CPP2067 Interpretation of epidemiology studies Flashcards
what are the 5 main questions to ask when assessing a study/paper?
- Bias?
- Confounding factors?
- Chance?
- is the association causal?
- are the findings generalisable/apply to all patients?
what is selection bias?
- when there is a systematic difference between the characteristics of the people selected for a study and those who were not
- people picked for some kind of reason (consciously or unconsciously)
- e.g. target all working age adults in England, but only sample people on campus at UCL on a Wednesday afternoon = picks only students who don’t play sport
what is measurement bias?
- when measurements or classifications of disease are inaccurate (they do not measure correctly what they are supposed to)
how do you identify selection bias in epidemiological studies?
- was the study population clearly defined?
- what were the inclusion and exclusion criteria?
- were refusals, losses to follow up kept to a minimum and reported?
how do you identify selection bias in cohort and intervention studies?
- are the groups similar except for the exposure status?
- is the follow up the same for all groups?
how do you identify selection bias in case-control studies?
- do the controls represent the population from which the cases arise?
- was the identification and selection of cases and controls influenced by the exposure status?
how do you identify measurement bias in epidemiology studies?
- were the exposures/interventions of interest clearly defined using standard criteria?
- were the measurements as objective as possible? (i.e. measure height with a ruler rather than asking)
- was the study blinded as much as possible?
- were the observers/interviewer rigorously trained?
- was information provided by the patient validated against existing records?
what is confounding?
- when an estimate of the association between an exposure and the disease is mixed up with the effect of another exposure on the same disease
- e.g. less sleep = increased mortality
- however, there is a confounding factor of age as less sleep is associated with older age which is associated with increased mortality
= age = confounder of the relationship between lack of sleep and mortality
how can confounding be resolved?
- stratification = stratify the analysis according to confounder status (adjust and include factor)
- statistical modelling = statistical adjustments can be made to control for confounding factors (e.g. multivariable analysis)
what is chance?
-when using a sample drawn from a population of interest we make inferences about the true value in the population based on the observed estimate from the sample
- however different samples from the same population can yield different estimates due to sampling variation or chance
how can the role of chance in studies be assessed?
- by calculating the confidence interval
what are confidence intervals?
- a range of values which we are 95% confident will contain the true values of the mean measure of interest in the overall population
what does p<0.05 mean?
= significant p value = reject null hypothesis
= in 5 times or less out of 100 samples there is no association between the exposure and the outcome (not necessarily causal)
what is p-hacking?
- only selecting preferable/better findings to have a significant p value
what does a narrow confidence interval mean?
- that is is more precise/accurate (more likely to be significant)
