Statistics + Epidemiology Flashcards
Recall 2 x 2 table
- PPV
- NPV
- Specificity
- Sensitivity
- Pos likelihood ratio
- Neg likelihood ratio
2x2 table. How to calculate:
-RR
- OR
- NNT
- NNH
What is positive predictive value and how to calculate?
Positive predictive value
* Predictive value of test
* Chance if the est is positive of patient actually having the disease
- precision of test
* PPV = True positives / all the positives
PPV = T+ / (T+ + F+)
What is negative predictive value and how to calculate?
Negative predictive value
* If the test is negative, what is the chance of patient actually not having the disease
* NPV = True negative outcome/ over all negative outcomes
NPV = T- / (T- + F-)
Sensitivity - what, how to calculate, when most useful?
Sensitivity
* True positive rate of test
* Out of all people with condition, how many does test report is positive?
* Only concentrating on people with the condition (Condition +)
* Sensitivity = True positives / (true positives + false negatives)
* Test has high sensitivity - reports or overreports condition, unlikely to miss condition
Test has sensitivity, negative result is useful for ruling out disease
Specificity - what, how to calculate, when most useful?
Specificity
* True negative rate
* Of all the people who don’t have condition, how many does the test report as negative
* Specificity = true negatives / (true negatives + false positives)
Test with high specificity, if have positive result useful to rule in disease
Calculate positive predictive value from Sens/Spec/Prevalence
- steps involved
1) True positives
2) False positives
3) Positive predictive value
Likelihood ratio + - what, how to calculate
- Assess value of performing a diagnostic test
- Uses sensitivity and specificity of test to determine whether a test result usefully changes the probability that a condition exists
- Probability of a true positive (with disease) to false positives (without disease)
- LR = sensitivity/ (100% - specificity)
Positive likelihood ratio - larger big number, better, increases post test probability - worthwhile testLikelihood ratio of a positive test
- LR = sensitivity/ (100% - specificity)
Likelihood ratio of negative test - what, how to calculate?
Likelihood ratio of negative test
* Probability of false negative (with disease) to a true negative (without disease)
LR - = (100% - sensitivity)/ specificity
Negative likelihood ratio - lower closer to 0, better, decreases post test probability, helpful to rule out - useful test
Pre-test probability - what, how to calculate
Pre-test probability
* Probability of patient having disease before diagnostic test is known
* Essentially prevalence within subject population
Pre-test probability = proportion of patients with disease/all patients with the symptoms (with and without disorder)
Bias - confounding, ecological fallacy, selection, recall?
· Confounding - distortion in measure of risk factor and outcome by mixing effect with another exposure
· Ecological fallacy - not applying clinical decision instrument to patient in front of you
· Selection bias - group selected for study not representative wider population
Recall bias - do not correctly remember events in past
What is incidence?
- Rate
Eg cases per 5 years
What is prevalence?
- “Point”/ cross section in time
- Point prevalence studies
Prevalence = Incidence x mean survival
- Point prevalence studies
Screening tests - what do you want?
- High sensitivity
- High pos likelihood ratio - bigger number, increase post test probability
Pos predictive value
- High pos likelihood ratio - bigger number, increase post test probability
WHO guidelines for screening? (Wilson’s criteria)
- Important health problem
- Need treatment
- Facilities to diagnosis and treatment
- Latent stage
- Test or examination
- Test acceptable to population
- Natural history - adequately understand
- Agreed policy on whom to treat
- Cost, economical
Continuous process
Relative Risk?
- RR = a/(a+b)/ c/ (c+d)
- Ratio of outcome between two exposures or treatments
- Probability of an event occurring in exposed group compared to the non exposed group
- RCT/cohort studies
- Incidence rate exposed/ incidence rate unexposed
- RR = 1 = 1 same
- RR > 1- increased risk in exposed
RR < 1 - decreased risk in exposed
Odds Ratio?
- Event occurring: event not occurring
- Compared to probability - event occurring/ total events
- Odds and probability of having a girl or boy
- Odds = 1:1, probability 0.5
OR = (a/b)/ (c/d) = ad/cb
* Used when population risk unknown in case-control studies With low prevalence, OR~ RR (mathematically)
Absolute risk reductions vs relative risk reduction?
Absolute Risk Reduction
* Population change
* Real difference in absolute terms between the two exposure or treatment groups
* Absolute risk reduction = events in control/total control - events in treatment/total treatment
* In above - relative risk 50% reduction (what drug companies like quoting) Absolute risk reduction - 20% die to 10% die = 10% (actual 10% difference)
Number needed to treat?
- How many patients needed to be treated to prevent one event
NNT = 1/ (probability in exposed - probability in non exposed)
NNT = 100/absolute risk reduction (%)
NNT = 100/10 = 10
Number needed to harm?
NNH = 100 /absolute risk increase (%)
Type 1 error (alpha error)?
- Alpha error
- False positive
- Pick up significant difference when there ISN’T one
- Reject null hypothesis when it is actually true
- Report your findings are significant when they have occurred by chance
- Alpha = probability of making Type 1 error
- Usually set at 0.05
Reduce by: randomisation, blinding
Type II error (beta error)?
- False negative
- Failing to reject a null hypothesis when it should have been rejected
- MISS a significant difference then there is one there
- Reduce by: larger sample size, power
- Concludes no significant effect when actually there is
Greatest chance of falsely rejecting a good treatment (type II error) is when the study is underpowered - need to good enough sample size
Phase 1 drug design?
initial safety
* Small number, healthy
* Is it safe? Safety, toxicity, PK
* Aim = safety
* Often (n=20-100) healthy volunteers to determine safe dosing ranges, identify some adverse events/side effects
* Subjects observed for several half lives
Safe dosing range
Phase II?
- Small number with disease, efficacy, optimal dosing, adverse affects
- IIA - dose
- IIB - efficacy
- Few hundred participants
Some complete IIA and IIB together - efficacy and toxicity
Phase III?
safety/efficacy compared to gold standard
* Aim - confirmation of safety and efficacy
* Large number, RCH, randomly assigned to placebo/best available treatment
* N = 100s to 1000s
* Confirm effectiveness vs placebo/active treatments “gold standard”
* Safety and efficacy and side effects compared to gold standard
* RCTs - often > 1
Involvement of regulators FDA/EMA to obtain approval
Phase IV?
(post marketing surveillance)
* Post marketing surveillance after approved, rare/long term side effects
* Aim - continual pharmaco-viligance
* New uses/populations
* Often paediatric patients at this stage
* Larger populations, less controlled
* Longer follow up
Interactions with other medications
Main study designs?
Systematic review/meta-analysis
* Most reliable
RCT
* Least biased- randomisation
* Can prove causality
* Intervention vs placebo
Cohort
* Have group of people with an exposure, measure for lots of different outcomes
* Observational assoc not causality
* Exposure vs non exposure
* Prospective vs retrospective
* RR > 3
Case Control
* Disease vs no disease
* Look back at exposures
* Observational assoc not causality
* Good for rare numbers
* Small number
* Most biased
* OR > 4
Cross sectional
* Frequency of disease and risk factors at particular point in time
* Risk factor association not causality
* Prevalence
Case Series/reports
Normal distribution SD numbers?
- 1SD = 68%
- 2 SD = 95%
- 3SD = 99.7%
Confidence Intervals?
- 95% CI = p 0.05
- If 95% CI includes 1 - NOT stat signif
- If CI between 2 groups overlaps - NOT stat signif
- If CI between 2 groups don’t overlap - sig sif
Narrower CI - more precise