EBM Flashcards
What is Cox proportional hazards analysis used for?
Compares dichotomous outcomes, except that time itself may be a variant.
Similar to logistic regression multivariant analysis (but time cannot be variant with this)
How is ARR (absolute risk reduction calculated) given the AE rate in placebo and treated groups?
ARR = AE rate in placebo (higher) – AE rate In treated (lower)
What is publication bias?
Publication bias is due to journals only publishing positive results.
Given the incidence of disease in the exposed is 0.10 and unexposed is 0.02. Calculate the RR (relative risk)
RR is an incidence ratio
RR = incidence of exposed (higher) / incidence of unexposed (lower)
RR = 0.10 / 0.02 = 5
How does the clinical utility of sensitivity/specificity differ from PPV/NPV?
Difference in clinical utility:
Sensitivity / Specificity = population studies
PPV / NPV = more relevant to the individual
Describe the calculation of specificity.
Specificity (column) = TRUE negative / ALL without disease
Describe the calculation of sensitivity.
Sensitivity (column) = TRUE positive / All with disease
Which of the predictive values (i.e. PPV/NPV) is 100% for the following:
- Sensitivity
- Specificity
1, If sensitivity = 100% then NPV = 100%
- If specificity = 100% then PPV = 100%
NB: think converse!
What does ANOVA analysis compare?
ANOVA compares >2 means
Compare type 1 (alpha) and type 2 (beta) statistical errors.
Statistical errors:
Type I (alpha) error:
- Denial of truth
- REJECTION of the a TRUE null hypothesis
- p-values
Type II (beta) error:
- Acceptance of falsehood
- FAILURE to reject (i.e. acceptance of) a FALSE null hypothesis
- Power
How is (OR) odd ratio calculated?
OR = (true positive x true negative) / (false positive x false negative)
‘Truth over lies’
Describe the differences in statistical analyses for variable that are discrete vs continuous (dependent vs independent).
Types of statistical analysis:
- Comparing 2 discrete variables = chi-squared
- t-tests are for continuous variable:
a. Unpaired – for independent groups e.g. birthweight in males and females.
b. Paired – for dependent groups e.g. pre- and post- BPs.
What is the fundamental difference between multiple regression multivariant analysis vs. logistic regression multivariant analysis?
Multiple regression multivariant analysis: compares multiple outcomes that are continuous e.g. BP
Logistic regression multivariant analysis: compares multiple outcomes that are dichotomous e.g. alive vs. dead
Considering a normally distributed population, what percentage of a population lie within:
1x SD
2x SD
For normal distribution:
66% within 1x SD
95% within 2x SD
Give the formulae for likelihood ratios LR+ and LR-
LR+ = sensitivity / (1-specificity)
LR- = (1-sensitivity) / specificity
Describe the calculation for NPV and PPV.
NPV (row) = TRUE negative / ALL test negative
PPV (row) = TRUE positive / ALL test positive
What is the difference between attributable/absolute risk and relative risk?
Attributable/Absolute risk (risk difference) = incidence of disease in UNexposed – incidence of disease in exposed
Relative risk (risk ratio) = incidence of disease in exposed / incidence of disease in Unexposed
What is the difference between SD and SEM?
SD: measure variability of data around a mean
SEM: measures the variability of a sample mean with respect to the likelihood of it being the true mean
SEM is not the same as SD
True/False: prevalence of disease plays an important role in the calculation of NPV and PPV.
True
What is an example of observation/measurement bias?
Example of observation/measurement bias: recall of weight by patients vs. actually measuring it.
What is the difference between intention-to-treat and per-protocol statistical analyses?
Intention-to-treat = analysis according to original designations (more realistic)
Per-protocol = analysis according to actual treatment received
What is an example of selection bias?
Example of selection bias: enrolment of hospitalised patient only with bias towards a sicker population of patients.
Describe the Kaplan-Meir curve.
Kaplan-Meir curves calculate the ratio of surviving patients divided by the total number of at-risk patients. Ratio is recalculated everytime the patient has an outcome.
What is lead time bias?
Lead time bias occurs in screening when the patient is diagnosed early by virtue of a screening process and therefore may have an artificially lengthened survival time that is not a function of the natural history of the disease or intervention.
The incidence of a disease is 2.5% in a population exposed to a protective factor/treatment and 3% in those not exposed to the protective for/treatment. Calculate the following:
- ARR
- NNT
- RR
Incidence exposed = 0.025 and incidence UNexposed = 0.03
ARR = 0.03-0.025 = 0.005 = 0.5% [Unexposed – exposed]
NNT = 1/0.005 = 200 people[1/ARR] RR = 0.025/0.030 = 83% [exposed/Unexposed]
Incidence of disease in a population exposed to a risk factor is 10% and incidence in a population unexposed is 2%. What is the relative risk?
RR = incidence exposed / incidence UNexposed = 0.1/0.02 = 5
RR is an incidence ratio
Given the AE rate for a placebo group and treated groups, how does one calculate the ARR?
ARR = AE rate in placebo group – AE rate in treated group
How is post-test probability calculated?
- Calculate LR+ = sensitivity / (1 - specificity)
- Pre-test odds = prevalence / (1 - prevalence)
- Post test odds = pre-test odds x LR+
Convert post-test ODDS to post-test probability:
- Post-test probability - post-test odds / (1 + post-test odds)
