Epidemiology
Application of the scientific method to the study of disease or ill health and its alleviation amongst defined population.
Prevalence - the number of cases at any one point of time
Incidence - the number of new cases over a set period of time
Population estimates
UK population is estimated using census data
Births are added, deaths are subtracted, allowances for migration
Crude birth rate is the ratio of live births in a year and the total estimated populations in the same year.
Estimated population growth uses crude death/birth rates
Risk
Disease distributions are rarely random
Relative risk - the ratio of the probability of disease occurring in a subpopulation compared to another
e.g. relative risk of smoking is 1.5 if early mortality due to CVD being 30% risk for smokers and 20% risk for non-smokers
Relative risk should be interpreted with absolute risk
To avoid bias, our samples need to be representative of the population where every member has an equal probability of being sampled
Confounding - other factors can influence the relative risks and the outcome of interest
Reducing Confounding
Case-control design
Statistical control
Compare like for like
CCD:
Cases in subpopulations are matched to control cases in other subpopulations using specific criteria
Cases - patients with a specific outcome
Control - find a control without a specific outcome with similar characteristics
Test all cases and controls for the outcome
Statistical: using analysis techniques to statistically adjust for all dimensions. Save the relevant available. Usually longitudinal
e.g. Framingham Heart Disease Study
Problem: the size of the sample required for appropriate analysis.
Effort and cost are required to collect relevant data.
Simpsons PAradox
Comparison of 2 treatments for kidney stones
A works better on both small and large stones
B works better oevrall
Stone size is a confounding variable
Randomised control trails are the best for demonstrating causality. It can’t be used for ethical or logistic reasons often.
Prospective (longitudinal)
Events can’t cause other events that occur prior to them
Prospective designs address this reverse causality by measuring variables at different time points to discover sequence.
There is always the possibility that an unmeasured variable may be the source of correlation.
Controlled means that you intervene systematically in one group but not another. It means the other group is treated exactly the same (except they do not receive treatment)
Randomisation means you allocate people randomly to the control or treatment. Everone has an equal chance of being selected. Random allocation removes pre-existing differences as there’s no bias.