EBM therapy

1
Compliance
Original Study Design
Randomised
Surgical care
Medical care
2
Reality - unplanned cross overs
Surgical care
Medical care
Refuse
Require
surgery
surgery
3
Intention to treat analysis

• In this example of unplanned cross overs it may
  be that elderly or high risk patients are refused
  surgery whilst some of the fitter, younger
  patients, although randomised to medical
  treatment actually end up having the surgical
  intervention. This will cause a bias - there will
  be more elderly patients in the medical group and
  they will have a worse prognosis. By analysing
  the results using intention to treat this bias
  will be avoided. If there is still a treatment
  effect then this is likely to be a true effect.
  It is still worth while analysing by actual
  treatment groups - this should reveal an even
  better outcome with treatment. However if the
  intention to treat shows no benefit, and the
  analysis by treatment group shows a positive
  effect then the reviewer should question whether
  the result is due to bias and loss of
  randomisation.

4
Reality - poor compliance
Medical care
Placebo
Refuse
Refuse
treatment
treatment
5
Intention to treat analysis

• With placebo controlled trials it has been shown
  that compliant patients who take their placebo
  have a better outcome (up to 30 better) than the
  non-compliant patients. If there is a large drop
  out in both the active and placebo arms of the
  trial it is attractive to analyse only those who
  received the active treatment (discarding the
  non-compliant patients in the active arm) but
  include all the patients entered into the placebo
  arm to increase the precision of the results. If
  the active treatment is actually of no benefit,
  because the non-compliant patients (who have
  worse outcomes) are only included in the placebo
  arm then the active treatment may falsely
  appear to be of benefit.. Intention to treat
  analysis removes this bias.

6
4. Were patients, health workers and study
personnel blind to treatment?

• In a well designed randomised trial the person
  giving one of two (or more) possible treatments
  should not know which treatment the patient is
  receiving. In a double blind trial the patient
  should also not know which treatment they are
  receiving.

7
5. Were the groups similar at the start of the
trial?

• In the paper there should be a table showing the
  characteristics of the two treatment groups.
  Sometimes by chance, particularly in small
  studies, the groups may be unequal (e.g. more men
  in one group) and this can cause bias.
• The larger the study the more likely the groups
  are to be similar and the less likely the
  difference between the groups will be due to
  chance. If 20 characteristics are looked at then
  by chance at 0.05 level, we would expect a
  significant difference in one characteristic
  between the groups.

8
Study size

• The larger the study the more likely the groups
  are to be similar and the less likely the
  difference between the groups will be due to
  chance. Thus big studies (mega trials) are to be
  preferred. This will also help avoid Type 1 and
  Type 2 error.

9
Type 1 and Type 2 Error
10
Statistical Power

• Relative frequency with which a true difference
  of specified size between populations would be
  detected by the proposed study.

11
Statistical Power

• Relative frequency with which a true difference
  of specified size between populations would be
  detected by the proposed study.
• It is equal to 1 minus the probability of Type 2
  error.

12
Sample Size

• Difference in response rates to be determined
• An estimate of the response rate in one of the
  groups
• Level of statistical significance
• The value of the power desired
• Whether the test should be one-sided or two-sided

13
6. Aside from the experimental interventions,
were the groups treated equally?

• This can sometimes be a problem, particularly if
  one treatment group is followed up more
  intensively. The better outcomes may then be due
  to something that is occurring in the follow up
  consultations rather than be due to the original
  intervention.

14
II. What are the results?

• 1. What are the overall results of the study?

15
II. What are the results?

• 1. What are the overall results of the study?
• Look at the Relative Risk (RR) of the main
  outcome in the two groups.

16
II. What are the results?

• 1. What are the overall results of the study?
• Look at the Relative Risk (RR) of the main
  outcome in the two groups.
• What about sub-group analyses?

17
What about sub-group analyses?

• First look at the intention to treat analysis.
• You may also want to look at the results in the
  groups that actually received the treatment.
• Is the result the same in men and women? For
  different age groups? Smokers and non-smokers etc.

18
II. What are the results?

• 1. What are the overall results of the study?
• Look at the Relative Risk (RR) of the main
  outcome in the two groups.
• What about sub-group analyses?
• Can you calculate the Number Needed to Treat
  (NNT) from the results presented?

19
Number needed to treat

• NNT is 1/ARR
• ARR Absolute risk reduction

20
Absolute risk reduction

• Absolute risk reduction (ARR) is the absolute
  risk in the untreated group minus the absolute
  risk in the treated group
• (see example)

21
II. What are the results?

• 2. How precise are the results?

22
II. What are the results?

• How precise are the results?
• Give both p values and confidence intervals for
  each result.

23
Confidence intervals

• A 95 confidence interval (95 CI) is the range
  within which, were the study to be repeated the
  true result would occur 95 of the time. When
  looking at a relative risk, if the 95 CI
  contains 1 then the results are not significantly
  different. A confidence interval is equal to or
  - 1.96 times the standard error

24
p-values

• p-values indicate the likelihood that the Null
  hypothesis is true. i.e. that there is no
  difference between the results. A p-value less
  than 0.05 is by convention considered significant
  but it gives you no idea of the range of the
  likely true result.

25
III. Will the results help me in caring for my
patients?

• 1. Can the results be applied to my patient care?

26
Will the results help me in caring for my
patients?

• 1. Can the results be applied to my patient care?
  Clinical significance.
• Refer back to the clinical problem
• Are the studies generalisable to our patient?
• Age, ethnicity, community or hospital patients
  etc?

27
Will the results help me in caring for my
patients?

• 2. Were all the clinically relevant outcomes
  considered?

28
Will the results help me in caring for my
patients?

• 2. Were all the clinically relevant outcomes
  considered?
• What about other outcomes - particularly harm.
• What about quality of life issues?

29
Will the results help me in caring for my
patients?

• 3. Are the benefits worth the harms and costs?

30
Will the results help me in caring for my
patients?

• 3. Are the benefits worth the harms and costs?
• Cost differences in treatments.
• Greater benefits and less side effects?

31
Example

• If 2000 patients with mild hypertension are
  randomly allocated to treatment or placebo and 4
  patients in the placebo group have had a CVA at
  the end of the year and only 2 in the treated
  group have suffered a CVA what are

32
(No Transcript)
33
(No Transcript)