Diagnostic Test Studies

1
Diagnostic Test Studies

• Assessing Validity
• Understanding Results

2
Questions about Validity

• Was there an independent, blind comparison with a
  reference standard?
• Did the patient sample include an appropriate
  spectrum of patients to whom the diagnostic test
  will be applied in clinical practice?

3
Independent, blind comparison with a reference
standard?

• Reference standard the defining test for the
  condition (gold standard) or the best available
  test.
• Independent Everyone gets both the new test and
  the reference standard
• Blind The person scoring the new test doesnt
  know the reference diagnosis, and vice versa

4
Appropriate patient spectrum?

• Characteristics of the patients who are studied
  should be similar to the patients in whom you
  want to apply the test
• Bad studying a test in patients in an advanced
  state of disease that you mean to apply to
  patients in an early stage
• Okay differences in prevalence of the condition

5
Typical study design

• The typically best study design is a prospective
  cohort study
• Identify cohort of appropriate spectrum
• Given each patient new test and reference
  standard in a blind fashion
• Compare test results

6
Working with results

• When a patient presents, the clinician thinks
  they have some (prior or pre-test)
  probability of being sick
• Based on clinician suspicions
• Based on prevalence of illness in population
• The purpose of a diagnostic test is to change
  that probability to a post-test probability that
  is high enough to act (or low enough to ignore)
• A positive test should increase the probability
• A negative test should decrease the probability
• Better tests lead to bigger changes in
  probabilities.

7
Getting it right and wrong

• Given that you know the patients true disease
  state (from the reference standard)
• A test can be right in two ways
• True positive (test says a sick person is sick)
• True negative (test says a well person is well)
• A test can go wrong in two ways
• False positive (test says a well person is sick)
• False negative (test says a sick person is well)

8
The 2x2 table
9
Example of a 2x2 table
Study prevalence 100/1000 (10) of those studied
were sick.
10
Sensitivity How good is the test when youre
sick?
This test correctly picks up 95/100 people who
are sick. Its sensitivity is 95
11
Specificity How good is the test when youre
healthy?
This test correctly classifies 100/900 people who
are healthy. Its specificity is 89
12
Using sensitivity/specificity

• Sensitivity and specificity are test
  characteristics that are independent of disease
  prevalence.
• With sensitivity, specificity, and your pre-test
  probability, you can compute your patients
  post-test probability if they have a positive or
  negative test.

13
Using sensitivity/specificity

• A negative result on a highly sensitive test
  rules out the disease, because if you were sick,
  a highly sensitive test would be positive.
  Mnemonic SnNout
• A positive result on a highly specific test rules
  in the disease, because if you were well, a
  highly specific test would be negative. Mnemonic
  SpPin

14
Likelihood ratios

• A likelihood ratio is the probability that, for a
  given test result, the patient is in the sick
  rather than the well population.
• Each test result (positive, negative) has a
  likelihood ratio (LR, LR-)
• LR should be greater than 1
• LR- should be less than 1 (fractional)
• LR of 1 means the test result adds no new
  information (result is equally likely to occur in
  a sick as in a well person)

15
Calculating LRs

• LR sensitivity / 1 specificity
• LR- 1 sensitivity / specificity
• But lifes too short, so let a computer or
  calculator do it

16
(No Transcript)
17
The nomogram