Why do we perform diagnostic tests?

Reduce uncertainty about the true condition of the patient

- Distinguish between disease and no disease
- Provide prognostic information
- Determine response to therapy
- Provide reassurance

What is an index test?

Why is it used?

The test that we are studying

Always want easier and cheaper tests that give same info as gold standard

What is a gold standard test?

A test generally accepted to be definitive in diagnosis of disease, reference test

Which tests do participants in a diagnosis study receive?

Every patients gets both index test and gold standard test

Why is diagnostic dilemma a criteria for performing a test?

A test is most helpful when it helps reduce uncertainty

Performing a test in a patient whose diagnosis is known is not helpful

Tests are most helpful when they distinguish between 2+ likely diagnoses

What is spectrum bias?

Performance of diagnostic test varies because the people tested are not part of a clinically relevant population

Occurs if a study population does not represent a true diagnostic dilemma

Affects external validity

Draw a 2x2 table for a diagnostic study

What is sensitivity?

How do you calculate sensitivity?

Given the presence of disease, the probability that a test will be positive

Sensitivity = TP / (TP+FN)

(use left column)

What is specificity?

How do you calculate specificity?

Given the absence of disease, the probability that a test will be negative

Specificity = TN / (FP+TN)

(use right column)

What is the formula for sensitivity?

Sensitivity = TP / (TP+FN)

Use left column

What is the formula for specificity?

Specificity = TN / (FP+TN)

Use right column

Is sensitivity used to rule in or out?

SNOUT - sensitivity, rule out

Is specificity used to rule in or out?

SPIN - specificity, rule in

Is sensitivity more useful for screening or confirming a diagnosis?

Sensitivity is useful for screening

Is specificity more useful for screening or confirming a diagnosis?

Specificity is useful for confirming a diagnosis

Increased sensitivity is related to [direction, sign] predictive value

Increased sensitivity is related to increased negative predictive value

Increased specificity is related to [direction, sign] predictive value

Increased specificity is related to increased positive predictive value

Describe the important points about sensitivity

Given the presence of disease, the probability that a test will be positive

Minimize false negatives

SNOUT - sensitivity, rule out

Most useful when test is negative

Increased negative predictive value

Screening

Sensitivity = TP / (TP+FN)

(use left column)

Describe the important points about specificity

Given the absence of disease, the probability that a test will be negative

Minimize false positives

SPIN - specificity, rule in

Most useful when test is positive

Increased positive predictive valuie

Confirming a diagnosis

Specificity = TN / (FP+TN)

(use right column)

What is the likelihood ratio?

What advantages does it have over sensitivity and specificity?

What is it used to calculate?

What is the formula for likelihood ratio?

Another way to describe the performance of a test

Can be used across a spectrum of test

Used to calculate post-test probability of a disease

LR = P(result in diseased person) / P(result in non-diseased person)

= true rate / false rate

What is the formula for positive likelihood ratio?

LR+ = P(positive result in diseased person) / P( positive result in non-diseased person)

= true positive rate / false positive rate

= [TP/(TP+FN)] / [FP/(FP+TN)]

= sensitivity / (1 – specificity)

What is the formula for negative likelihood ratio?

LR- = P(negative result in diseased person) / P(negative result in non-diseased person)

= false negative rate / true negative rate

= [FN/(TP+FN)] / [TN/(FP+TN)]

= (1 – sensitivity) / specificity

In calculating a likelihood ratio, sensitivity is in the _____ and specificity is in the _____

In calculating LR,

- sensitivity is in the numerator
- specificity is in the denominator

Negative LR is influenced by ______

Sensitivity

(important to rule out disease)

Positive LR is influenced by ______

Specificity

(important to rule in diseasea)

What is the best negative LR?

0

What is the best positive LR?

Infinity

What is a completely unhelpful LR?

1

What is the relationship between pre-test odds, LR, and post-test odds?

Pre-test odds x LR = Post-test Odds

How can you convert probability into odds?

Odds = probability / (1 – probability)

How can you convert odds into probability?

Probability = odds / (1 + odds)

What is a receiver operating curve?

Visual representation of the relationship between sensitivity and specificity

Slope = positive LR

X-axis: false positive rate = 1 – specificity

Y-axis: true positive rate: sensitivity

What do you look at to see if a test is a good test on an ROC?

Area under the curve

(the higher the AUC, the better the test)

What is a useless test on a receiver operating curve?

Area under curve = 0.5

What is a perfect test on a receiver operating curve?

Area under curve = 1

What is the X-axis on a receiver operating curve?

False positive rate = 1 – specificity

What is the Y-axis on a receiver operating curve?

True positive rate = sensitivity

What are three important uses of the receiver operating curve?

- Comparing the performance of two or more tests
- Determining the best cutoff value of a test to determine disease versus no disease
- Understanding how good a test is overall (area under the curve)

What is prevalence?

What is the formula for prevalence?

Proportion of individuals in a population who have a given clinical characteristic at a given point in time

At time point X: Prevalence = cases of disease / entire population

What is incidence?

What is the formula for incidence?

The number of new cases of disease that accumulate during a specified time period divided by the number of persons in population at risk

From time period A to B: Incidence = new cases of disease / population without disease at time A

What is post-test probability?

Given a certain test result, what is the probability of disease?

What is the formula for positive post-test probability?

Positive post-test probability = TP / (TP+FP)

(use top row)

What is the formula for negative post-test probability?

Negative post-test probability = FN / (TN+FN)

(use bottom row)

What is positive post-test probability?

If patient has a positive test result, the probability a disease is present

Positive post-test probability = TP / (TP+FP)

What is positive predictive value?

If test result is positive, the chances the patient has the disease

Positive predictive value = TP / (TP+FP)

What is the relationship between positive predictive value and positive post-test probability?

Positive predictive value = positive post-test probability = TP / (TP+FP)

(use top row)

What is negative post-test probability?

If patient has a negative test result, the probability the disease is still present but the test missed it

Negative post-test probability = FN / (TN+FN)

What is negative predictive value?

If the test result is negative, the chance the patient does not have disease

Negative predictive value = TN / (TN+FN)

What is the relationship between negative post-test probability and negative predictive value?

Negative post-test probability = 1 – negative predictive value

(use bottom row)

Differentiate sensitivity/specificity from predictive value

- Sensitivity and specificity are characteristics of the test itself
- Predictive value tells us how the test will perform in the population

Draw a 2x2 table with formulas for positive and negative likelihood ratio

Draw a 2x2 table with formulas for positive and negative predictive value and positive and negative post-test probability

If prevalence increases, what happens to positive and negative predictive value?

If prevalence (pre-test probability) increases:

Positiive predictive value increases

Negative predictive value decreases

What types of bias can be present in a diagnostic study?

- Spectrum bias (affecting external validity)
- Selection bias (test referral bias, in which haveing a positive test makes it more likely that a subject also gets the gold standard test)
- Measurement bias (non-blinding, inaccurate measurements, inaccurate gold standard)

When would you want a more sensitive test?

When you cannot miss anyone with the disease

- Dangerous, treatable illness
- Screening
- Early in workup when several diagnoses are being considered

When would you want a more specific test?

When you do not want to falsely label someone

- Late in diagnostic workup
- When false positives can harm patient (label is harmful, treatment is risky)

How do you convert pre-test probability to post-test probability?

- Pre-test probability to pre-test odds using odds = probability / (1 – probability)
- Pre-test odds to post-test odds using pre-test odds x LR = post-test odds
- Post-test odds to post-test probability using probability = odds / (1 + odds)

What three things do you need to calculate post-test probability?

- Sensitivity of the test
- Specificity of the test
- Pre-test probability (prevalence) of disease

What is serial testing?

When is it used?

Do tests in sequence

Only do next test if first one is positive, so all tests done must be positive to label w/ diagnosis

- Rapid assessment not needed
- Tests are expensive or risky
- Want to maximize specificity

When is serial testing used?

- Rapid assessment not needed
- Tests are expensive or risky
- Want to maximize specificity

What occurs with each subsequent positive result in a serial testing strategy?

False positives decrease

Specificity increases

What is parallel testing?

When is it used?

Do all tests at the same time

Need only one positive to label w/ diagnosis

- Rapid assessment necessary
- Patient cannot return easily
- Costs of tests relatively low
- Want to maximizine sensitivity

When is parallel testing used?

- Rapid assessment necessary
- Patient cannot return easily
- Costs of tests relatively low
- Want to maximizine sensitivity

What occurs if all tests are negative in a parallel testing strategy?

False negatives decrease

Sensitivity increases

When is sensitivity prioritized in diagnostic testing?

When you cannot miss a diagnosis

- Screening
- Early in the work up
- Terrible but treatable diseases (TB, syphillis, etc)

When are sensitive tests most useful?

When they are negative

(rule out the disease in question)

When is specificity prioritized in diagnostic testing?

When you need a very accurate diagnosis

- Treatment is very morbid (e.g. chemo)
- Label can be harmful (e.g. HIV)
- Later in the work up

When are specific tests most useful?

When they are positive and rule in the disease in question

Say you use a serial testing strategy with three tests. How can you easily calculate the post-test odds?

Pre-test odds x LR_{1} x LR_{2} x LR_{3} = Post-test odds