Stats, Epidem, EBM 2.5 Flashcards
A type of study with random allocation to intervention/control group but can be limited by practical/ethical problems
RCT
observational & Prospective study e.g. Framingham heart study?
What is the usual outcome measure?
Cohort study - e.g. follow a cohort who have all had exposure to an agent, to see how many develop an outcome
- measure Relative Risk
observational & Retrospective study?
what is the usual outcome measure ?
benefits?
Case-control study e.g. match cases with controls, collect data & identify possible causal agents
- measure Odds ratio
- inexpensive, quick results, useful for rare conditions
- but prone to confounding & Recall bias
A study that provides a snapshot in time?
X-sectional study (or prevalence study)
- weak evidence of cause & effect
What determines what type of significance test is used?
whether data is parametric (can be measured, usually normally distributed), or non-parametric
Examples of parametric tests for:
- paired/unpaired data?
- correlation?
- student’s t-test - paired (single group) or unpaired
- Pearson’s product-moment coefficient - correlation
Examples of non-parametric tests for:
- unpaired data?
- correlation?
- comparing 2 sets of observations on a single sample?
- comparing proportions/percentages?
- unpaired: Mann-Whitney U test
- correlation: Spearman, Kendall rank
- Wilcoxon signed-rank test (2 observations on 1 sample)
- chi-squared test (%)
What is pre-test probability?
- proportion of people with the target disorder in the population at a point in time or time interval i.e. point/period prevalence
What is post-test probability?
- proportion of pts with the test result who have the target disorder (think pp.)
= post-test odds / (1+post-test odds)
What is pre-test odds?
- odds the pt has a target disorder before the test is carried out
= pre-test probability / (1 - pre-test probability)
What is post-test odds?
- odds the pt has the target disorder after the test is carried out
= pre-test odds X likelihood ratio - where the likelihood ratio for a +ve test result = sensitivity / (1 - specificity)
Null hypothesis is rejected when it is true i.e. showing a difference between 2 groups when it doesn’t exist - a false positive
Type I statistical error
- not affected by sample size
- increased if no of end-points are increased
- determined against a preset significance level (alpha)
Null hypothesis accepted when it is false i.e. failing to spot a difference when one exists - a false negative
Type II statistical error
- probability of making this error = beta
- determined by both sample size & alpha
What is the power of a study?
The probability of correctly rejecting the null hypothesis when it is false i.e. probability of detecting a statistically significant difference
= 1 - beta (prob of a type II error)
- can be increased by increasing the sample size
% of values that lie within 1 SD of the mean
68.3%
% of values that lie within 2 SD of the mean
95.4%
% of values that lie within 3 SD of the mean
99.7%
What is standard deviation and it’s calculation
- a measure of how much dispersion exists from the mean
SD = square root (variance)
How to calculate the variance?
“mean of the squares - square of the mean”
What is and how to calculate the standard error of the mean?
= SD / square root (no. of values)
- a measure of the spread expected for the mean of the observations
- SEM gets smaller as the sample size (n) increases
Phase I of clinical trials?
- detemines pharmacokinetics & pharmacodynamics & side-effects prior to larger studies
- on healthy volunteers
Phase II of clinical trials?
- assess efficacy & dosage - involves small no of pts affected by a particular disease
IIa: assesses optimal dosing
IIb: assesses efficacy
Phase III of clinical trials?
- assesses effectiveness
- typically involves 100-1000s of people, often part of RCT comparing new & established treatments
Phase IV of clinical trials?
- Postmarketing surveillance - monitors for long-term effectiveness & side-effects
What are funnel plots used to demonstrate?
Existence of publication bias in a meta-analysis
- horizontal axis: treatment effects
- vertical: study size
What is bias?
situation in a trial where 1 outcome is systematically favoured
What is selection bias?
Error in assigning individuals to groups leading t odifferences which may influence outcome
- Sampling: subjects not representative of the population
- Volunteer: subjects more/less likely to volunteer
- Non-responder: e.g. those who smoke less likely to respond to smoking survey
- Loss-to-follow-up bias
- Prevalence/Incidence/Neyman bias: when a study Ix a condition characterised by early fatalities/silent cases
- Admission/Berkson bias: cases & controls in a hospital study different as exposure to risk & disease occurrence increases likelihood of admission
- Healthy worker effect
What is recall bias?
Difference in accuracy of recollections retrieved by study participants e.g. due to whether they have a disorder or not
- problem in case-control studies
What is publication bias?
Failure to publish results from valid studies, often as they showed a negative/uninteresting result
- important in meta-analysus where studies showing negative results may be excluded
What is Work-up/Verification bias?
In studies which compare new Dx tests with gold standards:
- clinicians may be reluctant to order the gold standard unless the new test is positive (eg if the gold standard is invasive e.g. tissue Bx)
- can really distort results of a study, altering specificity & sensitivity
- sometimes can’t be avoided
What is expectation bias (Pygmalion effect)?
- observers may sunconsciously measure/report data in a way that favours the expected study outcome
- problem in non-blinded trials
What is the Hawthorne effect?
Describes a group chaging its behaviour due to the knowledge it is being studied
What is late-look bias?
Gathering info at an inappropriate time e.g. studying a fatal disease many years later when some pts may have died alreadu
What is procedure bias?
When subjects in different groups receive different treatment
What is lead-time bias?
When 2 tests for a disease are compares, the new test diagnoses the disease earlier, but there’s no effect on the outcome of the disease
What is selection bias?
Error in assigning individuals to groups, leading to differences which may influence outcome
- Sampling: subjects not representative of population, e.g. may be due to volunteer bias, non-responder bias
- other examples inc loww to follow-up, prevalence/incidence aka Neyman, admission aka Berkson, healthy worker effect etc
What is information bias?
When measurement of info differs among study groups
- e.g. recall bias, reporting bias, Dx bias, Hawthorne effect, errors in measurement
What is confounding bias?
Distortion of exposure or disease relation by some other factor
What is detection bias?
Outcomes are sought after, more in 1 group than another
What is the p value?
= probability of obtaining a result by chance at least as extreme as the one actually observed, assuming that the null hypothesis is true
= the chance of making a type I error
What is a null hypothesis? H0
It sates 2 treatments are equally effective (hence negatively phrased)
What is a significance test?
it uses the sample data to assess how likely the null hypothesis is to be correct
Level of evidence Ia?
from meta-analysis of RCTs
Level of evidence Ib?
from at least 1 RCT
Level of evidence IIa?
from at least 1 well designed controlled trial (not randomised)
Level of evidence IIb?
from at least 1 well designed experimental trial
Level of evidence III?
evidence from case, correlation & comparative studies
Level of evidence IV?
evidence from a panel of experts
study design Grade A?
based on evidence from at least 1 RCT i.e. evidence Ia/Ib
study design Grade B?
based on evidence from non-RCTs i.e. evidence IIa/IIb/III
study design Grade C?
based on evidence from a panel of experts i.e. evidence IV
