Examination Paper for STAT0014 (Level 6 and 7)
Page 3 of 9
Section A
A1
A parallel two-group trial was to be carried out in a rural area of East Africa to ascertain whether giving
food supplementation during pregnancy increases birth weight. Women attending the antenatal clinic
were to be randomly assigned to either receive or not receive supplementation. The researchers
performed a sample size calculation and wrote the following statement.
“A sample size of 95 women per randomisation arm is sufficient to detect a clinically significant
difference of 0.25 kg using a z-test with 90% power at the 5% significance level. The variability in the
birth weights in the two groups of women was assumed to be the same and the sample size above
allows for a 10% dropout rate.
Read carefully the statement above and state which parameter value is missing in order to replicate
the sample size calculation above. Calculate this missing value. [3]
A2
Logistic regression was used to model the relationship between short-term mortality following CABG
surgery (1=died, 0=alive) and the predictors age as a categorical variable (0=less than 65 years, 1= 65
years or more) and sex (0=male, 1= female). The estimated intercept and coefficients for age and sex
were 0 = 3.10, age = 1.5 and sex = 0.5, respectively.
Data on two patients are given in the following table.
Patient ID Outcome Age (years) Sex
1 alive 65 Male
2 died 75 Female
a) Calculate the probability of short-term mortality for these two patients. [2]
b) Calculate the value of the likelihood for this model based on just these two patients. [1]
Examination Paper for STAT0014 (Level 6 and 7)
Page 4 of 9
A3
In a trial comparing Acupuncture (A) with Homeopathy (H), patients suffering from chronic headache
were allocated to treatment using a minimisation procedure based on sex (male/female) and type of
headache (migraine/tension/cluster). The number of patients with each characteristic for each
treatment group after eighty-nine patients have entered the trial are given in the table below.
Minimisation factors Acupuncture Homeopathy
Sex Male 19 16
Female 26 28
Type of headache
Migraine 23 24
Tension 17 17
Cluster 5 3
a) [Type] The 90th recruited patient is male and suffers from migraines. Given the information in the
table above, which group should randomisation be weighted towards Give a reason for your
answer. [2]
b) [Type] If stratified randomisation was used instead of minimisation, how many strata would be
needed to balance on these same factors. Justify your answer. [1]
c) In this trial, each patient’s quality of life will be measured at baseline and again 1 month after
treatment. Fully specify the most suitable and efficient model for analysis of this trial. Define all
terms in the model. [2]
A4
A randomised trial compared a new cream to a placebo cream for the treatment of athlete’s foot.
Using randomisation, 39 patients were allocated to the new cream and 40 to the placebo. Patients
were assessed at the end of a two-week treatment period. The infection was eradicated for 29 patients
in the new cream group and 23 patients in the placebo group. A difference of 1
0% in the proportion
of patients with an eradicated infection was considered to be clinically important.
a) Calculate an absolute measure that could be used to estimate the effect of the new cream
amongst people with athlete’s foot and quantify the precision of this estimate. Show your
workings. [3]
b) [Type] Interpret the results from part a), state a conclusion about the clinical benefit of the
new cream and comment on the size of the study. [3]
Examination Paper for STAT0014 (Level 6 and 7)
Page 5 of 9
A5
A cohort study was carried out to investigate the long-term outcome ‘loss of vision’ for patients,
following treatment for radiologically confirmed retinal artery occlusion. The table shows the results
for the two most common treatments.
Loss of vision Person-months
of follow-up
Rate of loss of
vision (per
year)
Yes No
Intravenous
thrombolysis (IT)
71 50 A 7.07
Antithrombotic
treatment (AT)
117 66 183.0 B
Relative rate of loss of vision (IT
compared with AT) C
For all calculations show your workings. Given the information above:
a) Calculate the values of A, B and C in the table. [2]
b) Calculate the relative risk of loss of vision. How do the relative risk and relative rate estimates
compare and what does this tell you about the study [2]
c) Calculate a 95% confidence interval for the relative rate given in the table. Then, draw a conclusion
regarding the relative effect of IT compared to AT. [3]
d) [Type] Explain why a modelling approach might be preferred for analysis of data produced by this
study design. [1]
Examination Paper for STAT0014 (Level 6 and 7)
Page 6 of 9
Section B
B1
A clinical trial of treatment for chronic back pain allocated 184 subjects to either 10 weekly sessions
of physiotherapy or a self-help booklet. Treatment was allocated using simple randomisation. The
outcome was a pain score measured at 6-month follow-up and also at baseline. A score of 24
represents the worst imaginable pain and 0 represents no pain. The following table summarises
baseline and follow up pain scores.
Pain score Baseline 6 Month follow up*
N Mean SD N Mean SD
Leaflet 96 5.646 3.931 88 3.739 4.484
Physiotherapy 88 6.591 3.996 81 2.988 3.441
* where available
a) [Type] Comment on the imbalance in numbers randomised to the groups in this trial and how this
might have been prevented. [2]
b) [Type] Based on the information in the table above, identify two potential sources of bias and
explain how such problems can impact the trial. [3]
The data from the trial were analysed in 3 ways. Results are given in the table below:
N Treatment effect estimate Standard Error
Unadjusted 169 -0.751 0.619
Change (6 month – baseline) 166 -2.114 0.667
ANCOVA, adjusted for baseline 166 -1.351 0.587
Note: correlation between change scores and baseline scores = -0.53
c) [Type] Explain in detail why for this trial:
i) the unadjusted treatment estimate is smaller than the estimates from the other analyses. [1]
ii) the change analysis produces a larger treatment estimate than the ANCOVA analysis. [2]
d) [Type] From the table, identify the analysis you consider to be most appropriate for this trial and
interpret its results. [3]