Diagnosis (part 1) Flashcards
What types of studies assess diagnostic accuracy?
Cross-sectional studies
RCTs
Screening
Systematic reviews
Describe cross-sectional studies
- Accuracy of the test at a point in time.
- Compare with gold standard
- Gold standard could be hindsight (correct diagnosis apparent with time)
What are the two types of cross-sectional studies?
Cohort - sample population suspected of having diagnosis (not known until after enrolment)
Case-control - one group clearly has diagnosis and other doesn’t (known at enrolment)
— prone to bias
When are RCTs used in diagnostics?
Usually to see if testing affects outcomes (e.g. Routine XR improve pt Mx with LBP?)
How is screening used?
Applies diagnostic tests to subjects who are NOT presenting with a disorder/complaint.
Effectiveness of screening by RCT, e.g. early detection helps…
What does SENSITIVity refer to?
Correctly identifying people who HAVE the disease.
How is SENSITIVity calculated?
From people who actually have the disease:
Correctly diagnosed WITH disease/total with disease
Is a high sensitivity always good?
No. Could just always be a positive test.
What does SPECIFity refer to?
Identify “well” people (no disease)
How is SPECIFity calculated?
From people who actually DON’T have the disease:
Correctly diagnosed WITHOUT disease/total without disease
What is the PPV?
Positive predictive value.
- chance that a positive test result will be correct.
How is the PPV calculated?
From only the POSITIVE test results:
Correctly diagnosed with disease/total DIAGNOSED
What is the NPV?
Negative predictive value.
- the chance that a negative test result will be correct.
How is the NPV calculated?
From only the NEGATIVE test results:
Correctly diagnosed WITHOUT the disease/total DIAGNOSED WITHOUT disease
How does the sensitivity, specificity, PPV and NPV change with a disease prevalence?
Sensitivity and specificity stay the same.
PPV falls with lower prevalence but NPV will rise.
If the specificity of a test is 88%, what percentage of patients will have the same symptoms but no disease?
12%
1 - specificity
What is the likelihood ratio and how is it calculated?
How much more likely the test result would be seen in someone with the condition, than someone without.
Sensitivity/(1 - specificity)
How are likelihood ratios interpreted?
> 1 = test is more often positive
3 = test is useful
10 = test is very useful
What can sensitivity/specificity be used to describe, and what can’t they predict?
Describe:
- how abnormality/normality predicts particular test results
Can’t predict:
- risk of abnormality
Assessment of diagnosis (3 steps)
- Is the evidence about diagnosis valid?
- Is it important?
- Can you apply it clinically?
- Is the evidence valid? (3)
Herbert et al 2005
A. Comparison with adequate reference standard
B. Comparison bind?
C. Study sample consisted of subjects with diagnostic uncertainty?
Describe 1A (4)
Ref standard
- ref standard will often have face validity, e.g. open surgical
- reliability of standard should be assessed
- ref standard may be hindsight
- imperfect ref standard = under estimate of the accuracy
Describe 1B (2)
Blinding
- assessor should be blind to result of the ref test
- – knowledge would bias the index test
Describe 1C (2)
Diagnostic uncertainty
- spectrum of patients enrolled similar as possible to clinical practice? (COHORT study)
- aim is to solve diagnostic dilemmas (not confirm obvious Dx/negative Dx) (CASE-CONTROL study)
Is this evidence about a Dx test valid? (other list) (4)
Straus et al 2005
- Independent, blind comparison with reference (gold) standard of Dx?
- Appropriate spectrum of pts?
- Ref standard applied REGARDLESS of Dx test result?
- Test/cluster of tests validated in a second, independent group of pts?
