Statistics Flashcards
How do you calculate NNT?
1/ ARR
(Absolute risk reduction)
How do you calculate ARR (Absolute risk reduction)?
ARR = ARC- ART.
(Absolute risk in controls - absolute risk in treatment)
ARR = Event rate in control group - event rate in experimental group
How do you calculate a mortality rate?
Deaths / number in the group
What is the best definition of a cohort study?
A cohort study follows a group of people over time, looking at a specific outcome (e.g. death) and whether exposure to a given risk factor (e.g. smoking) contributes to that outcome
How do you decide if a study or overall finding is significant in a box plot?
Does the whiskers (Confidence interval) cross the HR = 1 line
(Or the corners of the diamond if overall)
How do you measure hetero or homogenicity in a meta analysis?
Higgins I2
What Higgins I2 values would suggest:
a) Homogenicity
b) Statistically significant heterogenicity
a) Perfect homogenicity would be an I2 value of 0%
b) Significant statistical heterogeneity is often considered to be present if I2 is over 50%
How do you calculate odds ratio?
Odds of event in exposure group / odds of exposure in control group
So (event in exposure group/ non-event in exposure group ) / (events in controls/ non-events in controls)
How do you calculate sensitivity?
True positive / True positive + false negative
OR written another way
True positive results / All who have the disease
How do you calculate specificity?
True negative / True negative + false positive
True negatives results / All who don’t have the disease
How do you calculate positive predictive value?
True positives / all positive results
(True positive / true positive + false positive)
How do you calculate negative predictive value?
True negatives / all negatives
(True negative / true negative + false negative)
What is positive likehood ratio and how do you calculate it?
Compares odds of having disease if test positive vs. odds of having at baseline
(So a PLR of 2 means you are twice as likely to have a true positive if the test was positive compared to baseline population)
How do you calculate positive likehood ratio?
Sensitivity / 1-specificity
How do you calculate negative likehood ratio?
1- Sensativity / Specificity
Name two qualitative statistical techniques?
Delphi method - Expert opinion taken from several rounds of questionnaires
Ethnography - Study of social interactions by interview or observation in participant own environment
What are two alternative names for a pilot study?
Feasability or vanguard study
Small scale used before larger study, helps to test ideas
What is action research?
Done by the person who will use the results to improve their own service
Name 2 descriptive and 3 analytical types of observation studies?
Descriptive:
- Case report or case series
Analytical:
- Cohort
- Case control
- Cross-sectional
What is a case control study?
Retrospective
- Looks back to the past to dervive patterns
What is the main way to distinguish between case control, cross-sectional and cohort studies?
Case control - Retrospective, look at groups in the past
Cross sectional - Current perspective, snapshot in time now
Cohort - Prospective, follow two groups into the future
What is the main thing measured by case-control studies?
Odds radio
(looks at groups (with and without disease) and then works out OR of different exposures
Main disadvantage of case control study?
Recall bias
Also can’t do causation
Main disadvantages of cohort studies?
Time consuming and expensive
(lots of resource to follow up risk groups)
What is the main thing measured by cohort studies?
Risk ratio (RR)
Looks at outcomes of groups with different exposures over time
When would a cross-sectional study be used?
To assess prevalence at one point in time
What is recall bias?
Disease status of subject affects liklihood of recalling exposure
What is sampling bias?
Some members of population more likely to be sampled than others
What is lead time bias?
Lead time bias occurs when two tests for a disease are compared and although one of the tests diagnoses the disease earlier, this does not translate in to a survival benefit
Only results in an earlier diagnosis (so mortality not affected, morbidity may increase as patient has to live with knowledge for longer)
What is late-look bias?
When information is gather at inappropriate time, important to consider for fatal diseases
What is the hawthorne effect?
Study group changes their behaviour because they are being studies
What is the pygmalion effect?
High expecatons of treatment lead to improved performance and changes in the outcome
What is the incidence of a disease?
Rate of new cases over given time period
New cases/ population
What is the prevalence of disease?
Total number disease cases at a specific time
All cases/ population
How do you summise the difference between incidence and prevalence?
Prevalence = all cases
What is a meta analysis?
Combines data from 2 or more RCT’s
What is a crossover RCT?
Participants starts like a normal RCT, but after first treatment phase participants are reallocated to another treatment arm
What is a systematic review?
Qualitative review of multiple pieces of research
What two ways can you compile the results of multiple trials?
Qualitative = Systematic review
Quantitative = Meta analysis
Name the research hierarchy
Clinical guidelines
Meta analysus/ systematic review (MA/SR)
RCT
Cohort study
Cass control study
Cross-sectional study
Case report/ case-series
Expert opinion
Note - more letters in the name, lower down hierarchy
How do you calculate:
a) Mode
b) Median
c) Mean
a) Most common
b) Middle (line all up)
c) Add all up and divide by number of those
What is:
a) Continuous data
b) Discrete data
c) Categorical data
a) Can be anythinany value (height, weight etc)
b) Must be whole number (i.e. number of participants)
c) Has to be in specific category (i.e. colours)
What are the two types of categorical data?
Ordinal - Can put in order (small, medium, large)
Non-ordinal- Can’t put in order (i.e. blood type)
How do you calculate median number if even number of readings?
Take 2 numbers in the middle
Add together and divide by 2
How do you calculate range in a set of data?
Largest - smallest
What is true about mean, median and mode in normally distributed data?
All will be the same
Where is the mode on a distribution curve that is postive or negative curve?
Mode at the top of the curve
Right and left skew, which positive and which negative?
Left skew - starts big - Positive skew
Right skew - starts small - Negative skew
(Positive skew, peak is sooner - negative skew, peak is later)
In positively and negatively skewed data what is the mean, mode, median relationship?
Positive (peak sooner) - Mode (peak) < Median < Mean
Negative (peak later) - Mean < Median < Mode (peak)
What is parametric data?
Normally distributed
Non parametric would therefore be skewed data
In normal distribution how much data is within 1 SD?
68%
What is measured by variance?
Large - Numbers in set are far from mean and far from each other
Small - Numbers are close to mean and close to each other
Variance 0 = all values the same
How is standard deviation calculated?
Square root of variance
In normal distribution how much data is within 2 SD’s?
95%
In normal distribution how much data is within 3 SD’s?
99.7%
What is a type 1 error?
A type 1 error is rejecting the null hypothesis when it is in fact true
(An innocent person being convicted - we see a difference where there is none)
What is a type 2 error?
A type 2 error is accepting/ failing to reject the null hypothesis when it is false
(guilty person not being convicted - not seen a difference where there is one)
How do you calculate the power of a statistical study?
1 – (type 2 error)
The likelihood that the test is correctly rejecting the null hypothesis (i.e. “proving” your hypothesis)
Which SINGLE condition does the FAMCAT clinical case-finding algorithm identify?
Familial hypercholesterolaemia
How do you decide if ethical approval is needed for a study?
Clinical audit and service evaluation do not warrant mandated REC review, but research does.
If randomisation is used, it is research
With respect to urgent referrals for suspected cancer, what is the relationship between referral rates, mortality, and conversion rates?
Higher urgent referral rates result in lower cancer mortality
The conversion rate (the proportion of urgent referrals which led to a cancer diagnosis) was not associated with high or low use of the urgent referral pathway.
The department of general practice at the local university is proposing a study to investigate whether the use of proton pump inhibitors increases the risk of osteoporosis.
Which is the SINGLE MOST appropriate study design to use in this case?
Cohort study
(Allow you to calculate relative risk)
Case-control studies can show associations, but cannot establish causality so wouldn’t be appropriate
What p-value is the common cut off for significance, what does it mean?
<0.05%
Means that there is a 95% chance that the null hypothesis (i.e. that there is no difference between the control group and intervention group) is false and that the intervention has resulted in a real and significant difference
What is absolute risk?
Probability something will happen
Cases vs. total number population
How do you manage a NNT which you calculate as 33.3?
Always round up never down
So would be 34
When do you use:
a) Risk ratio
b) Odds ratio
c) Hazard ratio
a) For propective studies (cohort)
b) For retrospective studies (case-control)
c) For Kaplan Myer graphs
How do you calculate odds or risk ratios?
Absolute risk in event group / absolute risk in control group
How do you calculate relative risk reduction?
1-Relative risk
A statin demonstrates a relative risk of 0.8 of having a stroke. What is the relative risk reduction?
1- Relative risk
Or 0.2 in this case
How does Hazard ratio differ to relative risk/ risk ratio?
Hazard takes into account time
How does mortality rate differ with case fatality rate?
Mortality rate is deaths in population
Case fatility rate is deaths in all those with the disease
What is standardised mortality rate?
Number of deaths / number of ‘expected deaths’ x 100
How do you interpret standard mortality rate?
< 100 = Less deaths than expected
> 100 = More deaths than expected
What is Bayesian probability? How is it expressed?
More flexible than traditional probabilty
What is a DALY?
Disability adjusted life year
(measures disease burden in populations - number of years lost due to ill health, disability or early death)
DALY = Years of life lost + years lost due to disability
What is a QALY?
Quality adjusted life year
Considers quality and quantity (Utility value x years of life)
1QALY = 1 year of life in perfect health
0= Death
What is the difference in calculations between QALY and DALY?
DALY = Years lost + years lost due to disability
QALY - Quality x number of years
What is the difference between what is measured by QALY and DALY?
QALY - Measures benefit with and without intervention (more individual)
DALY - Measures overall burden
(population measure)
What is the null hypothesis, how is it noted?
H0 (null hypothessis)
There is no difference between two measured phenomenon
(Aim of an experiment is to reduce the null hypothesis)
What is the difference between type 1 and type 2 errors?
Type 1 (alpha) error - Incorrectly rejecting the null hypothesis when it was true (seeing a difference where there is none)
Type 2 - Accepting a false null hypothesis (not seeing a difference where there is one)
When is a result deemed?
a) significant
b) Highly signficant
a) p<0.05
b) p<0.001
What does “regression to the mean” refer to?
The more you measure a variable the closer you will get to the true mean
What is triangulation and what research is it used for?
Assessing a problem in multiple ways
Often in qualitative research
Aiming to gather wider and more broad understanding
What is validity of a trial?
Does the study measure what it is supposed to?
(internal validity - was design good?)
(external validity - was it analysed well)
What is reliability?
How repeatable or consistent are the results if repeated
What is generalisability?
The extent to which a study can be applied to other settings
What is the difference between validity, reliability, and generalisability?
Validity - Was the study designed (internal) and analysed (external) - well
Reliability - Are the results reproduceable
Generalisability - Will it apply in a variety of settings
If a test has a 95% confidence level (p=0.05) what is the chance of a type 1 error?
5% chance
You attend a local diabetes update course where a recent randomised controlled trial is presented.
What would BEST illustrate the flow of patients through the trial and the reasons for attrition and dropout?
CONSORT diagram
The CONSORT statement is an evidence-based, minimum set of recommendations for reporting RCTs.
The statement comprises a 25-item checklist and a flow diagram. The checklist items focus on reporting how the trial was designed, analysed and interpreted, and the flow diagram displays the progress of all participants through the trial.
