1. Which of the following is not correct about the principles of causal inference
  1. Lack of Temporal Ambiguity means that the cause should precede the effect
  2. A Dose-Response Relationship means that the size of an effect should vary with the size of the cause
  3. Coherence means that there is a biological or theoretical basis for the hypothesized cause and effect
  4. Specificity means that the effects are associated primarily with one cause
  5. Consistency means that the same relationship is seen over many studies involving a wide range of other factors
  1. An apparent association between two variables, other than representing a true causal relationship, could be due to
  1. Chance
  2. Confounding
  3. Bias
  4. Random error
  5. Systematic error

For the statements in question 3 to 9 match the scenarios to the type of bias which they either demonstrate or would be liable to demonstrate from the following list. NB each response can be used only once.

  1. Ecological fallacy
  2. Floating numerator
  3. Hawthorne effect
  4. Healthy worker effect
  5. Halo effect
  6. Interviewer bias
  7. Recall bias
  1. Reduction in drug use by patients suffering from schizophrenia in the course of an observational study of illicit drug use in psychiatric patients
  1. Decreased self reporting of psychotic symptoms in self report scale in patients whose first language is not English
  1. Increasing number of assaults by patients with schizophrenia as compared to other diagnoses on an acute psychiatric ward over succeeding months period after the initiation of a zero tolerance policy towards illicit drug use in in-patient wards.
  1. Increased reporting of birth complication by mothers of patients suffering from schizophrenia
  1. Attributing a causal role for drug misuse in schizophrenia in inner cities on the basis that rates of drug use and schizophrenia are higher in the inner city
  1. Reduced rates of psychosis in drug using doctors as compared to age matched drug users from the general population with similar drug histories
  1. Tendency of raters to rate all psychotic symptoms as more severe in patients with poor hygiene.
  1. Annual unadjusted incidence of 1st onset Schizophrenia as ascertained via a culturally validated version of SCAN is 800/100,000 in Egypt as opposed to 500/100,000 in Scotland. The disparity would be most plausibly attributed to;
  1. Ecological fallacy
  2. Ascertainment of mild cases
  3. Reduced access to services
  4. Better prognosis of psychosis outwith soi disant developed countries
  5. The influence of the pyramids

For questions 11 to 18 choose the study design from the list below that most closely approximates to the scenario.

  1. Case study
  2. Case series
  3. Case control
  4. Cohort
  5. Cross section
  6. Nested case control
  7. RCT
  8. Retrospective cohort
  1. 5 year hell of 9/11 survivor
  1. 5 year outcome of EMDR for PTSD diagnosed and treated at a specialist clinic at one year after 9/11
  1. The MAV 9/11 Study; Morbidity and Associated Variables in 9/11 survivors in the immediate aftermath of the event using a random sample drawn from Health Maintenace Organisation and Health insurance registries.
  1. Characteristics of PTSD cases in 9/11 survivors from MAV 9/11 at one year as compared to case nominated controls from the same workplace.
  1. The Goldman Sachs project; 5 year health status and associated variables in a random sample of individuals entering employment with Goldman Sachs worldwide in 1999
  1. Pre-morbid predictors of development of PTSD following 9/11 in individuals from the 1999 Goldman Sachs intake.
  1. Factors associated with development of PTSD in the 1999 Goldman Sachs intake cohort. Cases vs. age matched controls
  1. Prospective 5 year follow up of subjects presenting with PTSD up to 1 year after 9/11 among those who declined vs those who accepted counselling within the first 6 weeks. Data derived from the American Association of counsellors (who had been tasked with offering all individuals involved in 9/11 immediate counselling) cross referenced with MAV 9/11.
  1. Individuals with the CATCH 22 mutation have a 40% risk of psychosis. The RR of developing psychotic symptoms with prolonged heavy cannabis use was shown to be as much as 6 in the Swedish conscript study. Consequently an individual with the CATCH 22 mutation who indulged in prolonged heavy cannabis use
  1. Would inevitably develop psychotic symptoms
  2. Would have a risk of developing psychotic symptoms of 2.4
  3. Could be reassured that cannabis is less bad for you if you´ve got CATCH 22
  4. The risks are so high that they can´t be independent
  5. None of the above.
  1. Individuals with the CATCH 22 mutation have a 40% risk of psychosis. The OR of developing psychotic symptoms with prolonged heavy cannabis use was shown to be as much as 6 in the Swedish conscript study. Consequently an individual with the CATCH 22 mutation who indulged in prolonged heavy cannabis use could have a risk of developing psychotic symptoms as high as
  1. 240%
  2. 100%
  3. 95%
  4. 80%
  5. 24%

Questions 21 to 27 refer to the data in the table below, drawn from the cross sectional study; ‘Alcohol, cannabis and tobacco use among a representative community sample of 10641 Australians: a comparison of their associations with other drug use and use disorders, affective and anxiety disorders, and psychosis’ by Degenhardt et al, Addiction (2001) 96, 1603 – 1614.

Bivariate and adjusted odds ratio (OR) and 95% confidence intervals (95%CI) of DSM-IV affective disorder according to alcohol, tobacco and cannabis use

Prevalence
(95% CI) / Bivariate
OR / 95% CI / Adjusted
OR* / Adjusted
95% CI*
No Alcohol Use / 7.3 (0.4) / 1.00 / 1.00
Alcohol Use / 5.5 (0.3) / 0.82 / 0.70 – 0.97 / 0.98 / 0.81 – 1.19
Alcohol Abuse / 6.1 (2.0) / 0.90 / 0.50 – 1.60 / 0.64 / 0.34 – 1.21
Alcohol dependence / 24.0 (4.0) / 4.47 / 3.48 – 5.74 / 1.98 / 1.45 – 2.72
No cannabis use / 6.2 (0.3) / 1.00 / --- / 1.00 / ---
Cannabis use / 12.1 (2.7) / 2.24 / 1.73 – 2.91 / 1.30 / 0.94 – 1.79
Cannabis abuse / 18.6 (5.3) / 2.88 / 1.61 – 5.17 / 1.46 / 0.73 – 2.90
Cannabis dependence / 13.6 (2.6) / 2.85 / 1.86 – 4.35 / 0.91 / 0.54 – 1.54
No tobacco use / 5.4 (0.4) / 1.00 / --- / 1.00 / ---
Tobacco use / 10.9 (0.8) / 2.20 / 1.90 – 2.54 / 1.48 / 1.24 – 1.76

* Adjusted for age, gender, educational attainment, marital status, employment status and neuroticism

21. Regarding the figures for prevalence attaching to the various levels of alcohol use in the second column from the left.

  1. They would add up to 100 if the study had more subjects
  2. Alcohol use is more common than no alcohol use
  3. Alcohol abuse is more common than alcohol use
  4. If Australians get depressed they tend to drink at a dependent level
  5. Depression is apparently more common in alcohol dependent subjects but the figures do not allow any comment on significance

22. The unadjusted OR for depression in subjects with alcohol use vs no use is

  1. 1.
  2. 0.90
  3. 0.74
  4. 0.60
  5. Not calculable from these figures

23. Numerically the RR would be

  1. Much lower
  2. Slightly lower
  3. The same
  4. A little higher
  5. A lot higher

24. The study reports OR rather than RR for all of the following reasons EXCEPT

  1. Because logistic regression is being used
  2. Because use of RR entails the possibility of derived probabilities of greater than 1
  3. Because the rare disease assumption is satisfied
  4. Because OR is a reasonable approximation to RR at low prevalence
  5. Because OR is more powerful.

25. For alcohol use vs no use the OR for depression is 0.74 prior to bivariate adjustment and 0.82 after. The implication is that

  1. Alcohol and cannabis together protect against depression more than alcohol alone
  2. Smoking cigarettes and/or cannabis have a protective effect against depression but the effect is confounded by alcohol use
  3. Alcohol is an effective antidepressant
  4. Alcohol use is positively and independently associated with depression
  5. The negative association of alcohol use and depression is partially confounded by smoking status for cannabis and/or cigarettes

26. Assuming that there is a dose response relationship between cannabis use at various levels and risk of depression the failure of the study to demonstrate such a relationship may be accounted for by all of the following reasons apart from

  1. Type 2 error
  2. Confounding by gender, given that heavy cannabis use is an exclusivlely male preserve and depression is less common in males
  3. Problems in definition of cannabis dependence vs abuse
  4. Problems in definition of depression in the context of cannabis intoxication
  5. Association of cannabis abuse as opposed to dependence with higher levels of use of other depressogenic drugs (eg opiates) not accounted for in the analysis

27. Assuming that the OR of 1.48 for depression in Australian Tobacco users vs non users operated at the population level in Scotland, that OR was a reasonable approximation to RR, that the relationship was causal and unconfounded, and that the prevalences of depression and smoking were 5% and 20% respectively, the population attributable risk for smoking and affective disorder would be;

  1. 0.4%
  2. 1%
  3. 5%
  4. 8.8%
  5. 20%

28. Which of the following correctly ranks the various kinds of epidemiological study in ascending order of reliability in their ability to gauge an association between an exposure and a disease?

  1. Case study, case series, cohort, RCT, case control
  2. Case study, case series, case control, cohort, RCT
  3. Case control, case series, RCT, case study, cohort
  4. Case study, case control, case series, cohort, RCT
  5. Case control, case study, cohort, RCT, case series,

29. Case control studies would be most inefficient in the study of

  1. Rare disease, common exposure, high attributable risk
  2. Rare disease, rare exposure, high attributable risk
  3. Common disease, rare exposure, low attributable risk
  4. Common disease, common exposure, low attributable risk
  5. Common disease, rare exposure, high attributable risk

30. Cohort studies would be most inefficient in the study of

  1. Rare disease, rare exposure, low attributable risk
  2. Rare disease, rare exposure, high attributable risk
  3. Common disease, rare exposure, low attributable risk
  4. Common disease, common exposure, high attributable risk
  5. Common disease, common exposure, low attributable risk

31. Case control studies yield an OR but this can be used as an estimate of the RR…

  1. Because OR and RR are arithmetically derived in the same way
  2. If the rare disease assumption is met
  3. As long as you remember that RR tends to give a more numerically extreme result
  4. The result is statistically significant
  5. If the 95% CI for OR and RR overlap

32. A case control study was conducted to study the relationship between cannabis use prior to the age of twenty and subsequent psychotic illness. Cases were drawn from consecutive admissions to the RoyalEdinburghHospital who expressed an interest in the research. Controls were from a random sample from a local church congregation in Morningside. On average cases tended to be younger, of lower social class, and male

psychosis
+ve / psychosis
-ve
cannabis
+ve / 9 / 1
cannabis
-ve / 91 / 99

The OR in favour of an association between cannabis and psychosis is

  1. 100/1
  2. 0.11
  3. 0.98
  4. 9.8
  5. >0.95

33. However the result is likely to have been influenced by all of the following apart from

  1. Self selection bias
  2. Recall bias
  3. Failure of the rare disease assumption
  4. Inadequate matching
  5. Embarrassment

34. The same study was repeated using cases were drawn from a random sample of patients in current follow up at the RoyalEdinburghHospital. Controls were age, sex and post code matched subjects from a national GP registration database

psychosis
+ve / psychosis
-ve
cannabis
+ve / 7 / 3
cannabis
-ve / 93 / 97

The OR is now

  1. 2.4
  2. (21/93)/97
  3. 1/2.4
  4. Significant
  5. Insignificant

35. The second study is less likely to have yielded a confounded result because of improved

  1. Restriction
  2. Matching
  3. Stratification
  4. Regression
  5. All of the above

36. A further case control study to assess the relationship between cannabis use before 20 and psychosis was undertaken, this time in Jamaica. Cases were drawn from consecutive admissions to the RoyalKingstonHospital

Cases and controls were matched for age, sex, social class and religion (37 % of the sample were Rastafarians)

psychosis
+ve / psychosis
-ve
cannabis
+ve / 58 / 45
cannabis
-ve / 42 / 55

The OR on this occasion is

  1. 0.7
  2. 1.7
  3. 2.7
  4. 3.7
  5. 4.7.

37. The OR is different from that obtained in the second Edinburgh study, which may reflect…

  1. An underestimate because of undermatching
  2. An overestimate because undermatching
  3. An underestimate because of overmatching
  4. An overestimate because of overmatching
  5. None of the above

38. A case control study was conducted to assess the relationship between an enthusiasm for Pink Floyd and a diagnosis of schizophrenia. Cases were drawn from a random sample of out-patients Controls were friends of the subjects of around the same age.

60 % of cases were male as against 40% of controls

psychosis
+ve / psychosis
-ve
Pink Floyd
+ve / 35 / 28
Pink Floyd
-ve / 65 / 76

The OR for the association of Pink Floyd and psychosis is

  1. 1.06
  2. 1.26
  3. 1.46
  4. 1.66
  5. 1.86

The data were re-analysed separately for males and females as follows

psychosis
+ve / psychosis
-ve / psychosis
+ve / psychosis
-ve
Pink Floyd
+ve / 30 / 20 / Pink Floyd
+ve / 5 / 8
Pink Floyd
-ve / 30 / 20 / Pink Floyd
-ve / 35 / 56

39. The re-ananlysis is an example of

  1. Restriction
  2. Randomisation
  3. Matching
  4. Stratification
  5. Regression

40. The re-analysis reveals that

  1. The positive finding of the first analysis was a type 2 error
  2. The positive relationship was confounded by gender
  3. The relationship between Pink Floyd and schizophrenia was obscured by gender
  4. The relationship is likely to be confounded because people who like Pink Floyd smoke a lot of cannabis
  5. The relationship between gender and schizophrenia is mediated by musical taste

41. A study was undertaken to assess the presence or absence of a preceding life event in patients presenting with depression in general practice

200 cases of new onset depression were recruited from GPland. Controls were patients from the same practice matched for age and sex and post code, life events in both groups being assessed with the Life Events and Difficulties Scale

Depression
case / Control
Life event
+ve / 110 / 80
Life event
-ve / 90 / 120

OR = 1.83

Cases and controls were then stratified on the presence of high or low levels of social support

High Social support (n = 225) Low social support (n = 175)

Depression
case / Control / Depression
case / Control
Life event
+ve / 40 / 65 / Life event
+ve / 70 / 15
Life event
-ve / 50 / 70 / Life event
-ve / 40 / 50

OR = 0.86 OR = 5.83

The stratified analysis reveals all of the following except

  1. There is an interaction between level of social support and the effect of life events on the development of depression
  2. Life events are irrelevant to subsequent development of depression in the presence of adequate social support
  3. The effect of life events on subsequent development of depression is modified by level of social support
  4. High level of social support may even mediate the effect of life events into a protective factor, possibly by encouraging resilience and mastery
  5. Social support might be both a mediator and a modifier

Questions 42 to 53 refer to the following paper, given in précis below

‘Co-proxamol and suicide: A study of national mortality statistics and local non-fatal self poisonings’ Keith Hawton et al; BMJ 2003; 326;1006-8

Abstract

We ascertained the numbers of suicides from poisoning with co-proxamol, paracetamol and tricyclic antidepressants and then compared fatal and non-fatal self poisonings to estimate the relative fatality of overdoses with these three drugs.

Methods

Mortality data – We obtained data from the Office for National Statistics on deaths in people aged 10 years and over in England and Wales for 1997-9. We considered those deaths that involved drugs and medicines in which a verdict of suicide (ICD 9 E 950.0 – E959.5) or undetermined cause (E980.0 – E989.5) was recorded.. Deaths involving co-proxamol, paracetamol, or tricyclic antidepressants were identified by searching for all variants of descriptions of these drugs in the textual fields for cause of death

Non-fatal self poisonings – Non-fatal self poisonings with coproxamol, paracetamol or tricyclic antidepressants alone for 1997-9 were identified through the Oxford monitoring system for attempted suicide. This records all presentations to the general hospital in Oxford of deliberate self harm. The comprehensiveness and reliability of the data has previously been shown. The pattern of drugs used for self poisoning in the Oxford area is similar to that seen elsewhere.

Statistical analysis - We used Poisson regression to compute estimates, confidence intervals and comparison of death rates and presentation rates according to drug, age and sex. Odds ratios for the relative lethality of different drugs were calculated by computing ratios of relative death rates to relative non fatal presentation rates. Confidence intervals for relative lethality were calculated with Monte Carlo methods.

RESULTS

In England and Wales in 1997-9, 15,299 deaths were recorded as suicides or open verdicts. Of these 4,162(27%) were drug related (such as self poisoning). There were more drug related deaths in men than in women, although a higher proportion of women who killed themselves did so by self poisoning (1835/3762 (49%) v 2327/11537 (20%) in men)

Table 1. Details of suicide by coproxamol, antidepressants, and paracetamol; England and Wales, 1997-9. Figures are numbers of suicides (percentage of all drug related suicides in age group)

Drugs used for self poisoning

/ All drug related suicides plus open verdicts
Co-proxamol
alone / Tricyclic antidepressants alone / Paracetamol
alone
Male
Age (years):
10-24 / 55 (23.8) / 53 (22.9) / 8 (3.5) / 231
25-34 / 112 (19.2) / 115 (19.8) / 29 (5.0) / 582
35_54 / 155 (15.8) / 221 (21.6) / 77 (7.9) / 978
>55 / 97 (18.1) / 63 (11.8) / 59 (11.0) / 536
Total / 419 (18.0) / 452 (19.4) / 173 (7.4) / 2327

Female

Age (years):
10-24 / 42 (25.3) / 51 (30.7) / 13 (7.8) / 166
25-34 / 45 (14.6) / 72 (23.4) / 29 (9.4) / 308
35_54 / 118 (17.2) / 176 (25.6) / 75 (10.9) / 688
>55 / 142 (21.1) / 176 (26.2) / 78 (11.6) / 673
Total / 347 (18.9) / 475 (25.9) / 195 (10.6) / 1835

Total

Age (years):
10-24 / 97 (24.4) / 104 (26.2) / 21 (5.3) / 397
25-34 / 157 (17.6) / 187 (21.0) / 58 (6.5) / 890
35_54 / 273 (16.4) / 397 (23.8) / 152 (9.1) / 1666
>55 / 239 (19.8) / 239 (19.8) / 137 (11.3) / 1209
Total / 766 (18.4) / 927 (22.3) / 368 (8.8) / 4162

Table 2. Comparison of numbers (95% confidence intervals) of drug related suicides and undetermined deaths in England and Wales with non-fatal self poisoning in Oxford, 1997-9, for coproxamol, tricyclic antidepressants and paracetamol (used alone). Odds ratios shown with 95% confidence intervals

Co-proxamol
/ Tricyclic antidepressants / Paracetamol
Deaths in England and Wales/year / 255
(238 to 274) / 309
(289 to 330) / 123
(110 to 136)
Non fatal self poisonings in Oxford/year / 26
(21 to 33) / x
(64 to 83) / 356
(335 to 378)
Odds ratio for relative lethality compared with paracetamol / w
(24.9to 32.9) / 12.3
(11.5 to 13.2) / 1.0
Odds ratio for relative lethality compared with tricyclics / 2.32
( to 2.5) / 1.0 / y
(0.08 to 0.09)

42. What is the hypothesis?

  1. That more people commit suicide in Oxford than elsewhere
  2. That more people commit suicide with paracetamol in Oxford
  3. That people attempting suicide with coproxamol are at greater risk of dying
  4. That suicide with paracetamol is commoner than with coproxamol
  5. That subjects attempting suicide with different drugs are likely to differ clinically

43. What sort of study is this?

  1. Case series
  2. Case control
  3. Retrospective cohort study
  4. Nested case control
  5. Cross section

44. What specific features of the study led you to this conclusion?

  1. Retrospective, observational comparison of cases and controls on odds of exposure
  2. Prospective, observational study of relative risk of outcome with various exposures
  3. Prospective, experimental study of risks of exposure
  4. Because it studies incidence rather than prevalence
  5. Observational and controlled but not randomised

45. Regarding your answer to question 41; all of the following advantages of the chosen method over the other types of study would have influenced the investigator’s choice EXCEPT;

  1. Less liability to confounding than cohort studies
  2. Better for rare outcomes
  3. Ethics
  4. Cost
  5. All of the above

46. Regarding the investigators’ use of Coroner’s data, they have gone about this in such a way as to make their ascertainment of suicide.

  1. Less specific but more sensitive
  2. More specific but less sensitive
  3. More specific and more sensitive
  4. More valid
  5. None of the above

47. The authors claim that ‘the comprehensiveness…of the data [in the Oxford monitoring system for attempted suicide] has previously been shown’. In practice how could they have gone about showing that this was the case?