Critical Appraisal Flashcards
(107 cards)
1
Q
Confounder
A
confounders make it appear as if there is a direct relationship between an exposure and an outcome. They may also mask a true relationship. A confounder must have an association with exposure but not be a consequence of it. It must also be associated with the outcome but independently of the exposure.
2
Q
Selection bias
A
where the individuals, group or data selected for a study is not appropriately randomised meaning the sample is not representative of the desired population. This will normally be identifyed by differences in the baseline characteristics between exposure and control groups.
3
Q
Bias
A
Anything that influences the study in a non random way deviating the results from the truth.
4
Q
observation bias
A
When there is differences between how the data is gathered in a study and how the outcome is measured. Such that the results may be unduly influenced by the expectations of the researchers or subjects.
5
Q
cross sectional studies
A
studies a group at a specific point in time
6
Q
ecological studies
A
study a community or population
7
Q
pragmatic studies
A
take place in a real life setting e.g. hospital or clinic, gives good effectiveness data and external validity
8
Q
case reports
A
observational and descriptive, prone to chance association and bias. useful for generating hypothesis and as for setting for clinical reminders.
9
Q
cohort studies
A
recruiting subjects based on a risk factor being present or absent and then study going forward. May be expensive to set up and long time from exposure to outcome. Also bias may be present if subject drop otu. They are prospective, good if risk factor is rare
10
Q
case control studies
A
recruiting subjects who either have or haven’t had an outcome and then look back on data (retrospective). The disadvantages of these are that they rely on memory, old records. and can be difficult to recruit a matching control group, good if outcome is rare
11
Q
temporality of exposure
A
did exposure proceed outcome
12
Q
gradient of exposure
A
increased exposure leads to increased risk
13
Q
analogy of exposure
A
is the association analogous to any previous proved causal association
14
Q
factorial studies
A
use more than one intervention at a tme to see how they interact
15
Q
meta analysis
A
combines results of more than one study to create a quantitative assessment
16
Q
phase 0
A
microdosing for pharmacokinetics/bioavailabilty etc
17
Q
phase 1
A
in healthy individuals
18
Q
phase 2
A
in relevant illness subjects
19
Q
phase 3
A
large group in clinical setting
20
Q
phase 4
A
post marketing surveillance
21
Q
exclusion criteria
A
a common exclusion criteria is to exclude those too unwell, co-morbid or with confounding factors. Too much exclusion criteria and may reduce effectiveness of data/ cause selection bias
22
Q
simple random sampling/representative sampling/proportionate sampling
A
everyone in TP has equal chance of being selected
23
Q
systematic sampling
A
every nth person selected after first randomly chosen
24
Q
stratified sampling
A
subjects are divided into suubgroups and equal number are drawn from each
25
cluster sampling
subjects are divided into clusters and some are sampled some are not. disadvantages are that differences between clusters and that different sized clusters may be given same weighting
26
SB: berkson admission rate bias
sample taken from hospital setting
27
SB: diagnostic purity bias
excludes those with comorbidities
28
SB: survival bias/neyman/incidence-prevalence
end up analysing those who have survived or lasted long enough, therefore milder forms of disease are represented
29
SB: membership bias
those from a specific group e.g. cancer charity
30
SB: historical control bias
where groups are chosen at different times this may mean they encountered different treatments/RFs/disease
31
SB: response bias
those more likely to respond are more likely to be motivated etc
32
randomisation
allocation of different subjects to intervention groups
33
adaptive randomisation
adjustment of study arms to maintain similarity
34
open label trial
no blinding, therefore observation bias is an issue
35
triple blinding
subject, researcher and analyst dont know
36
surrogate endpoints
a biomarker used as a substitute for a clinical endpoint e.g. LDL level. Useful as can be shorter as dont need to wait for outcome but not necessarily reflective of clinical outcome
37
composite endpoints
composite of several clinical endpoints e.g. death, MI, stroke
38
validity
the extent to which a test measures what it is supposed to measure
39
reliability
how consistent a test may be on repeated measurements
40
criterion validity
compared to an existing test
41
relevance of a large drop out
if in the experimental arm could suggest intolerable treatment or lack of efficacy of treatment, if in placebo arm may indicate individuals in group may have required treatment
42
difference between systematic error and random error
random error is variable, systematic error is consistent e.g. scales that measure 1gram too low
43
test-retest reliability
level of agreement of two assessments of the same data/sample
44
accuracy
how close is the measurement to its true value
45
precision
how close repeat measurements are
46
incidence
number of new cases over a period of time/population size
47
mortality rate
incidence of death, morbidity rate is incidence of new non fatal cases
48
prevalence
number of people with a given disease at a specific time/population size. i.e. the proportion of a defined population with the disease at a given time
49
lifetime prevalence
proportion of a population who has had or has a given disease at a given point in time
50
discrete vs continuous data
discrete includes whole numbers continuous is infinite
51
qualitative data
non numerical, can be converted to quantitative data by using cut off points e.g. HTN 130/90
52
nominal scales
no relation to eachother
53
ordinal scales
inherent order
54
interval scales
start at a set point, 2 degress is not twice as hot as 1 degrees
55
probability distribution
links all the possible values of a random variable with the likelihood of that occuring, e.g. flipping a coin once, heads has a probablity distribution of 0.5
56
a binomial experiment
a fixed number of runs in which a random variable can have two possible outcomes
57
binomial distribution
the probability distribution of a binomial random variable, yes or no
58
poisson experiment
two possible outcomes of a random variable but number of runs is not fixed
59
poisson distribution
can be used to determine the probablility of an outcome from a poisson experiment i.e. if you know on average 5 babies are born a day on a ward you can calculate the probablility that exactly 6 will be born tomorrow
60
bimodal distribution
two modes therefore two peaks
61
normal distribution/gaussian
mean=mode=median, bell shaped
62
weighted mean
increased significance or weight is attached to some values
63
variance
is the distance of all the different values from the mean, average distance from the mean
64
standard deviation
a statistical measure that describes the degree of data spread about the mean, it is the square root of variance. 1SD 68%, 2SD 95%, 3SD 99.7%. can be used to calculate the proportion of variables that lie between any two values
65
z score
the distance of a value from the mean using standard deviation as a unit i.e. +1.2 is +1.2SD from the mean
66
coefficient of variance
The coefficient of variation (CV) is the ratio of the standard deviation to the mean. The higher the coefficient of variation, the greater the level of dispersion around the mean.
67
effect size/standardised mean difference
compares the mean difference between experimental and control groups and devides it by the standard deviation. if bigger then bigger effect
68
positively skewed
mean>median
69
negatively skewed
mean
70
kurtosis
how peaked is the distribution
71
central limit theorum
the idea that if you take lots of samples, the mean of all the sample means will be the same of the population mean. The standard error of the mean is the standard deviation of the sample mean
72
95% confidence interval
if normally distributed, 95% of the means in a sample will like within 1.96 standard errors above or below the population mean. The confidence interval may be smaller with a larger sample size and bigger with a smaller population size. It is the range around the mean that we think the true population mean falls into 95% of the time. can be worked out with the sample mean and the standar error. the wider/larger the confidence interval the less certain the result
73
per protocol analysis
only those who complied with the trial protocol and completed are analysed
74
intention to treat analysis
all sibjects who were randomised are included in the analysis regardless of if they dropped out, this is more representative of real life; includes worst case scenario, hot desk (similar subject results transposed), last observation carried forward
75
last observation carried forward
can either underestimate or over estimate effect
76
absolute risk
the percentage chance of an outcome, can be either CER (control event rate) or EER (experimental event rate)
77
absolute risk reduction
the change in risk between the control group and the experimental group = CER - EER
78
relative risk
the ratio of risk in the control compared to the experimental group = EER/CER, if = 1 there is no increased risk, if 2 then twice as likely, if 0.5 twice as unlikely
79
relative risk reduction
is the drop in risk from the experimental group from the control group = CER-EER/CER
80
numbers needed to treat
number of subjects who need to have the intervention to have benefit compared with the control. reciprocal of the absolute risk reduction = 1/CER-EER
81
odds ratio
used to compare how likely outcomes are between groups, the odds ratio is the ratio of the odds of having the outcome in the experimental group relative to the control group, an odds ratio of 1 reflects the sasme outcome in both groups, if >1 than odds are greater in the experimental group
82
null hypothesis
any difference is due to chance
83
p value
is the probability of getting the expressed value by chance, the smaller the p value the less likely it is to be due to chance and therefore the more likely it is to be statistically significant
84
type 1 error
false positive, usually attributable to bias or confounding or data dredging
85
type 2 error
false negative, usually due to small sample sizes, can use power calculations to figure size of sample required to avoid type 2 error
86
regression
expresses the relationship between two or more variables, regression lines appear on correlation tables
87
confounding management: restriction
using inclusion and exclusion criteria to restrict confounding factors from entering the sample population
88
confounding management: matching
people with confounding factors are allocated evenly into different study arms
89
confounding management: randomisation
confounding factors known or unknown are randomised into different study groups
90
sensitivity
the true positive rate, the proportion of patients with the disorder who have a positive test, if sensitive a negative test will rule a disorder out (SNOUT)
91
specificity
the true negative rate, the proportion of patients without the disorder who have a negative test, if specific a positive result will rule a disorder in (SPIN)
92
positive predictive value
proportion of patients with a positive result who actually have the outcome,
sensitivity is if a patient has a disorder what are the chances they have a positive test, whereas, PPV is if the patient has a positive test what are the chances they have the disroder
93
negative predictive value
proportion of patients with a negative result who do not have that outcome
94
likelihood ratios
calculated from specificity and sensitivity, shows how many more times likely a patient with a disorder is to have a particular test than a patient `without the disorder.
95
hazard rate
probability of an endpoint event at a time interval divided by the duration of the time interval . hazard ratio compares experimental arm to control arm
96
Quality Adjusted Life Years
number of extra years of life obtained x value of the QoL during those years. death =0 QALYs, 1 year of perfect health = 1 QALY
97
minimising bias in qualitative studies
reflexivity/bracketing, member checks, triangulation of data by cross referencing
98
meta analysis
aids detection of statistical significance by pooling together studies often with smaller samples, results shown in a forest plot
99
PRISMA
27 item checklist for systematic reviews
100
publication bias
large studies whether with significant or insignificant findings tend to be published, small studies with significant findings tend to be published, we tend to miss small studies with negative findings in MAs/SRs leading to a bias towards +ve results. Can be detected with funnel plots
101
Methodology CA
```
Is there a focused clinical question?
How were the groups chosen and allocated?
Was there blinding?
Are the groups similar?
What are the baseline demographics?
Are there any confounders?
Are the endpoints relevant?
Does exposure proceed outcome?
Is length of follow up sufficient?
Is there consistent measurement between groups?
```
102
Results CA
```
appropriate use of;
What are the NNT/NNH?
Sensitivity/specificity?
PPV/NPV?
Confidence intervals?
Are the results reliable and valid?
```
103
Discussion/Applicability CA
```
Are the patients similar to TP?
Are RFs similar to TP?
ShouldRF be stopped/minimised?
Did this answer the clinical question?
Is the null hypothesis proved or disproved?
Is it possible to apply this to your clinical practice?
Is there any evidence of bias?
Are there any conflicts of interest?
```
104
cost effectiveness analysis
analyses the cost compared to the effectiveness of a treatment to figure out if it provides good value for money
the cost of a treatment divided by its effectiveness e.g. the cost of a treatment/QALYs.
105
incremental cost effectiveness ratio
difference in costs between two treatments divided by the difference in effectiveness
106
a good screening test
Wilson Junger criteria:
- an important health problem
- natural history well understood
- detectable early stage where treatment is more beneficial than at a late stage, test should aim for this stage
- test should be acceptable i.e. SEs/risks/invasiveness
- benefits outweigh the risks
- cost effective
107
power of a stady
the probability of finding a difference between the two study arms when a difference does exist, 1-type 2 error, probability of correctly rejecting a false null hypothesis
