Evidence about Diagnosis Flashcards
Can you test for diagnostic accuracy with RCT?
NO- as diagnostic accuracy depends on whether Px has or doesn’t have disease
What is diagnosing?
classifying individual as having ot not having disease
what are things optometrists consider when making diagnosis?
symptoms signs characteristics Hx Examination Results
How can we measure a diagnostic test?
Against a ‘gold standard’ or ‘reference test’
Can a diagnostic test be more than one test?
YES- an examination as a whole can be considered to be a diagnostic test
Best type of study to estimate diagnostic test accuracy?
- Consecutive sample
- Defined clinical presentation
- gold standard and test of interest performed in every individual
- SYSTEMATIC REVIEWS BEST
What is better for testing diagnostic accuracy- consecutive or non-consecutive sample?
Consecutive sample
non-consecutive- sample bias
Can we use case control studies to test diagnostic accuracy?
yes- but weak study
want to avoid
Why are case control studies not great for diagnostic accuracy testing?
- subject to selection bias as non-consecutive sampling
- less likely to enrol Px’s with whole spectrum
PICO for diagnostic accuracy
P- population
I- diagnostic test we’re interested in (rather than intervention)
C- gold standard
O- diagnosis in question
For the following diagnostic accuracy question- what is the question:
You are working as an optometrist in a clinic
that performs refractive surgery on adults.
Your routine is to use both objective as well
as subjective refraction. A patient asks why
you do both. The question prompts you to
wonder if objective methods are as accurate
as the gold standard (subjective methods).
P: adults
I: objective refraction
C: subjective refraction
O: accuracy of measuring refractive error
Clinical question: In adults, how does objective refraction tests compare in accuracy to subjective refraction tests for Px’s considering Rx error surgery
What Test used to Appraise Diagnostic Research Quality? What about RCT? What about prognosis? systematic reviews?
QUADAS-2
RCT- PEDro
(prognosis- CASP)
(sys reviews- AMSTAR or CASP)
Is it appropriate to exclude difficult to diagnose Pxs from diagnostic accuracy study?
NO
What types of studies report sensitivity and specificity?
Diagnostic accuracy studies
What is Sensitivity?
Px’s with the disease who test +ve
Sensitivity = a/ (a+c)
D+ D- T+ a b T- c d
What is Specificity?
Px’s without the disease who test -ve
Specificity = d/ (d+b)
D+ D- T+ a b T- c d
What is a true positive and false positive
true positive- Px’s with the disease that get +ve result
False +Ve- Px without disease that gets +ve result
What is a true negative and false negative
true neg- Px without disease that gets neg result
false neg- Px with disease that gets neg result
Complete a table for a test on a sample of n=1000 Px’s with prevalence of 1% with disease
Test 1 D+ D-
T+ 6 99
T- 4 891
totals 10 990
How do we diagnose?
Set a criterion
What effect does decreasing criterion have on specificity and sensitivity?
increases sensitivity (more Px's tested as pos) decreases specificity (more Px's with disease classified as +ve)
High criterion- effect on sensitivity and specificity
low sensitivity
high specificity
What is more useful that sensitivity and specificity?
Predictive values
What are predictive values?
probability that the test result is accurate
Does sensitivity and specificity change with the population?
NO- it is a property of the test itself
calculate Positive Predictive Value (PPV):
PPV = probability of having disease when test is +ve
(go down test column)
PPV = a / (a+b)
D+ D- T+ a b T- c d
calculate Negative Predictive Value (NPV):
Probability of not having disease when test it -ve
(go down test column)
PPV = d/(d+c)
D+ D- T+ a b T- c d
Calculate PPV and NPV: TEST 1 D+ D- T+ 6 10 T- 4 980 totals 10 990
PPV = 6/16 = 0.375
NPV = 980/984 = 0.995
True or false- prevalence is not important for PPV and NPV
FALSE
Effect of high prevalence on PPV and NPV?
increases PPV
decreases NPV
Does prevalence effect sensitivity and specificity?
NO- property of test
Likelihood ratio for positive test result
how much more likely a Px’s +Ve test would be for someone with disease compared to someone without disease
LR+ = sensitivity / (1-specificity)
Likelihood ratio of negative test result
how much more likely a Px’s -ve test result would be for someone without disease compared to someone with disease
LR- = (1-sensitivity) / specificity
What does a PPV of 0.03 mean on a RAT test
3% chance of having COVID if test result is +ve
not reliable test
What does NPV of 0.998 mean on RAT test?
99.8% chance of not having COVID is test result is -ve