Evaluation of Diagnostic test

In which circumstances do we want to MINIMIZE false negatives ?

If the disease is -often asymptomatic -is serious (high mortality) -Progresses quickly -can be treated more effectively at an EARLY age -spreads one person to another e.g. Squamous cell carcinoma, TB

(Sensitivity/ Specificity/ PPV/NPV) are highly affected by prevalence of the disease in population

PPV + NPV

"The sensitivity is 87%" → Out of 100 people with OLP, ______ will test positive and _________ will test negative.

87 13

_________________ : Probability of a positive test result when condition is present

sensitivity

_______________ bias occurs when non-blinded testing leads to overestimation of the accuracy of the diagnostic test.

Review bias

Any given PPV and NPV are only accurate for YOUR patients if :

your patients have the same pre-test probability as the population being studied (e.g. general population = low risk for OLP & your patients = also low risk for OLP)

If you DECREASE the prevalence of a disease, there is a RELATIVE (increase/ decrease) in false positives. This occurs because the (PPV/NPV) (decreases/ increases).

INCREASE PPV decreases

SpPIn & SnNOut

SpPIn = for a highly specific (95%) test, a positive result rules in the disease SnNOut = for a highly sensitive (95%) test, a negative result rules out the disease

_______________ : Probability that a subject who is disease free will test negative. _______________ : Probability that a subject who has disease will test positive.

Specificity Sensitivity

_____________ bias occurs when the patients recruited to a study are NOT representative of those on which the test will be used in practice.

Spectrum bias

You want to minimize false (negative/ positive) for patients with CARIES. You want to minimize false (negative/ positive) for patients with SCC.

positive negative

Predictive values is also termed ________________

post-term probability

Name all biases commonly found in diagnostic test papers:

-Spectrum bias -Review bias -Verification bias

Likelihood ratios can be calculated from __________________ Likelihood ratios (do/ don't) depend on the prevalence of disease.

Sensitivity+ specificity of a test do NOT

If you decrease the prevalence of a disease, -false positives go (up/down) -PPV goes (up/ down) -NPV goes (Up/ down)

Up Down Slightly up

_______________ bias is when people in a study of a diagnostic test don't have an equal chance of getting the reference standard.

Verification bias

________________ : Probability of a negative test result when condition is absent

specificity

"The NPV is 96%" → Out of 100 people who test negative, ___________ will not have OLP

96 (+ 4 false negatives)

If we adjust the cut-off line to minimize the false positives, there's a side effect of _________________.

Increase in false negatives.

"The PPV is 49%" → Out of 100 people who test positive, ___________ will have OLP.

49 (+ 51 false positives)

The denominator for sensitivity is ________________

All patients who have the disease or condition

The denominator for Positive predictive value is __________________

All patients who test positive

(T/F) Sensitivity / Specificity tell you how well the test itself works (T/F) Sensitivity/ Specificity tells you how the results apply to your patient.

True false (doesn't tell you much about how the results apply to your patient just about accuracy of diagnostic test- so it may not be clinically useful)

As the prevalence of a disease decreases, PPV of diagnostic test gets (worse/ better), due to (increase/ decrease) in false positives. This leads to risk of (under/ over)-treatment.

Worse ; Increase Over-treatment

"The specificity is 73%" → Out of 100 people without OLP, ______ will test positive and _________ will test negative.

27 73

PICO in Papers dealing with Diagnostic tests :

Intervention - diagnostic test being STUDIED Comparison - Reference standard or the GOLD standard

_______________ measures help you SHIFT your suspicion of a particular diagnosis from "pre-test probability" to the "post-test probability"?

Likelihood ratio

_____________ for positive result tells you How much the ODDS of the disease increase when a test is positive. _____________ for negative result tells you How much the ODDS of the disease decrease when a test is negative.

Likelihood ratios

________________ : Probability, given a positive test result, that the patient has the condition. ________________ : Probability, given a negative test result, that the patient does not have the condition.

Positive predictive value negative predictive value

_____________ : Probability of those testing/ screening positive truly having the disease _____________ : Probability of those testing/ screening negative NOT actually having the disease

Positive predictive value (PPV) Negative predictive value (NPV)

Define a positive predictive value (PPV) :

Probability of those testing/ screening positive truly having the disease

Define Specificity:

-Probability that a subject who is DISEASE-FREE will test negative -A highly specific test will NOT falsely identify many people as having the condition, when in fact they do not

Define Sensitivity:

-Probability that a subject with DISEASE will test positive -A highly sensitive test will not miss many people with the condition

Likelihood ratios:

-Provide direct estimate of how much a test result will change the ODDS of having a disease -Tells us the ability of a diagnostic test to move us from a pre-test probability to a post-test probability

3 types of statistics for RESULT of diagnostic test:

-Sensitivity & specificity -Predictive values (PPV, NPV) -Likelihood ratio

_______________ measures remain CONSTANT, no matter what setting a test is used in.

sensitivity + specificity

Purpose of diagnostic tests:

-To Rule in disease -To Rule out disease

(T/F) Sensitivity and specificity are usually clinically useful.

false

Who should be the representative participants of study regarding diagnostic tests?

-Those that have conditions that are different from each other but are commonly CONFUSED - in "grey areas" of diagnosis

What are predictive values?

-"If I test positive, what is the chance that I HAVE the condition?" -"If I test negative, what is the chance that I DON'T have the condition?" -represented by positive (PPV) and negative predictive (NPV) values

What is the requirement of choosing a "gold standard" reference diagnostic test?

-Must be INDEPENDENT, stand-alone of the diagnostic (index) test -Must have no overlap with the diagnostic test of interest

Define a negative predictive value (PPV) :

Probability of those testing/screening negative NOT actually having the disease

In which circumstances do we want to MINIMIZE false positives ?

-When costs/ risks of treatment are high -When treatment is non-reversible -When disease itself is not life-threatening -e.g. caries in patients at low risk -e.g. prostate cancer in very elderly men, unlikely to develop clinically significant disease (minimization of cancer treatment, since it can be aggressive)

<Rule of thumb> A test with (+)Likelihood ratio bigger than ________ is good to Rule IN a disease. A test with (-)Likelihood ratio smaller than ________ is good to Rule OUT a disease.

10 0.1

Spectrum bias leads to diagnostic tests appearing (more / less) accurate than it would in appropriate select of patients.

MORE