusions we can draw from the following potential studies:
– The ACE inhibitor group had a 10% (95% CI +7% to +13%) higher chance of survival
– The patients who were treated with the drug had a 2 % (95% CI -3% to +7%) increase in T cells.
1) The ACE inhibitor group had a 10% (95% CI +7% to +13%) higher chance of survival
ACE inhibitor group had improved survival (because the entire confidence interval lies in positive percentages)
2) The patients who were treated with the drug had a 2 % (95% CI -3% to +7%) increase in T cells.
Cannot be certain of effect of drug on T cell because confidence interval includes negative values so could actually cause a decrease in T cells.
Which of confidence intervals or p values are better ? Why ?
Confidence intervals are better than p values, because:
1) Gives a range of possible effects sizes
2) Embrace the value of no difference between treatments (not significantly different from control)
3) Help interpret clinical trial data by placing upper and lower bounds on the true effect size
4) Statistically significant does not mean clinically important- the size of the effect determines importance! (e.g. treatment has higher mean survival than control but their confidence intervals overlap so cannot definitely say treatment group has higher survival)
Interpret graph on first slide of page 3 of lecture on “presenting the evidence”
Hospital A: If the 95% confidence interval crosses over the reference line of 1.0, the hospital’s infection rate is similar to ‘expected’ (predicted).
Hospital B: If the 95% confidence interval falls completely below the reference line of 1.0, the hospital’s infection rate is significantly lower than ‘expected’.
Hospital C: If the 95% confidence interval falls completely above the reference line of 1.0, the hospital’s infection rate is significantly higher than ‘expected’
Define Number need to treat (NNT). Is a bigger NNT good or bad ?
Number of patients you need to treat for one patient to benefit (treatment specific and describes difference between treatment and control in achieving a particular clinical outcome.
Bigger NNT means less effective intervention.
Calculate NNT for the following:
• 100 people given an analgesic tablet
– 70 had their pain relieved after 2 hours
• Same 100 given placebo
– 20 had their pain relieved
► Analgesic tablet responsible for pain removal of 50 of the patients (20 would have irrespective of analgesic), so ARR (Risk for treatment - Risk for control) is 50/100.
► NNT is reciprocal of ARR so 100/50, so 2.
2 people have to be treated for one to obtain effective pain relief
Identify a drug with a low NNT, and another one with a high NNT.
Mamography has a high NNT
Head lice treatment has a low NNT
Distinguish odds from probability.
Probability = number of favorable outcomes/number of all possible outcomes
Odds = number of favorable outcomes (e.g. that diseased were exposed)/number of unfavorable outcomes (e.g. that controls were exposed)
Define risk, relative risk, relative risk reduction, absolute risk reduction, NNT.
Risk = (number of outcomes in a group/number of people in group) x 100
Relative risk (= Risk ratio) = Risk of disease if exposed/Risk of disease if unexposed
Relative risk (= Risk ratio) = Risk of outcome if on drug/Risk of outcome if on placebo
Relative risk reduction = 1 - RR
Absolute risk reduction = risk exposed - risk unexposed
NNT = 1 / ARR
What situations should we use risk ratio in ? odds ratios ?
♪ Retrospective study (esp. case control studies), use odds ratio
♪ Prospective study (esp. cohort studies, clinical trials), use risk ratio
Calculate the risk ratio in the following, and then again doubling the number of patients in each category:
EXPOSED
Diseased: 24
No Disease: 60
Total: 84
UNEXPOSED
Diseased: 15
No Disease: 100
Total: 115
Then, do the same with odds ratio. What is the implication of this ?
RISK RATIO
• Risk in exposed = 24/84 = 0.29
Risk in unexposed = 15/115 = 0.3
Risk ratio = 0.29/0.13 = 2.19
This means that those exposed were 21.9% more likely to get disease than those who were unexposed.
• When doubling the number of patients in each category, the risk ratio becomes 1.93 (i.e. 19.3% more likely).
ODDS RATIO
• Odds of exposure in people with disease = 24/15 = 1.6
Odds of exposure in people without disease = 60/100 = 0.6
Odds of exposure = 1.6/0.6 = 2.7
• When doubling the number of patients in each category, the odds ratio remains the same
This is why, in retrospective studies, we should use odds ratios rather than risk ratios.
What is the meaning of a risk ratio inferior to 1 ?
Risk in exposed is lower than risk in unexposed
Describe the main features of a Forest plot.
-Each square represents an individual study
-Size of square is proportional to study weight
-Length of the line passing through each square is the 95% confidence interval
-Diamond at the bottom is the combined results for all the studies (i.e. mean + line of confidence interval)
-X-axis scale is logarithmic for ratios (odds, risk ratios) but linear for differences
-Line of effect, below which treatment is said not have an effect (i.e. good studies proving effectiveness of a drug should be to the right of this line)
