Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
65 Cards in this Set
- Front
- Back
PICO
|
P = patient description
I = intervention tested (superior to) C = comparison O = expected outcome |
|
Steps required to analyse a study
|
1) PICO
2) Clinical question - is it therapy/intervention, diagnosis, prognosis, risk, observation 3) Compare study design with ideal for this clinical question 4) Comment on quality of results (RAMBO) |
|
Formula for sensitivity
|
S = TP / (TP + FN)
|
|
Formula for sensitivity
|
Se = TN / (TN + FP)
|
|
Positive predictive value, what does it imply
|
PPV = TP / (TP + FP)
High PPV means good diagnostic test |
|
Negative predictive value, what does it imply
|
NPV = TN / (TN + FN)
High NPV means few false -ves, therefore, good screening test |
|
When is RCT appropriate
|
1) effectiveness of new therapy
2) assess risk factor 3) compare aetiology |
|
Why cannot an RCT always be performed
|
1) obtaining adequate sample size and retention of enough individuals for follow up
2) ethics (eg testing potentially harmful drugs on pregnant women) 3) Costly |
|
Why is RCT the gold standard for drug research
|
1) biases removed by blinding researcher and subjects, as well as by randomisation
2) Prospective, therefore establishes timing and direction implying causality |
|
Why is an inception cohort study considered less convincing than RCT
|
1) selection of cohort & no double blinding > selection bias
2) may be looking at past records > measurement bias |
|
Why is case controlled study not a determinant of causality
|
1) begins with selection at future date for groups with disease and control group then looking into past for exposure factors > causality cannot be determined
2) multiple confounding factors may be in control and test groups > selection bias 3) looking at past records and recollections > measurement bias |
|
In evaluating a diagnostic test, what is the best form of study, where does it rate on the hierarchy of EBM
|
1) Dx: Cross-sectional reference study (compared to "gold standard")
2) Rates below meta, RCT, inception cohort & case-controlled study |
|
We wish to determine the prevalence of a disease. What is the best method
|
Cross sectional study using a random sample
|
|
In risk prediction, what types of studies are most suitable
|
Who will develop the problem, therefore:
1) RCT 2) inception cohort 3) case control study |
|
What type of study would be appropriate for diagnosing an outcome
|
random cross sectional sample and testing with the "gold standard"
|
|
In critical appraisal, what are the 5 categories of questions investigates (note the most appropriate type of study)
|
1) Frequency (random or consecutive sample)
2) Risk prediction (incidence: RCT or cohort, CCS possible) 3) Diagnosis (random or consecutive sample) 4) Prognosis (inception cohort) 5) i Intervention (RCT) |
|
If 40 died is a treatment group and 50 in the control group and there were 200 in each group, what is the absolute risk reduction
|
ARc = 25%
ARt = 20% ARR = 5% |
|
If 40 died is a treatment group and 50 in the control group and there were 200 in each group, what is the number needed to treat
|
No of lives saved = 50 - 40 = 10 ie. 10 lives were saved by the treatment but we needed 200 to save 10. Therefore:
NNT = 200/10 = 20 patients to save 1 person |
|
If 40 died is a treatment group and 50 in the control group and there were 200 in each group, what is the relative risk
|
RR is the risk of treatment compared to the risk of doing nothing ie 40/200 / 50/200 = 80%
|
|
If 40 died is a treatment group and 50 in the control group and there were 200 in each group, what is the relative risk reduction
|
First calculate relative risk of the treatment ie 40/200 / 50/200 = 80%. If the treatment was ineffective then RR=1 (same as placebo). Therefore, RRR = 1-0.8 = 0.2
|
|
If the absolute risk reduction = 5%, what is the number needed to treat
|
If ARR = 5%, then 5 people were saved out of 100. Therefore, NNT = 100/5 = 20
|
|
What is the ideal study for the 5 main types of clinical question
|
1) therapy - RCT
2) diagnosis (test): cross section compared with gold standard 3) prognosis: inception cohort 4) risk: can be cohort of CCS 5) prevalence: random sample |
|
Why are placebos and blinding important in trials
|
Blinding the patient and researcher minimises biases
|
|
In conducting a critical appraisal of a study, what is RAMBO
|
R = representative & randomised
A = Ascertin if all subjects finished (< 20% losses OK) M = Measurements (same for all groups) B = blinded O = objective results and conclusions |
|
What is the meaning of 100% sensitivity
|
All people who were actually sick were diagnosed as sick ie no one was diagnosed as no sick (no negatives)
|
|
What does 100% specificity mean
|
All people from the healthy group were diagnosed as not sick and noone was diagnosed as sick (no positives)
|
|
What is 100% positive predictive value mean
|
All people tested +ve with the particular test were correctly diagnosed.
|
|
What is PPV
|
No of people correctly diagnosed as sick compared to the number diagnosed as sick in total (ie true positivves / (true + false positives)
|
|
What is the NPV
|
No of people corrrectly diagnosed as not-sick compared to the total number diagnosed as not-sick (includes false negatives)
|
|
A study found an association between carrying matches and cancer. Is this a true association or what is the likely confounder
|
There is an association, therefore it is either true or confounded. The likely confounder is that cigarette smokers are likely to carry matches
|
|
What is the key difference between a confounder and an associated variable
|
A confounder must be independent to the variable and directly associated or a cause of the effect eg diet is not a confounder of cholesterol and chd because cholesterol is dependent on diet
|
|
How can you minmise confounders in a cohort study, what about case control
|
restriction to eliminate variation in the level of the confounding factor eg match carrying and non smokers
Matching: two groups (eg exercisers v non exercisers for CV disease) and match confounders eg (gender, smokers, age). With cc need to match analysis also? |
|
If after stratifying, we work out the odds ratio for both situations (with and without the suspected confounder) we find odds ratios equal and 1. What does this mean
|
the suspected confounder is a true confounder and if the odds ratios are equal and one, the variable is not associated variable
|
|
What is effect modification
|
A confounder acts differently with different associated variables eg low birth weight is assiciated with leukaemia but gender is also. With girls, high birth weight is associated and girls low. Therefore the adjusted ORs will not be equal
|
|
If you stratify for gender, smoking and age (3 groups) how many strata and adjusted odds ratios are formed
|
12
|
|
Two groups are treated with different antihistamines. What is the flaw in this study
|
no comparison with placebo to determine placebo effect
|
|
what are the Bradford Hill criteria of plausability
|
Temporal relationship between cause and effect (problem with case control studies)
dose response (ie strength of the effect) plausability (mechanism) |
|
Give 3 examples where communication improves outcomes
|
info on drug side effects improves compliance
preoperative info saw patients needing less post op analgesia patients with chronic back pain reduced amount of analgesia |
|
how do patients perceive risk
|
over assess low probability, high consequence, unfamiliar v familiar, short time frame v long time frame
|
|
Give examples of nominal and continuous variables
|
Nominal: marital status
Continuous: height |
|
To get a normal distribution, what assumptions are made
|
Large number of independent variables acting upon a single continuous variable
|
|
What does it mean if the distribution is not normal
|
Median <> mean
Could have a mixed population, meaning that the mean and SD are irrelevant eg males and females measured together re heights, weights etc |
|
What different populations are within 1 and 2 SDs
|
1sd is 68%
2 sd is 95% |
|
If the mbw is 3.5kg and the sd is. 5 kg, what is the birth weight range of 95% of newborns
|
range is +- 0.5x2 =1 kg, hence 2.5 to 4.5kg
|
|
The sd of weight is 0.8 and the mean is 10.4. A girl presents with a weight of 8 kg. How many SDs is she from the mean
|
Diff is 2.4 kg from mean = 3 sds therefore she is less than 99.9 % of the population ie severely malnorished.
|
|
What is the z score
|
the value/sd
|
|
A girl has a z score of 1.2, and an SD of 0.8 and mean of 10,, calculate the value
|
mean + sd x z = value (could be height, weight or any continuous variable)
|
|
What is a problem regarding all experimental models and interpreting the result
|
we compare 2 groups (affected v unaffected) involving 2 different populations with different means
|
|
What sort of number is used for determining standard error and confidence interval
|
sample statistic rather than population parameter
|
|
What does the confidence value indicate
|
it shows the variability between the means of different samples to make an inference regarding the population
|
|
Formula for standard error of sample relative to a population sd
|
sd of population divided by the square root of the size of the sample
|
|
What does the standard error indicate
|
An estimate of the difference between the true mean and the sample mean
|
|
Define error
|
Sample mean - true mean
|
|
Why is there uncertainty in diagnosis. What are the ethical issues involved
|
Normal overlap of pathological and normal
False negatives and positives of tests Little knowledge about less serious diseases Some diseases have no diagnostic test Variation of severity of symptoms Causes over investigation and over treatment |
|
What is Murtagh's diagnostic methods and what are the pitfalls
|
Probability diag, not to be missed, masquerades
Problems: leads to habit forming and missed conditions |
|
What is the red flag/alarm signals approach to diagnosis
|
Red flags: symptoms absent rules out a serious condition
Alarm signals: if present then signals serious illness Problems: may not have good evidence base, may develop incorrect heuristics |
|
How is probabilistic reasoning used to make a diagnosis. What are the problems.
|
Identify pre test probability based on population studies
Determine sensitivity and specificity of test Determine post test probability Problems: what if post test probability is inconclusive, may still have to rely on other methods Tests not available for all conditions |
|
What are the pitfalls of pattern recognition diagnosis
|
May not recognise the pattern
Pattern changes with time and type of patient eg childhood v adult asthma |
|
Why is CAM a problem for medicine
|
1/6 use CAM as primary health care
Extrordinary variability and little evidence base Variability of training Legitimacy without responsibility Self care interacting with medical care or resulting in avoidance of timely medical treatment until condition became more severe. CAM not communicated to doctor Opposition to EBM baded care by medical prefession by CAM therapists |
|
Why has CAM use increased
|
Feeling of involvement and personal power over illness
Loss of faith in medical profession and drug companies Stigmatisation by doctors impairs communication regarding any CAM treatments which can spill over into patient management |
|
What is risk
|
Probability of an event occurring
|
|
What is a risk factor. What are the types. How can they affect yor results
|
Risk factors contribute to increasing risk
Modifiable and non modifiable Risk factors can act as confounders |
|
Attributable risk. Why is it important
|
How much of the incidence of a disease is attributed to a certain exposure.
Usually the target of treatment trials |
|
What is the problem with RR and RRR in identifying if a treatment is clinically relwvant
|
Become problematic when the AR is very low, therefore a small decrease in AR can result in a large RRR which is a distortion of the clinical picture
|
|
The relative risk of developing lung cancer in smokers is 14 whereas it's 1.6 for CVD. Why is the social impact greater for CVD
|
The attributable risk is greater for CVD because there is a far greater incidence of CVD in the population, therefore, a small increase in relative risk causes a large increase in the numbers affected and the corresponding burden of disease
|