Statistics Flashcards
Systematic review/meta analysis
Most reliable
Randomised control trial
Least biased due to randomisation
Can prove causality
Intervention versus placebo
Cohort studies
Observational. Association not causality
Exposure versus no exposure
Prospective or retrospective
Relative risk >3
Case control studies
Observational. Association not causality Disease versus no disease Good for rare diseases Can do with small numbers Most biased Odds ratio >4
Cross sectional
Frequency of disease and risk factors at a particular point in time
Risk factor association not causality
Prevalence
Forest plot
Size of square - weight given to the study
Placement of square - mean effect of the intervention (ODDS RATIO)
Horizontal lines - size of the CI (narrower = more precise)
Diamond - meta analysis findings (overall result)
Edges of diamond = CI
if the edges of the diamond crosses the vertical line = NOT statistically significant
if p <0.05 then statistically significant
Clinical trials - phase 1
SMALL number, HEALTHY
Is it safe? SAFETY, TOXICITY, pharmacokinetics
Clinical trials - phase 2
SMALL number, with the DISEASE
Treatment EFFICACY, optimal DOSING, adverse effects
Clinical trials - phase 3
LARGE number, RCT, randomly assigned to treatment under investigation or placebo/best available treatment
Compares new to current/no treatment. SAFETY compared to gold standard, SIDE EFFECTS
EFFICACY of experimental therapy, compare to other treatments
Clinical trials - phase 4
Post marketing SURVEILLANCE of patients after treatment is approved
Detects rare or long term side effects
Prevalence
All cases = # existing cases/ total # people in population
Incidence
New cases = # new cases / # people at risk
Incidence down = PPV down
Normal distribution
1 SD = 68%
2 SD = 95%
3 SD = 99.7%
Confidence interval
95% CI = p 0.05
If 95% CI includes 1 - NOT statistically significant
CI between 2 groups overlaps - NOT statistically significant
CI between 2 groups don’t overlap - statistically significant difference
Narrower the CI = more precise (less likely due to chance)
Sensitivity Specificity
Screening test - sensitivity **
Sensitivity and specificity are independent of prevalence
Prevalence same as PPV but opposite to NPV
(Incidence ~ Prevalence ~ Pretest Probability ~ Positive Predictive Value)
High sensitivity = less false negatives therefore miss less people
POS for disease correctly tests POS
High specificity = less false positives
NEG for disease correctly tests NEG
Positive likelihood ratio
Probability of a positive test in a person with the disease compared to a person without disease
= SENSITIVITY / (1 - SPECIFICITY)
Negative likelihood ratio
Probability of a negative test in a person with disease compared to a person without disease
= (1 - SENSITIVITY) / SPECIFICITY
Absolute Risk
Rate of occurrence of a disease (incidence in exposed)
Absolute risk reduction
= CONTROL EVENT RATE - EXPERIMENTAL EVENT RATE
Number needed to treat
How many patients need to be treated to prevent one event
= (1/ARR) x 100
Relative risk
RCT/cohort studies
Incidence rate in exposed / Incidence rate if not exposed
RR = 1 then risk A = risk B
RR >1 INCREASED risk of outcome in exposed person
RR <1 REDUCED risk of outcome in exposed person
RCT small RR, cohort RR >3, case control RR >4
e. g. RR 1.36 = risk DISEASE is INCREASED by 36%
e. g. RR 0.8 = risk DISEASE is reduced by 20%
Relative risk reduction
Population of the baseline risk which was reduced by a given intervention
Absolute risk reduction / Control event rate
e.f. RR 0.39 therefore 61% less likely to get disease
But if baseline population is 20% then what is the likelihood?
61% less than 20%. ~60% of 20% = 12%
So 20-12 = 8. 8% risk of Dx or 0.2 x 0.39 = 0.078 ~8%
Odds ratio
Case control study
Odds of EXPOSURE in those WITH DISEASE /
Odds of EXPOSURE in those WITHOUT DISEASE
OR = 1 No change in frequency of exposure
OR <1 decreased frequency of exposure among cases
OR >1 increased frequency of exposure among cases
Hazard ratio
Probability that individual at time t has an event at that time
Similar to RR but takes time into account
HAZARD IN TREATMENT ARM /
HAZARD IN CONTROL ARM
Null hypothesis
No difference between the same or population being compared, differences due to random variation
Type 1 error
Picking up significant difference when there ISN’T one
Risk of FALSE POSITIVE result
Reduces with randomising, blinding
Type 2 error
MISS a difference when there actually is one
Risk of FALSE NEGATIVE
REDUCES with LARGER sample size***
