Academic Flashcards
What study reporting guidelines exist (6)?
CONSORT - RCT
IDEAL - Surgical Innovation
MOOSE - Meta-analysis of observational studies
PRISMA - systematic Reviews
STROBE - Observational studies
TRIPOD - Predictive models
What is a null hypothesis?
The initial hypothesis stating there are no differences between the items studied
What is the sensitivity?
Proportion of true positives correctly identified
Proportion of patients with a disease that test positive
What is Specificity?
Proportion of true negatives correctly identified
Proportion of patients without a disease that test negative
What is the absolute risk reduction?
The difference in event rates between two groups
The NNT is the inverse of this
What is a regression analysis?
A statistical means of establishing the dependence of a variable on one or more different variables.
Practically, regression analyses are often used to either generate predictive models or to establish the indepent effect of different variables
What is the variance and SD?
- Summary of variability of a dataset. SD is the square root of the variance.
- In normally distributed data, 68% of the data lies within 1 SD and 95% in 2 SD
- measure of the central tendancy of the sample
What is a confidence interval?
A measure of the precision of study estimates compared to population values
How can the quality of RCTs be assessed?
JADAD score
1) Was the study randomised
2) Was the study double blind
3) Was there a description of dropouts and withdrawals
4) Was the randomisation appropriate?
5) Was the blinding complete?
CONSORT guidelines
How is an impact factor defined?
IF 2021 = (total citations in 2021 for articles published in 2019+2020)/number of articles published in 2019+2020
What is a power calculation?
- Usually conducted a priori
- Calculation of sample size required to reach a specified power level (usually 80% chance of avoiding a type. 2 error)
- requires an expected effect size from previous work
What are the levels of evidence?
1a - SR/MA RCTs/ Consensus Guidelines
1b RCT
1c All or nothing trial
2a SR/MA Cohort studies
2b Individual Cohort studies
2c Outcome studies
3a SR/MA Case-control
3b Case-control
4 Case series
5 Expert opinions
What is the GRADE recommendations?
A - Level 1 studies, or consistent 2a/2b/3 studies
B - generally consistent findings from Level 2/3 studies
C - inconsistent findings from Level2/3 studies
D - no systematic evidence
GP - good practice guidelines
What is a RCT?
Level 1b evidence where patients are randomised to different treatments and outcomes compared
Pros
- Provide strong evidence of cause and effect
- randomisation minimises most types of bias that affect other studies
- blinding reduces observer bias
Cons
- Time consuming and expensive
- can lack equipoise
- analysis of subgroups difficult
- require long follow up period
What is a cohort study?
Group of patients identified for study without reference to an outcome. Can be prospective or retrospective
Pros
- Can study multiple outcomes
- Good for rare exposures
- Can measure incidence
- May infer causality
Cons
- Bias increased (particularly for retrospective)
- Confounding variables inevitable
- Loss to follow up
- Not useful for rare outcomes
How might randomisation be conducted?
Simple usually algorithmic
Block — small subgroups of patients are given equal allocations to keep the overall sample sizes equal throughout the study
Stratified — patients allocated so that measured covariates (e.g. gender, comorbidities) are equally distributed between groups
How are RCTs analysed?
Either with
1) Per-protocol —- only including patients who completed treatment as prescribed (a better measure of the true effect of a treatment)
2) Intention to treat — analysing all patients who were allocated to a treatment arm (a better measure of the real world effect of a treatment)
What is a case-control study?
Comparison of two groups, one with outcome and one without outcome
Advantages
- Useful for rare outcomes
- Can assess multiple exposures
- Quick/easy/cheap
Disadvantages
- Cannot establish causality
- Retrospective and biased
- Cannot establish prevalence
- Can only calculate OR
What is the OR?
Odds = number with/number without
OR = Odds 1/Odds2
What is the Risk Ratio?
Relative Risk = proportion with/proportion without
Risk Ratio = RR1/RR2
What is a Kaplan Meier?
Typically this represents a graph of treatment failure over time such as death.
The Kaplan Meier method is a statistical approximation that incorporates censoring of patients that have unequal follow up. As if these patients were excluded then the analysis would be very pessimistic
What is a hazard ratio?
The instantaneous risk of an outcome at a particular time in follow up
What is a ROC curve?
Used with diagnostic tests and predictive models to assess performance at different probability cut-offs
A chart of sensitivity on the y axis against 1-specificity on the x axis for all possible thresholds.
The AUROC is equivalent to the C-index
A non discriminative model would have an AUROC of 0.5
Moderate 0.6, good 0.7 and excellent 0.8. A threshold of 0.8 is reasonable
What is an H-index?
Number of papers (‘h’) cited at least ‘h’ times
H-index of 5 means 5 papers cited at least 5 times each
What is the difference between fixed and random effects models in Meta-Analysis?
Fixed = universal effect size assumed = weighting based on size
Random = variable effect size = weighting varied = wider confidence intervals = more conservative
What are the phases of randomised trials?
Phase 0 Exploratory, first in human trials at sub therapeutic doses (<12)
Phase 1 Assess for any beneficial effect (20-100)
Phase 2 - A = dosing B = efficacy
Phase 3 - Multicentre RCT
Phase 4 - Post hoc
What is allocation concealment?
Prevention of clinician from knowing allocation in advance
What are the requirements for screening?
UK National Screening Committee criteria (20)
Condition - important, primary prevention undertaken, natural history understood
Test - good, accurate, well studied and acceptable test. Agreed policy for next steps after a positive result
Intervention - treatment and evidence available for treatment
Screening programme - evidence of efficacy, benefits outweigh harm, cost effective
Various implementation criteria
How would you assess the quality of a subgroup analysis?
Design
Analysis
Reporting
Applicability
What is propensity score matching?
Create balanced groups for comparison
Approximate randomization by regressing out specified series of confounders
Can make most cohort studies better
Can also do ‘weighting’
Need to measure residual imbalance using standardized mean difference
What is a collaborative authorship model?
Distributive authorship.
Reducing the incentive to publish small single-centre data of questionable relevance
Cons minimise the importance of people who actually do most of the work, although writing groups etc
What is a non-inferiority trial?
Trial conducted to ensure a new treatment is not unacceptably worse than an existing treatment (inferiority margin)
Works with Reversed null hypothesis - lack of significance of standard trial does not mean treatments are equivalent
How can an impact factor be manipulated?
Self citation
Journal self citation
Some journals only include original articles/review in denominator and include others (letters, news, editorials) in numerator
What is a cost-benefit analysis?
Comparison of costs of an intervention and those resulting from it, including productivity gains, other costs averted and monetised value of health improvements.
NICE QALY threshold is £20-30k
What are the impact factors of:
JAMA Surgery
Annals
BJS
BJS open
Surgical Endoscopy
EJSO
JAMA Surgery -16.7
Annals - 13.79
BJS - 11.12
BJS open 3.7
Surgical Endoscopy 3.7
EJSO 4.4
What are some common types of bias?
Selection sample not reflective of population
Confounding unconsidered variables affecting outcome
Reporting Bias selective reporting of outcomes
Lead-time bias screening detects disease earlier and patients appear to survive longer
Confirmation bias investigator bias towards particular outcome
Publication bias +ve papers > -ve papers published
Attrition bias drop out rates differ between groups
Measurement bias (Hawthorne effect) measurement changes behaviour
Classification bias definitions of groupings too vague
Recall bias study events recalled preferentially by some patients
What types of bias are relevant in screening?
Selection
Lead time
Length (aggressive tumours less likely to be screen detected)