Stats definitions Flashcards
Should studies be registered?
Yes, all studies should be registered on a publicly accessible database
What is the null hypothesis?
Hypothesis stating that there is no real difference between the two groups, and any difference is due to chance
What is the alternate hypothesis
Hypothesis stating that there is a real difference between the two groups
What is bias?
Any tendency that influences the results of the trial causing over / underexageration of the results, other than the experimental intervention
What is blinding?
Technique used to eliminate bias by hiding intervention from the patient, clinician (double) and data analyst (triple)
What is evidence based medicine?
Using current best evidence judiciously to make the best decisions for the patient
What is efficacy?
Performance of an intervention under ideal and controlled circumstances
What is effectiveness?
Performance of an intervention under real world conditions
– think that “effective” is layman word > reflects real world, while “efficacious” is scientific
What is alpha?
Alpha = probability of rejecting Ho due to chance = False positive error = T1Error
What is alpha usually set as
0.05
What is beta?
Beta = probability of accepting Ho when it should have. been rejected = False negative = T2 Error
What is power (definition and as maths terms)
Power is the study’s ability to accept H1 when true (True positive)
Power = 1 - beta
What is power usually set as?
0.8
How can you increase power in a study?
By increasing the sample size
What is a type 1 error?
False positive.
You reject Ho even though Ho was true
What are causes of T1 error?
Bias
Confounding
Data dredging (cherry picking with multiple hypothesis testing)
What is a T2 error?
False negative
You accept Ho even though Ho was false
What are causes of a T2 error?
Sample size too small
Variance is too large.
What studies to heterogeneity and homogeneity apply to?
systematic reviews
What is heterogeneity?
the amount of incompatibility of trials included in the review, whether clinical (studies clinically different) or statistical (results different from each other)
What is homogeneity?
When studies included in a systematic review are clinically and statistically similar
What is incidence v prevalence?
Incidence = number of NEW cases occurring over a specific time
Prevalence = proportion of population with the condition in a specific time
What is internal. validity
Indicates how well the study backs the conclusion
i.e. the extent to which study methodology accomplishes what it set to accomplish
What is external validity?
the generalisability of the results to non-study population
depends on incl/exclusion criteria, patient demographics
What is a confounding variable?
A variable which is not the experimental variable but that may affect trial results (i.e. independent factor associated with both exposure and outcome)
What are ways to reduce confounding variables?
stratification, regression or randomisation
What is a confidence interval
the range in which the population value lies 95% of the. time
What is the NNT
Number needed to treat
So number of patients that need to be treated in order. for one to benefit
What is the ARR
Absolute Risk Reduction
So the amount by which your therapy reduces the risk of a bad outcome
So if a drug reduces risk from 50% to 30%, ARR = 0.5-0.3 = 0.2
How do you calculate the NNT based on ARR
NNT = 1/ARR
What is the NNH and what should it be ideally
Number Needed to Harm
Ideally as BIG as POSSIBLE.
What is the Hazard Rate
The probability of an endpoint in the time interval / duration of that time interval
What is the Hazard ratio and what does the hazard ratio give you that other stats dont?
Hazard in the intervention group / Hazard in the control group
It tells you the effect of an intervention on an outcome in the INTEREST OF TIME
What is the absolute risk?
incidence rate of the outcome
= outcome in either control or experimental arm / total n participants in arm
What is the absolute risk reduction (ARR)?
AR in control - AR in experimental group
What is the Relative Risk?
prob (risk) of event in exposed group : prob. of event in non-exposed group
= EER/CER
What is RR used in?
In cohort studies - it requires knowing the total number of people at risk (exposed)
So in retrospective case control studies it cannot be calculated
What is the Odds ratio?
And what is it used in
Ratio of odds of outcome occurring in experimental group vs. control group
Cohort or case control
What its the differentce between OR and RR
OR = ratio of two odds
RR=ratio of two probabilites
When would you use. RR over OR
RR requires total people at risk in the denominator > can only be used in cohort studies
What is sensitivity
True positives on test / everyone with the disease
What is specificity
True negatives/ everyone without disease
What is PPV
True +ve on test / everyone testing +ve on test
What is NPV
Those with negative. result
Who do not have the result
What are two key advantages of intention to tx studies
Represent what happens IRL
Ensures maintainance of comparability between groups (for randomisation, maintaining sample size, eliminating bias)
What is another important piece of stats data you need when looking at OR/RR?
the confidence interval
When is the confidence interval for absolute difference and OR /RR indicative of no significant difference and why?
absolute difference: when it crosses 0
RR/OR: when it crosses 1
What does it mean when OR or RR = 1
that there was no differennce between the two groups
what contexts are HR useful for?
time-to-event analysis
OR
Survival analysis
What does it mean we say HR looks at the context of time?
It tells you the probability that an individual woulld experience an evennt at a particular given time point after the intervention
i.e. at any particular time
what is the confidence interval?
range between which the population mean value willl lie in 95% of the time
== 95% sure the population mean is contained within the CI
Why do we need to calculate the CI
because the sample point estiimate mean will always be different to the true population mean
what is a per protocol study
only data from subjects who complied with trial protocol through to completion are considered
what are +s and -s of per protocol study
ADVANTAGES: shows true treatment effect (accurate representation of event)
DISADVANTAGES
- attrition bias - loss of randomisation
- exclusion bias - excludes patients who have had bad side effects / have failed to improve so stopped taking the drug
What is intention to treat analysis
all randomised subjects are included in the analysis, regardless of their completion / adherence to study
What are +s and -s of intention to treat analysis
ADVANTAGES:
- mirrors real life results (effectiveness > efficacy)
- ensures maintainance of comparability between groups obtained through randomisation, maintaining sample size and eliminating bias
DISADVANTAGE:
-reduces statistical power and may thus fail to demonstrate a real effect
what is another name for a TIME TO EVENT CURVE
a KAPLAN MEIR curve
What are stat ways to analyse a time to event curve / survival distribution???
Cox proportional Hazard
Log-rank
Wilcoxon two-sample test
What does each downward step represent on KAPLAN MEIR curve
Each downward step represents an event experienced by a patient
What does each small vertical tick reeprsent on KM curve?
Each vertical tick represents a censored observation (death, lost to follow up, or study period ends)
what is a set of criteria you can use for causality?
Bradford HIll Criteria
What is the Relative Risk Reduction
Proportion of risk reduction attributable to an intervention
RRR = (CER-EER)/CER
What are good outcomes?
Outcomes determined A PRIORI
That are patient oriented and meaningful to pt and HC
What do the Bradford HIll criteria for assessment of causality include=
- Strength of effect (the larger an association, the more likely it is to be causal)
- Biological plausibility
- Coherence (with otgher studies)
- Consistency
- Temporality….
What is log-rank used for
To statistically compare two groups. Assumes proportional hazard
What are proportional hazards
the assumption that HR remain constant over time
What is Cox prop hazard model used for
Compares two groups bu including other factors (covariates) - similar to multiple regression
also assumes proportional hazards
What model is used to analyse a KM curve when you CANNOT assume proportional hazards?
Wilcoxon test - as it gives more weight to deaths/events early on in time
What is the benefit of a Wilcoxon test
It gives more weight to deaths or events early on in time
So it is good at detecting EARLY differences between the two groups
What is a linear regression analysis?
Looks at whether there is a linear relationship between dependent variable and independent variables
What type of linear regression analysis exist
Univariable (1 independent variable only)
Multivariable (>1 independent variable)
What number do we look at for linear regression
R2
What does R2 indicate
Coefficient of determination
Measures % of variation in dependent variable that is explained by the variation in the independent variable
= fraction of variation in dependent variable that is explained by the model
So if R2 = 52%, smoking accounts for 52% of inhaler use. Other factors need to be considered.
what are strategies to reduce confounders?
- Randomization (aim is random distribution of confounders between study groups)
- Restriction (restrict entry to study of individuals with confounding factors - incl/excl criteria, although risks bias in itself)
- Matching (of individuals or groups, aim for equal distribution of confounders)
- Stratification (confounders are distributed evenly within each stratum)
- multivariate analysis (only works if you can identify and measure the confounders)
when is logistic regression used?
Used to estimate the association of INDEPENDENT VARIABLES to a BINARY DEPENDENT VARIABLE (the outcome)
I.e. the outcome MUST be a YES/NO situation
what are methods to deal with missing data in an intention to treat protocol?
- Worse-case scenario
- Hot deck imputation (fill in missing vaules from similar subjects with complete records)
- last observation carried forward