At St. George's we use two types of automatically marked question. We use multiple choice questions of the stem and 5 branch type, where each stem may be True or False, usually referred to as "MCQs". We use extended matching items questions, where there is a scenario and any number of questions with answers chosen from a long list, usually referred to as "EMIs". In our usual practice, MCQs have negative marking, i.e. you get +1 for a right answer and -1 for a wrong answer, and EMIs do not have negative marking.

In Term 2, 2002, a question of the EMI type is to be used.

*Each branch is True or False.*

1. The paired t test is:

(a) impractical for large samples.

(b) equivalent to a chi-squared test.

(c) suitable for very small samples.

(d) used for independent samples.

(e) requires the assumption that differences between paired observations
follow a Normal Distribution.

2. For a t test for two independent samples to be valid:

(a) the numbers of observations must be approximately the same in the two
groups.

(b) the standard deviations of observations must be approximately the same
in the two groups.

(c) the means must be approximately the same in the two groups.

(d) the observations must be from distributions which are approximately
Normal.

(e) the sizes of samples the must be small.

3. A simple linear regression equation:

(a) always describes a line which goes through the origin.

(b) always describes a line with zero slope.

(c) is not affected by changes in the units in which variables are
measured.

(d) describes a line which goes through the point defined by the means of
the two variables.

(e) is affected by the choice of dependent variable.

4. The product-moment correlation coefficient between two variables,
*r*:

(a) must lie between -1 and +1 inclusive.

(b) can only have a valid significance test carried out when one variable
follows a Normal Distribution.

(c) is expected to be zero when there is no relationship between the
variables.

(d) depends on which of the two variables is chosen to be the dependent
variable.

(e) measures the magnitude of the change in one variable associated with a
change in the other.

5. The standard chi-squared test for a 2x2 contingency table is not valid
unless:

(a) all the expected frequencies are greater than five.

(b) both variables are continuous.

(c) at least one variable is from a Normal Distribution.

(d) all the observed frequencies are greater than five.

(e) the sample size is at least 100.

6. To analyse the relationship between blood pressure and height in a sample
of students, we could use:

(a) paired t test

(b) correlation coefficient

(c) chi-squared test

(d) regression

(e) two sample t test

7. Advantages of random sampling for studying the general human population
include:

(a) it can be applied to any population.

(b) likely errors can be estimated.

(c) estimates obtained are not biased.

(d) it is easy to do.

(e) the sample can be referred to a known population.

8. In a case-control study to investigate oral contraception and breast
cancer:

(a) cases would be women with breast cancer.

(b) controls would be given oral contraception.

(c) cases would be given oral contraception.

(d) women with breast cancer would be randomly allocated to be cases or
controls.

(e) controls would be observed for several years to see how many developed
breast cancer.

9. In a case-control study, patients with lung cancer had a highly
significantly lower cholesterol level than did controls. This provides strong
evidence that:

(a) low cholesterol causes lung cancer.

(b) there is evidence for a relationship between low cholesterol and lung
cancer in the sampled population.

(c) low cholesterol is not related to lung cancer.

(d) low cholesterol and lung cancer always go together.

(e) low cholesterol is a risk factor for the development of lung cancer.

10. In a case-control study, 101 stroke patients were compared with 137
healthy controls. Among the results were:

Ever smoked? | Total | ||
---|---|---|---|

Yes | No | ||

Cases | 71 | 30 | 101 |

Controls | 36 | 101 | 137 |

Chi-squared = 45.5, 1 degree of freedom, P < 0.0001 Odds ratio = 6.6 (95% confidence interval = 3.8 to 11.8) |

(a) These data would be unlikely if smoking and stroke were unrelated.

(b) It is estimated that the relative risk of stroke for ever-smokers
compared to never-smokers is 6.6.

(c) In the population from which cases and controls come, the odds ratio is
estimated to lie between 3.8 and 11.8.

(d) We can conclude that smoking causes stroke.

(e) There is good evidence that smokers have an increased risk of
stroke.

**
For each question, choose an appropriate answer from the following list.
Choose only one answer unless the question asks for more.**

a. Binomial

b. Case-control study

c. Chi-squared test

d. Cohort study

e. Confidence interval

f. Controlled trial

g. Cross-sectional study

h. Double blind

i. Highly significant

j. Large sample comparison of proportions

k. Little or no evidence

l. Logarithmic transformation

m. Negatively skew

n. Normal

o. Not significant

p. Odds ratio

q. Paired t

r. Positively skew

s. Regression

t. Significant

u. Some evidence

v. Standard error of mean

w. Strong or very strong evidence

x. Symmetrical

y. Two sample t method

z. Weak evidence

Magnetic resonance (MR) spectroscopy was used to estimate n-acetylaspartate (NAA) in 43 patients with Motor Neuron Disease and 14 healthy volunteers recruited by advertisement. Results were presented as the ratio of NNA to creatine (Cr). The patients were grouped according to the type of Motor Neuron Disease: amyotrophic lateral sclerosis (ALS), with or without probable upper motor neuron signs, primary lateral sclerosis (PLS), and non-upper motor neuron disorders. The following table was given:

Group and Motor Cortex | No. of Patients | Mean | SD | SEM | P Value* |
---|---|---|---|---|---|

Control | |||||

Right | 14 | 3.12 | .34 | .09 | NA |

Left | 14 | 3.04 | .40 | .11 | NA |

Combined | 14 | 3.08 | .32 | .09 | NA |

ALS | |||||

Right | 11 | 2.42 | .50 | .15 | <.05 |

Left | 11 | 2.37 | .46 | .14 | <.05 |

Combined | 11 | 2.40 | .42 | .13 | <.05 |

ALS with probable upper motor neuron signs | |||||

Right | 8 | 2.54 | .46 | .16 | NS |

Left | 8 | 2.38 | .45 | .16 | <.05 |

Combined | 8 | 2.44 | .41 | .15 | <.05 |

PLS | |||||

Right | 18 | 2.46 | .53 | .13 | <.05 |

Left | 18 | 2.68 | .50 | .12 | NS |

Combined | 18 | 2.57 | .38 | .09 | <.05 |

Non-upper motor neuron disorders | |||||

Right | 6 | 2.83 | .67 | .27 | NS |

Left | 6 | 3.03 | .57 | .23 | NS |

Combined | 6 | 2.93 | .58 | .24 | NS |

* For comparison with the control group. NA = not applicable, NS = not significant |

(*Radiology *1999; 212:763-769)

11. What should be understood by the abbreviation `SEM’?

12. Which method could be used to calculate the P values?

13. What is the strength of the evidence for a difference in combined NAA/Cr
between ALS and control subjects?

14. What is the strength of the evidence for a difference in combined NAA/Cr
between patients with non-upper motor neuron disorders and control
subjects?

To investigate the use of the toe-touch test as a predictor of hamstring injury in Australian Rules footballers, 67 players were videotaped while performing a test from the erect standing position. The Peak Motion Measurement System was used to obtain measurements of end range hip flexion, lumbar flexion, toe-touch distance (TTD) and the ratio of lumbar spine flexion to hip flexion. Over the following football season, eight subjects (11.9 per cent) sustained a hamstring strain. The following date were given:

Measurement | Hamstring strain n = 8 | No hamstring strain n = 59 | |
---|---|---|---|

Toe-touch distance (cm)* | 1.3 (9.1) | 2.6 (9.2) | P=0.76 |

Lumbar spine flexion (degrees) | 43.6 (9.6) | 43.5 (7.9) | P=0.92 |

Hip flexion (degrees) | 79.7 (10.3) | 77.8 (10.9) | P=0.82 |

Lumbo-femoral ratio | 0.56 (0.16) | 0.57 (0.14) | P=0.90 |

* can be negative if the subject cannot touch toes |

(*Australian Journal of Physiotherapy *1999; **45**: 103-109)

*QUESTIONS ABOUT SCENARIO TWO*

15. How could the P values be best described?

16. What is the strength of the evidence that the test is a good predictor
of hamstring injury in Australian Rules footballers?

In a study of the treatment of scorpion stings with antivenom, 825 consecutive patients older than 10 years, who presented to the accident and emergency department of the hospital in Tozeur, Tunisia, were randomly assigned placebo (n=413) or 20 ml bivalent intravenous scorpion antivenom (n=412). The solutions were prepared by the hospital pharmacist and were identical in appearance so that physicians in charge and patients were masked to treatment status. All patients were observed for 4 h. Patients who developed life-threatening symptoms were admitted to the intensive-care unit. The cure rates were 55% in the scorpion antivenom group and 66% placebo (absolute difference, 11% [95% CI 4.8 to 26.8]; p=0.234). Preventive effects were seen in 94% and 96% of patients in the scorpion antivenom and placebo groups, respectively, who remained symptom-free (absolute difference, 2% [95% CI 1.27 to 5.27]; p=0.377).

(*Lancet *1999; **354**: 906-09)

17. Which technique was used here to prevent both assessment and response
bias?

18. What should be understood by the abbreviation 'CI'?

19. Which method could be used to calculate '[95% CI 1.27 to 5.27]'?

20. Which method, different from that in question 19, could be used to
calculate 'p=0.377'?

21. What is the strength of the evidence that antivenom is more effective
than placebo in curing and preventing symptoms from scorpion stings?

22. What study design was used in Scenario One?

23. What study design was used in Scenario Two?

24. What study design was used in Scenario Three?

The relationship between type 2 (adult onset) diabetes and smoking was studied using the British National Child Development Study (NCDS). This is based on a survey of about 17,000 births from 3 to 9 March 1958. These children were studied at several ages up to 33 years. Cohort members signed consent forms at age 33 years allowing access to medical records. A personal interview at age 33 years asked about diabetes. Those with only gestational diabetes were also excluded: 15 men and 13 women with an onset of diabetes between 16 and 33 years were identified.

At birth midwives recorded information on the child's sex, birth weight, mother's age, her age on leaving full time education, family social class, and smoking during pregnancy (after the 4th month) categorised as non-smokers, medium (1-9 cigarettes/day) heavy (>10), and variable (a balance of medium and heavy). Details of maternal smoking were again recorded in 1974 when children were 16. The children's own smoking behaviour was recorded during an interview at age 16.

Of the 521 children born to heavy smokers during pregnancy, 9 (1.7%) developed diabetes, compared to 12 (0.3%) of the 3430 children of non-smokers during pregnancy. The odds ratio was 4.94 (95% CI 2.07 to 11.77, P<0.001). After adjustment for sex, mother's age at birth of cohort member, age mother left school, family social class at birth, birth weight, own smoking at age 16 years, maternal smoking in 1974, and BMI at age 33 years, the odds ratio was 4.02 (95% CI 1.14 to 14.14, P= 0.030).

(*BMJ* 2002; **324**: 26-27)

25. What kind of study design is this?

26. Which statistical method could be used to test the null hypothesis that
the odds ratio in the population is 1.0?

27. How strong is the evidence that maternal smoking and type 2 diabetes are
associated?

28. How strong is the evidence that maternal smoking during pregnancy and
type 2 diabetes are associated, after the effect of maternal smoking outside
pregnancy (i.e. in 1974) has been removed?

To examine the relation between self reported eating frequency and serum lipid concentrations in a free living population, data were analysed from 14 666 men and women aged 45-75 years and living in Norfolk. A health check was carried out and subjects completed a dietary questionnaire. (This was part of a large study into diet and cancer.) Mean concentrations of total cholesterol and low density lipoprotein cholesterol decreased with increasing daily frequency of eating. No consistent relation with eating frequency was observed for high density lipoprotein cholesterol, body mass index, waist to hip ratio, or blood pressure.

The mean cholesterol measurements (mmol/l) were as follows:

Eating frequency (No of times a day) | P value for linear trend | |||||
---|---|---|---|---|---|---|

1 or 2 | 3 | 4 | 5 | 6 or more | ||

Men | (n=353) | (n=2176) | (n=2525) | (n=1211) | (n=625) | |

Cholesterol | 6.11 | 6.06 | 5.98 | 5.94 | 5.94 | <0.01 |

LDL cholesterol | 4.06 | 3.98 | 3.91 | 3.86 | 3.86 | <0.001 |

HDL cholesterol | 1.24 | 1.24 | 1.23 | 1.22 | 1.22 | 0.42 |

Women | (n=362) | (n=2182) | (n=2877) | (n=1511) | (n=844) | |

Cholesterol | 6.26 | 6.25 | 6.19 | 6.12 | 6.08 | <0.001 |

LDL cholesterol | 4.00 | 3.97 | 3.95 | 3.90 | 3.85 | 0.02 |

HDL cholesterol | 1.57 | 1.60 | 1.56 | 1.55 | 1.56 | <0.01 |

(*BMJ* 2001; **323**: 1286-1288)

29. What kind of study is this?

30. Which statistical method could be used to calculate the P values?

Back to Research and Critical Skills Course Page.

This page maintained by Martin Bland.

Last updated 29 June, 2004.