Epi Flashcards
Primary prevention
Stop getting the disease
Eg wearing suncream
Secondary prevention
Early identification and treatment
Treat RFs
Cure / prevent progression
Tertiary prevention
Rehabilitation of people with established disease
Aims to reduce numbers/ impact of complications
When should primary prevention start
As early as possible. E.g. During or pre conception
Why cant do RCT for questions about prognosis
unethical to make them wait if there is a treatment
Sensitivity calculation
proportion of people truly positive with the disease = true positive / total
Specificity calculation
probabiluty of -ive test result in people without the disease= true -ive /tatal
How to tell whats true positive or negative
if gold standard test is positive and the new intervension is +, then true +
if gold standard test is negative and the new intervension is -, then true -
which one is good to rule things out? in? a specific or sensitive test?
SnNout Sensitivity; rules out (eg D-dimer) (if high and -, then they most likely dont have the condition)
SpPin- specific; rules in ( if high and + then they most likely have the disease)
Positive predictive value
the probability of having disease if you test positive
Negative predictive value
the probability of not having disease if you test negative
work up bias
- when gold standard test is too expensive/invasive
- so you only use it in advanced disease eg kidney biopsy
- so you may end up overestimating the sensitivity
reporting bias
was it blinded to investigators?
publication bias
difficult to publish -ive results
spectrum bias
A type of sampling bias
Pt mix in one clinic may be completely different from another
Therefore performance of a diagnostic test may be completely different
+ive likelihood ratio
LR+= Sensitivity / 1-specificity
Likelihood ratio
values close to 1 indicating no better than random
if higher than 10, it is likely and conclusive of disease presence
Intension to treat analysis
Analyse data from everyone randomised no matter whether they dropped out or not throughout the trial
Per protocol analysis
Just analyse people who completed/complied with the study protocol, exclude dropouts
Sensitivity analysis
Doing the analysis using different assumptions, eg if seeing if certain drug gives you hyponatraemia, analyse for Na < 135 mmol/L, then Na < 130, then etc.
See if lower Na levels have an effect on the outcome
Pragmatic study
Very inclusive, generalisable
Assesses real world
Explanatory RCT
Selected (exclusive) patients
Assesses best case, efficacy
Cluster randomisation
Groups rather than individuals
Advantages of cluster randomisation
Avoid performance bias (treating similar patients differently)
Avoid contamination
Disadv of cluster randomisation
Greater sample size required
Randomised before individual consent taken
Prevelance calc
No individuals / total population at risk
Incidence calc
No of new cases/ population at risk (disease free)
Incidence rate calc
No of new cases/ (population at risk * time interval)
Risk ratio
risk in exposed/risk in unexposed
Odds ratio
ods in exposed /odds in unexposed
Risk vs odds calculation
Risk: 1 in 10 people gets MI , Risk 1/10
Odds: 1 (MI) / 9 (healthy)
Precision vs accuracy
Precision - all the darts on a board in a similar area; big sample size shows true value
Accuracy - all the darts on bulls eye, low bias shows true value
SD vs SE
SD of samples means is SE
95% CI calc
+/- 1.96 SE
Quality adjusted life year
measure of health= morbidity (quality of life) *mortality (quantity of life)
QUALY interpretation
1= best imaginable health
0.5 intermediate health state
0 worst- death
ICER calc
Incremental cost-effectiveness ratio
Difference in cost/difference in effect (QUALY)
Problems with ICER
hard to interpret, is ICER negative because negative QUALY or negative cost
Alternative to ICER
Net monetary benefit
If NMB is positive, intervention is cost effective
(takes into account how much health we get when we spend money)
Horizontal equity
equal treatment of equals
Vertical equitiy
unequal treatment of unequals
Fixed effect meta analysis assumption and weighting
homogeneity
Larger studies get higher weighting in the ultimate conclusion
When use a random effect meta analysis and not a fixed effect
If there is risk that there is heterogeneity of data (difference in results of studies due to methods used and not random error)
i.e. if I squared value is bigger than 25, a random effect meta analysis should be used
Heterogeneity tests
Q score (if p value is significant, the heterogeneity) I2 statistics (if above 25 then more likely for heterogeneity)
Which one is better fixed or random effect meta analysis
fixed if possible provides stronger evidence
NNTB
number needed to treat to benefit
1/ (Risk difference)
risk difference = R0 (risk of exposed) - R1(risk of unexposed)
NNTH
number needed to treat to harm
1/ (Risk difference)
risk difference = R1 (risk of unexposed) - R0 (risk of exposed)
type 1 error
suggests there is a statistical difference when there isnt any
type 2 error
suggests there is no difference when there is
Berksons bias
controls selected from hospital patients
Attrition bias
loss from one group more than the other in RCT