Medical stats Flashcards
Hazard ratio
Chances of an event occurring in the treatment arm/chances of event occurring in the control arm. Displayed in tandem with survivorship curves ie shows temporal progression of events (per unit time). Used in survivor analysis
Layman: Chances of an event happening per unit time.
Odds
Expression of relative probability of event happening / event not happening. Odd = 1 mean that the event is a likely to have happened as it was not to have happened. Example is a case control study if we take a group of 100 smokers and 20 of them had cancer, the odds would be 20/80 = 0.25
Odds ratio
Odds of having been exposed to something in case arm/odds of having been exposed to something in control arm
Primarily used in case-control studies (ie retrospective)
Layman: X% more likely to have been exposed to Y than in control group
Absolute risk
Just risk of something happening in either control or treatment group.
Equation = Number of events/total population in group
Layman: X% chance of something happening
Absolute risk reduction
Absolute difference in chances of something happening between intervention as opposed to control
Equation = Absolute event rate in treatment arm – Absolute event rate in control arm
Layman: Difference in chance of X% of something happening
Recommended by CONSORT to be reported alongside relative risk reduction in all RCTs to make sue risk is not overestimated
Relative risk/risk ratio – NB rate ratio is a similar concept but with rates rather than risk
Risk of event occurring in treatment arm relative to control arm
Equation = Absolute risk in treatment arm/Absolute risk in control arm or 1-RR X 100
Layman: X% more likely to happen than in control group
Alternative: Risk of event in group 1 / risk of event in group 2. Risk is events / all in a group. Example RR of 2.5 means that participants in group are 2.5 more likely to have developed the outcome,
Relative risk reduction
Essentially reduction in risk compared to original risk of control
Equation = Absolute risk reduction/Absolute risk in control arm (think of it as percentage increase/decrease)
RRR = 1 - RR
Layman: X% more likely to happen than in control group (you don’t need to do maths in your head for the 1 – as you would for a RR/OR interpretation)
Number needed to treat
Number of patients needed to obtain one outcome of interest e.g. if a certain procedure reduced the risk of death by 10% then you would need to do it 10 times for this effect to be visible. In cohort studies can be expressed as numbers needed to expose (as no intervention)
Equation = 1/Absolute Risk Reduction
Layman: Ten patients would be needed to see one less X event
95% Confidence intervals
Range of estimates for an unknown parameter. In 95% CIs that means that if you run the experiment 100 times, the true value will be within the interval in 95 times. Derived from SEM
Alternative: Range within which you are 95% sure the true mean of the population lies.
Receiver operating characteristic (ROC) curve – used in making scoring systems
Graph showing diagnostic capability of binary classifier
Y axis: Sensitivity (true positives)
X axis: 1 – specificity (false positives)
Area under curve represents predictive performance (0.5 for a random test, so the higher above this the better)
Sensitivity and specificity
Sensitivity
Chances that a test will detect someone who actually has the disease (test focussed)
Equation = True positive/total with disease
Specificity
Chances that a test will not detect someone who actually doesn’t have the disease (test focussed)
Equation = True negative/total people without disease
PPV and NPV
Positive predictive value
Chances that someone who receives a positive result actually has the disease (patient focussed)
Equation = True positive/total positive results
Negative predictive value
Chances that someone who receives a negative result actually doesn’t have the disease (patient focussed)
Equation = True negative/total negative results
P-value
The P value is defined as the probability under the assumption of no effect or no difference (null hypothesis), of obtaining a result equal to or more extreme than what was actually observed.
The P stands for probability and measures how likely it is that any observed difference between groups is due to chance.
Usually accepted at threshold of <0.05 in medical sciences meaning that there is a 5% chance of the outcomes we found being a false positive (i.e. we found a result and the true results is non-significant)
Null hypothesis
The baseline hypothesis in research starting that there is no difference between two groups. The role of RQ and following methods is to refute null hypothesis.
Power
Probability that the test rejects the null hypothesis correctly, when alternative hypothesis is true. Often marked by 1 - beta. Often at 0.8 or 0.9 in medical sciences trials. Ability to pick up true positives / not have false negative. Beta - probability of type 2 error. 1 - beta = probably of true positive.
Underpowered = sample size to small to pick up differences and possibility of not rejecting null.
