Statistics Flashcards
How to calculate SENSITIVITY:
How many ACTUALLY positive people, tested positive
True Positives
__________________
True Positives + False Negatives
How to calculate SPECIFICTY:
How many ACTUALLY negative people, tested negative
True Negatives
____________________
True Negatives + False Positives
How to calculate POS PRED VALUE:
Likelihood that a positive test is a TRUE positive. How likely patient is to HAVE the disease, in light of a positive test
True Positives
_________________
True Positives + False Positives
How to calculate NEG PRED VALUE:
Likelihood that a negative test is a TRUE negative. How likely patient is to NOT HAVE the disease, in light of a negative test
True Negatives
__________________
True Negatives + False Negatives
Type 1 and Type 2 error:
Type 1 - False Positive
Type 2 - False Negative
Confidence Interval:
A range within which it is thought the true result lies.
Chooses a MEAN and a STANDARD DEVIATION either side of it. This is the range, the CI.
A 90% confidence interval means a 10% chance of being wrong. A 95% CI means a 5% chance of being wrong. (ie. that the true result does NOT lie within that range).
The wider a confidence interval (the more results it spans) the less precise it is.
Can’t test/measure everyone in a population- use a SAMPLE instead. CI helps us interpret how likely that sample is to reflect the whole population
If a CI includes a null (eg. relative risk 1), then it is not statistically significant.
Heirarchy of literature quality:
Systematic Review vs Meta-Analysis:
Systematic review answers a defined question by selecting a number of studies that fit pre-set eligibility criteria. These studies are then scrutinised for validity, and kept or discarded. The combined data is then summarised.
Sometimes the data undergoes meta-analysis: it is pooled and reanalysed (as though there was a huge sample).
–> This isn’t always appropriate (eg. if the samples aren’t sufficiently comparable)
A metanalysis will actually re-analyse the data. Some systematic reviews simply compare/ integrate.
What is Likelihood Ratio:
PLR: How much more likely a person is to have a disease, following a positive test result.
NLR: How much less likely a person is to ave a disease, following a negative test result.
It is a measure of utility of the test.
Ie.
T1 bleed has 50:50 chance of being miscarriage. Once USS shows a FHR, only 10% chance. USS/FHR has strong NLR
Types of bias in clinical trials:
Selection bias: baseline characteristics of groups allocated to control vs intervention, are systematically different. (Randomisation helps).
Performance Bias: The groups systematically receive different care or exposures (blinding helps).
(eg. Those in control group receiving more frequent ABG from nurses than those in Tx)
Detection Bias: Systematic differences between groups in whom a finding is/isn’t found.
(eg. assessor may note more post-op pain in someone who DIDN’T receive the analgesia intervention) (Blinding helps).
Attrition Bias: Systematic differences in groups who withdraw/ are omitted from study
(eg. low SES withdrawing from a study that required frequent drives to hospital)
Reporting Bias: Which analyses the authors choose to actually report on in their study.
If bias and result are in opposite directions, study may still be deemed valid
Relative risk:
The chance of a negative outcome in one group, versus another.
Eg. risk of a stroke WITH and WITHOUT aspirin. Risk of a heart attack in MEN vs WOMEN.
RR 0.5 = HALF the risk of a bad outcome.
RR 1 = unchanged.
RR 3 = 3x MORE likely to have a bad outcome.
Odds ratio:
The strength of connection between an exposure and an outcome.
Eg. Odds ratio of smoking and lung cancer is 40!!
Similar, but not the same, as risk ratio.
Risk = chance of something happening (out of all possibilities)
Odds = Chance of something happening vs NOT happening.
Accuracy:
How close to the REAL value a measurement is.
(Precision = how repeatable a measurement is)
Major Stats
SENSITIVITY
- Pos tests/ all actual pos cases (TP+FN)
SPECIFICITY
- Neg tests/ all actual neg cases (TN+FP)
POS PREDICTIVE
- Actual pos cases/ all who tested pos (TP+FP)
NEG PREDICTIVE
- Actual neg cases/ all who tested neg (TN+FN)
LIKELIHOOD RATIO
- Likelihood of a disease NOT existing after a negative test, or existing after. positive test (about the test performance)
—> (50:50 risk of miscarriage in T1 bleed. Once USS neg (FHR), risk 10%)
RELATIVE RISK
- Likelihood of a negative outcome in one group, vs another
—> (RR of MI in men vs women is 1.5)
ODDS RATIO
- Likelihood of an outcome happening, vs not happening, in the setting of an exposure
—> (odds ratio lung Ca with smoking is 40)
