Epidemiology Flashcards
validity
how close results are to true value
reliability
if u repeat the experiment, how similar will ur results be to each other
type 1 error
reject null hypothesis when it should be accepted-think it has an effect wen actually it doesn’t
type 2 error
accept null hypothesis when it should be rejected (is false)-ur saying it doesn’t have an effect wen in fact it does
prevalence
no of existing cases in one particular pnt in time
incidence
occurrence of new cases over time period
• Cumulative incidence
new cases over a specific period
• Fatality
frequency of deaths
• Validity
how close to the true value
• Reliability
if u repeat the experiment lots of times, u will get the same result if reliable
T test
compares means from 2 groups
measures continuous data
chi square test
- association bw 2 categorical variable exposure n outcome
- (e.g. smoking [Y/N] vs. lung cancer [Y/N])
- for categorical variable (
Correlation & Linear Regression:
linear relationship between two continuous variables (e.g. age and knee cartilage volume.)
Logistic Regression:
Outcome = binary (Y/N event). Output = OR.
Survival Analysis:
“time to event” analysis – outcome is continuous time
diagnositic test vs screening test
screening= mass testing, see if unsymptomatic/apparently healthy ppl have disease
diag= • tests those who are likely to get disease (ie from screening/from symptoms). confirms if someone has a disaease or not
true +ve
have disease n test positive
true -ve
no disease n test negative
false +ve
no disease but test positive
false -ve
have disease but test negative
sensitivity
true positive/total diseased
specificity
true negative/total non-diseased
PPV (positive predictive value)
true positive/total positive (i.e. probability person has disease wen test positive
NPV (negative predictive value)
true negative/total negative (i.e. probability don’t have disease wen test negative)
sensitivity n specificity
dont change with prevalence of disease
PPV n NPV
change with disease
PPV low wen prevalence low
NPV high wen prevalence low
