statistical issues

Question 1

How can you blind studies?

Answer 1

1. Centralised preparation of similar treatments. This ensures that the placebo and experimental tablets appear the same.
2. Tablets are embedded in hard gelatine capsules. This ensures they have identical appearance.
3. Topical application. In an RCT comparing fluorouracil cream and placebo, the 2 creams were indistinguishable in texture, colour, smell and packaging was identical. You always wouldn’t need to worry about taste.
4. Use of flavour to mask taste. Peppermint or Wild cherry flavoured syrup can be used.
5. Centralised preparation of distinguishable treatments in opaque syringes/bottles. You can’t see inside to tell the difference. Opaque covers may be used to conceal infusion bags and catheters.
6. Injection of treatments administered by an unblinded nurse not involved in the study. In this case the nurse must not provide any information to anyone about the contents of the syringe.
7. Use of double dummy. In a trial assessing the abciximab infusion for 24 hours or 48 hours vs placebo. Abciximab was given for 24 hours then placebo for 24 hours. Or the placebo was given for 48 hours, or the abciximab was given for 48 hours. In this case the patient would never know what treatment they were receiving as all are being treat the same way.

Question 2

What does randomisation aim to reduce?

Answer 2

Selection bias. It does this by randomly allocating a patient to one of the treatments. It tends to produce comparable treatment groups for both known and unanticipated prognostic factors. Also allows valid statistical inference.

Question 3

What are the advantages and disadvantages of block randomisation?

Answer 3

This method ensures that there is a balance between the two treatment groups in terms of numbers. However, it may be relatively simple to unblind if the blocking factor is known as the sequence become predictable after the first two patients.

Question 4

What is stratified randomisation?

Answer 4

This is when there is a balance of certain characteristics, e.g. risk factors between each arm. The risk factors must be measured beforehand.

Question 5

What is minimisation?

Answer 5

This is the process of adjusting the new patients to a treatment group depending on the imbalances already occurring within the groups, e.g. if age > 60 is a risk factor, then it may allocate a patient to a group if that group is imbalanced for >60 year olds. This ensures that differences between the treatment groups are minimised.

Question 6

What is central based randomisation?

Answer 6

Subjects are assigned to a treatment according to the randomisation schedule, regardless of what centre they’re in.

Question 7

What is centre-based randomisation?

Answer 7

This is when there is a different randomisation schedule for each centre. This way, the number of subjects in each treatment group is around the same for each centre.

Question 8

What is allocation concealment?

Answer 8

This refers to when the randomisation schedules are concealed from the study investigators to avoid selection bias. This must occur as clinicians may consciously or subconsciously select patients which are most likely to benefit, to the intervention arm. It has been shown that trials without allocation concealment have a 41% increase in the effect for the active treatment, compared to trials with adequate allocation concealment.

Question 9

What are effective methods to achieve allocation concealment?

Answer 9

1. Sealed envelopes (must be opaque)
2. Central telephone randomisation
3. Internet-based Randomisation.

Question 10

What is a superiority trial?

Answer 10

This is defined as a trial with a primary objective of showing that the response to the investigational treatment is superior to a comparator.

Question 11

When is it ethical to demonstrate superiority to placebo?

Answer 11

This is only done if it is safe to do so, e.g. in cases of depression, not in cancer.

Question 12

What is a non-inferiority trial?

Answer 12

These are trials that are used to show that an investigational produce is good or better than a comparator. The aim is to show that any reduction in the response is not greater than a clinically unimportant amount. They may be aiming to show secondary benefits for the new product.

Question 13

What is the non-inferior margin?

Answer 13

This is the margin which concerns the maximum allowable difference between the responses, before it is considered clinically important. This must be pre-specified and can be determined via guidelines or other articles. Sometimes you can use a proportion of the superiority margin seen for the active control.

Question 14

What are the issues with non-inferiority margins?

Answer 14

Although the drug may not be considered non-inferior in terms of efficacy/response. There may be inferiority seen in other end points, such as safety. There is potential that this could outweigh the primary end point of efficacy.

Question 15

What are equivalence trials?

Answer 15

These are trials used to show that an investigational product is comparable in efficacy to a comparator. Want to show that any difference in response for the new treatment is not greater than a clinically unimportant amount.

Question 16

Why are equivalence trials used?

Answer 16

1. You may want to show that a treatment is comparable in subgroups, e.g. in patients with or without diabetes or in patients in different age groups, etc.
2. Bioequivalence studies to show that 2 formulations of a drug have similar bioavailability and therefore equivalent therapeutic effect.

Question 17

What are the rules a comparator drug must follow in bioequivalence trials?

Answer 17

The comparator drug must contain the same active ingredient, must follow the same route of administration, the same dosage form, the same strength. It therefore has the same therapeutic effect and targets the same pathway.

Question 18

How can you conclude that bioequivalence exists?

Answer 18

If the bioavailability of the test formulation is within 20% of the reference formulation, we can assume bioequivalence.

Question 19

What is the trial design for bioequivalence studies?

Answer 19

Usually cross-over design, specifically Williams design. The length of the washout period must be at least 5.5-8.5 half lives.

Question 20

What is ordinal data?

Answer 20

Data that has more than 2 categories which are ranked, e.g. pain grade (none, mild, moderate, severe).

Question 21

What is nominal data?

Answer 21

This is data that has more than 2 categories which are unordered, e.g. blood groups.

Question 22

What is survival data?

Answer 22

This is data which combines a continuous variable such as time, with a binary variable (e.g. event or no event). Examples include: time to death (advanced cancer therapies), time to disease remission.

Question 23

What is the best measure of central tendency for skewed data?

Answer 23

Median and interquartile range.

Question 24

What is type I error?

Answer 24

This is when we reject the null hypothesis when it is in fact true.

Question 25

What is a type II error?

Answer 25

This is when we accept the null hypothesis, when it is in fact false.

Question 26

What are the consequences of a high type 1 error rate?

Answer 26

There is a risk that we will conclude that a treatment is beneficial when this is not the case. This can lead to further investments in an ineffective compound and ultimately putting an ineffective drug on the market. Often this is regulatory or public health risk.

Question 27

What are the consequences of a high type 2 error rate?

Answer 27

The sponsor invests money in a study which does not have a high chance of delivering the information required. Often this is the sponsors risk.

Question 28

What determines study power?

Answer 28

1. Effect size. Larger difference in effect between two groups means there is more power.
2. Sample size. Larger sample size, easier to demonstrate any difference between two groups.
3. Variance. Lower variability in effect between subjects, easier it is to demonstrate a difference.
4. Significance level. Higher alpha, greater chance of detecting an effect. However, significance level must be low enough to ensure positive results are convincing, but not too low that its too hard to detect a difference.

Question 29

What are the determinants of sample size?

Answer 29

1. Power
2. Significance level
3. Clinical event rate - frequency of outcome being measured in the study population.
4. Clinically relevant effect size
5. Variance
6. Compliance/Drop-out rates
7. Subject allocation ratio (usually 1:1)

Question 30

Name measures of precision.

Answer 30

Standard error measures precision, e.g. larger spread means larger standard error and less precision. Another measure of this is coefficient of variation which is the ratio of the standard deviation to the mean. Higher coefficient, less precise.