Sensitivity
The ability of a test to correctly detect a condition.
True positive.
True positive / True positive + False negative
Specificity
The ability of a test to exclude subjects without the condition. True negatives.
True negative / True negative + False positive
2 tests that make statement about the test in general
Sensitivity and specificity
2 tests that makes a statement about what a test means to the patient
Positive and negative predictive value.
Positive predictive value
Probability of having the condition when testing positive.
True positive / True positive + False positive
Negative predictive value
Probability of not having the condition when testing negative
True negative / True negative + False negative
Accuracy
The proportion of all tests providing the correct result
True positive + true negative / All
Likelihood ratio of a positive test
Odds of a disease if the test result is positive
Sensitivity / 1 - specificity
Likelihood ratio of a negative test
Odds of a disease if the test is negative
1- sensitivity / Specificity
10 questions to evaluate a paper claiming to validate a diagnostic or screening test
- Is the test relevant to clinical practice?
- Has it been compared with the gold standard?
- Was an appropriate spectrum of participants included?
- PPV/NPV are dependent on prevalence. - Verification bias avoided?
- Confirm all subjects were tested with both the gold standard and the new test. - Was expectation bias avoided?
- Do this by blinding the interpreter of the tests. - Test reproducible both within and between observers ?
- Intra observations: same observer, same result.
- Inter observations: different observers, same results - What are the test features?
- Sensitivity, specificity, PPV, NPV - Were confidence intervals provided?
- Normal range derived from results?
- Has the test been placed in the context of other similar tests?
Intra observations vs inter observations
Intra: Same observer, same result.
inter: Different observer, same result.
Confidence interval
Possible range of results within which the true value will lie.
Ideal: narrow and based on large sample size.
What happens to the CI with a small sample size?
CI will be sensitive to small changes.
Variables (parameters)
What we are measuring or manipulating in a study
Two types of variables
Numeric - continuous and discrete
Categorical (non numerical)- ordinal and nominal
Continuous variable (branch off numeric)
One that takes on an infinite number of values in a certain range.
Ex: distance, temp, time, some biomarkers.
Discrete variable (branch off numeric)
Countable whole number. Ex: 10 people in the room. Not 10.5
Dichotomous variable (apart of categorical nominal)
only have 2 values.
Male-female
Employed- non employed
Ordinal (apart of categorical)
Distance between categories is unknown.
Usually questionnaires use this.
Ex: Leikert scale.
Likelihood ratio
Ratio of the likelihood of a test result in one with disease to likelihood of the same test result in one without the disease.
Preferred way of expressing and comparing the usefulness of different tests.
Defines risk of actually having the condition.
Useful for continuous variables.
Calculated using sensitivity and specificity.
Likelihood ratio is calculated using
Sensitivity and specificity
Likelyhood ratio defines ____ of having the condition
Risk
Clinical prediction
Numerical estimate of chance of person having or developing disease. Can use likelihood ratio to compare pre-test probability to post-test probability.
Three stages of developing a clinical prediction rule (standard of care)
- Establish indecent and combined effects of symptoms, signs, and diagnostic tests.
- Assess explanatory variables above in different populations to be more generalizable.
- RCT to assess the impact on patient outcomes applying the rule. Most critical part.
