Diagnostic Tests Flashcards
Definition of diagnostic test
Performed to aid diagnosis/disease detection
Definition of sensitivity
Correct identification of diseased with test
Definition of specificity
Correct identification of healthy
Definition of positive predictive value
Probability that the positive result is found in the diseased
Definition of negative predictive value
Probability of the correct identification of the healthy
Definition of the likelihood ratio
Use sensitivity and specificity of test to determine whether a test result usefully changes the probability that a condition exists
Definition of bias
If the average is not close to the true value in a distribution graph
Definition of imprecision
If there is a large distribution of values
Definition of line identity
Line where x=y at 45 degrees from origin
If values of the 2 measurements are almost equal => high levels of agreement
Definition of the line of regression
Line that describes how x changes y
Definition os association
Relationship between the exposure and outcomes
What are diagnostic tests
When are they used
How can they be used
Performed to aid diagnosis/detection of disease
Used in
- radiology
- laboratories
- physiology
Can be a definitive diagnosis
Part of a diagnosis
Part of a screening test
Describe the ideal diagnostic test
Why is this never going to happen
Correct identification of diseased with +ve test result
Correct identification of healthy with -ve test result
Cheap and painless
False positives and false negatives will occur
-will be 4 possible outcomes
What are the 4 ways of assessing the effectiveness of a diagnostic test
Sensitivity
-correct identification of diseased
Specificity
-correct identification of healthy
Positive predictive value
-+ve results due to diseased individual
Negative predictive value
—ve result due to healthy individual
How would you calculate the confidence interval for proportions or percentages
P+- 1.96 √p(1-p)/N
N can be the denominator of the previously used equation
How would you calculate the sensitivity
No of diseased +ve to test/Total no of diseased
How would you calculate the specificity
No of healthy -ve to test/total no of healthy
How would you calculate the positive predictive value
No of diseased +ve/no of +ve results
How would you calculate the negative predictive value
No of healthy -ve/no of -ve results
How would you calculate the accuracy of the test
TP + FN / (TP+TN+FP+FN)
How would you calculate the positive likelihood ratio
Sensitivity/1-specificity
How would you calculate negative likelihood ratio
1-sensitivity/specificity
How can you involve disease prevalence values in the positive likelihood ratio
PPV/1-PPV = P/1-P x LR+
How would you calculate the amount of agreement between qualitative test results
What is k?
K = observed % agreement - expected % agreement/100 - expected % agreement
K=probability that judges will agree if there is no agreement by chance
How would you calculate the observed % agreement
How would you calculate the expected % agreement
Observed % agreement = no of agreements/all obs
Expected % agreements = (no of normal obs in 1 x no of normal obs in 2) + (no of abnormal obs in 1 and no of abnormal obs in 2) / total obs ^2
How can you interpret kappa values
<0, no agreement
- 0-0.20, slight agreement
- 21-0.40, fair agreement
- 41-0.60, moderate agreement
- 61-0.80, good agreement
- 81-1, excellent agreement
How would you process continuous measurements
Continuous measurements may be dichotomised => qualitative +ve or -ve result
How would you choose the cut off point for continuous measurements in diagnostic tests
Choose sensitivity or specificity required
If you move cut point to increase sensitivity, specificity will be sacrificed
Why do we have overlap between the ranges that give positive and negative results in continuous measurements
Patient variation (stress) Measured variation (GP, nurse) Measurement conditions (temp) Instrument condition variation (sphygmomanometer)
All are sources of bias and imprecision
What is the difference between bias and imprecision
Bias, if average is not close to the true value
Imprecision, if there is a large distribution of values
How is agreement between sets of continuous measurements measured
Line of identity drawn
How would you test for agreement
Bias, mean diff with 95% CI
Precision, SD of 95% range
If both are small => good agreement
What is the difference between agreement and association
Association,
- relationship between exposure and outcomes
- tested for with a correlation coefficient
- can involve the line regression
Agreement
- assess the departure of 2 measurements from line of identity
- tested for with SD and CIs
