Scenario

You are a liaison psychiatrist.

You notice that many patients presenting to A&E, despite having documented hysterical seizures (aka pseudoseizures, or non epileptic attack disorder (NEAD)) are maintained on antiepileptic drugs (or AEDs) because of an erroneous diagnosis of epilepsy. You wonder about the effect of withdrawing medication; you think that leaving individuals with pseudoseizures on AEDs reinforces the patients perception that they have epilepsy and as such their liability to have pseudoseizures, however your colleague thinks that pseudoseizures are a proxy of an underlying impulse control disorder which will worsen if mood stabilising AEDs are withdrawn.

But how to investigate this further?

Q 1

As a first step you do a literature search, looking for either observational or experimental studies with particular regard to withdrawal or maintenance of AEDs in patients with pseudoseizures. In the way of these things an astonishing array of baseline and outcome variables were assessed in the various studies, some of which are given below

  1. Prescription of AED (yes/no) at baseline and follow up
  2. Presence and type of coexisting mental disorder
  3. Receipt of DLA (high, middle, low)
  4. Baseline Dissociative Experiences Scale (DES) scores. DES sums the scores of twenty phenomenathought to represent dissociative diathesis rated as present/absent at a mild moderate or severe level

These scales are respectively

  1. Categorical, ordinal, ordinal, ordinal
  2. Categorical, categorical, ordinal, interval
  3. Categorical, ordinal, interval, ratio
  4. Ordinal, ordinal, categorical, ratio
  5. Ordinal, categorical, ordinal, ordinal

In the papersidentified by your search you find baseline attack frequency is often illustrated by recourse to diagrams

Q2

With regard to the stem and leaf plot belowshowing baseline attackfrequency/wk in 25 patients with pseudseizures,you think that there is something a bit fishy about the summary statistics on the right; and reader, you are correct because one of them is wrong. Can you tell which one?

Q3

Which statement is correct with regard to the box and whisker plot below showing baseline attack frequency in a consecutive series of patients with pseudoseizures?

  1. The plot is highly significant as indicated by the two asterisks
  2. The mean attack frequency is approximately 50, the mid point of the long horizontal line
  3. Approximately 75% of values are less than 50
  4. There are more values in the right hand side of the box than the left
  5. The long horizontal line is the confidence interval

As outlined above youare particularly interested in the effect on outcome (ie remission or persistence of pseudoseizures) in patients who continue to be prescribed AED despite a clear diagnosis of non epileptic attack disorder.

The only finding that directly addresses this issue is a case series (Abraçadas,Petons, et al, 2005). As part of the work up for patients with intractable epilepsy a series of patients were identified whose attacks were exclusively non epileptic. Subjects were given a diagnosis of pseudoseizures and the referring professional advised that AEDs could be withdrawn. Subjects were assessed at follow up (range 3 months to 6 years) to see if their attacks persisted and if this bore any relation to their being continued on AEDs

Results at end point are given in the table belowQuestions 4 to 12 refer to the table

Pseudo seizures
remit / Pseudo seizures
persist /
No AED / 15 / 10 / 25
AED / 25 / 40 / 65
40 / 50 / 90

Q4

With regard to individuals not currently receiving AEDs as compared to those who arethe RR of pseudo seizure remission is

  1. 1.56
  2. 0.65
  3. 0.61
  4. 0.375
  5. Not possible to calculate as risk only refers to negative outcomes

Q5

With regard to individuals not currently receiving AEDsas compared to those who arethe OR of pseudoseizure remission is

  1. 4
  2. 3
  3. 2.4
  4. 0.75
  5. 0.25

Q6

Assuming that withdrawal of AED preceded and caused remission of pseudo seizures the ‘number needed to treat’(NNT), rounded up to the nearest whole number, is

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5

Q7

The confidence interval for the ARR is; – 0.0321 to 0.4308. (p = 0.1080) As such the confidence interval for the NNT is

  1. 2.32 to 31.15
  2. 31.15 to 2.32
  3. 0 to 2.32
  4. 0 to 31.15
  5. 2.32 to ∞

Q8

The RR suggesting a positive effect on pseudoseizure status by AED withdrawal might be due to all of the following except

  1. Bias
  2. Confounding
  3. Chance
  4. Use of an inappropriate statistical test
  5. Reverse causality

Q9

With reference to the p value of 0.0655 this indicates….

  1. Clinical significance
  2. A trend towards significance
  3. Statistical significance
  4. High statistical significance
  5. Clinically insignificance

Q10

What is the mystery test?

  1. t test
  2. Mann Whitney U
  3. Chi square
  4. ANOVA
  5. McNemar’s Test

Q11

As compared to the two-tailed p value of 0.0655; the one tailed value would be numerically

  1. Much greater
  2. Slightly greater
  3. The same
  4. Slightly less
  5. Much less

Q12

All of the following are true of Fisher’s exact testEXCEPT

  1. It is generally (but not necessarily) reserved for 2x2 tables
  2. It is particularly useful for small samples.
  3. It has developed by Fisher to test the claim of a colleague that she could tell if the milk had been poured before or after the tea.
  4. Must be used if more than 20 % of observed values are less than 5 or any value is less than 1.
  5. Can be used if a cell value is 0.

Q13

Anyway, you are very excited by the figures fromAbraçadas,Petons, et al, 2005 and you wish to get a more robust estimate of the possible effect of AED withdrawal in pseudoseizures by doing an RCT as follows

Sequential patients newly diagnosed as suffering from pseudoseizures at a dedicated clinic at the Southern general hospital will be randomized to either immediate or delayed (6 months) withdrawal of anti epileptic drugs. All other treatment, in particular delivery of the diagnosis and psychological treatment will be the same such that, assuming randomisation succeeds, the immediate and delayed groups will only differ on the fact that they do or do not continue on AED

For your power calculation your grasping ambitious boy or girlfriend, who is desperate for you to get a publication and trample on your colleagues on the greasy psychiatric pole, tries to persuade you to do the power calculation on the assumption that the fall in pseudoseizure frequency is normally distributed, although this flies in the face of everything in your literature search..

What’s the point?

  1. He/she wants to use a non parametric statistic because they are more powerful
  2. He/she wants to use a parametric test because they are more powerful
  3. He/shefeels that the finding is important enough to risk a type 2 error
  4. Parametric statistics are easier to do
  5. Non parametric statistics are easier to do

Following randomisation baseline characteristics for the two groups were compared as followsQuestions 14 to18 refer to the table

variable / Summary statistic / Immediate withdrawal / Delayed withdrawal / p
A / Age / Mean (sd) / 37.3 (5.6) / 42.1( 6.3) / 0.043
B / Gender / % Female / 72 / 64 / 0.567
C / Hx of sexual abuse / % +ve / 32 / 52 / 0.089
D / Baseline seizure frequency wk / Median, mode
(IQR) / 13
24
(2 – 38) / 9
8
(1 – 47) / 0.741
E / Years on AED / Mean (sd) / 3.2 (12.6) / 4 (10.0) / 0.245

Which comparison was analysed by

Q14

t test

Q15

t test on log transformed data

Q16

Mann Whitney U

Q17

As regards the baseline differences in hx of sexual abuse and their effect on the endpoint, ‘50% reduction in pseudoseizure frequency’

  1. The difference is unlikely to influence the results as indicated by the non significant p value
  2. The difference is highly likely to influence the results as indicated by the significant p value
  3. Sexually abused subjects will have to be excluded from the study
  4. The study will have to be abandoned
  5. The existence of an interaction may be apparent in the analysis

Q18

The existence and extent of an interaction could be assessed at analysis via

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

Q19

This study was undertaken under the auspices of the liaison service at the Southern General. In the aftermath of the study dissemination of the results to local health care professionals you are interested to see if admissions to intensive care units for patients with pseudoseizures at the Southern General differed from those in the catchment areas of the other Glasgow teaching hospitals. The admission rates would be expected to follow

  1. The binomial distribution
  2. The poisson distribution
  3. The F distribution
  4. The t distribution
  5. The normal distribution

Q 20

All of the following would be likely to follow a poisson distribution EXCEPT

  1. Out of area admissions per month to ward 1d Crosshouse hospital.
  2. Incident cases of Huntington’s per year per GP practice in Ayrshire
  3. Phone calls to NHS 24 per day
  4. Referrals to North Cunninghame CMHT per week
  5. Suicide rates per year in individuals on the psychology waiting list