Conventional Analysis Of Diagnostic Test Data

Question 1

What’s been conventionally the standard way of describing the accuracy of diagnostic test? When is that performed?

Answer 1

two-by-two table.

When the test results are recorded as dichotomous outcomes (positive/negative results).

Question 2

Gold standard? How is that measured?

Answer 2

This standard may be another test but more expensive diagnostic method or invasive method but more accurate or combination of tests may be available in clinical follow up, surgical verification, autopsy, biopsy or by a panel of experts

Question 3

From the frequency of test results among patients with and without disease based on gold standard, one can derive?

Answer 3

probability of a positive test result for patients with disease and probability of negative test results for patients without disease

Question 4

What are the two other useful indices in clinical practice?

Answer 4

The positive predicted values (PPV) and the negative predicted values (NPV)

Question 5

PPV?

Answer 5

Probability of disease for positive test results

Question 6

NVP?

Answer 6

Probability of being healthy for negative test results

Question 7

Although these two measures (NPV JA PPV) are useful for clnical decision they are?

Answer 7

They are influenced by the prior prevalence of disease in population. PPV is elevated with a higher prevalence of disease while the NPV decreases with a higher prevalence

Question 8

The PPV and NPV can also be calculated from?

Answer 8

Bayes’ theorem using the estimates of sensitivity and specificity and the prior probability of disease (or prevalence of diseased in population) before the test is applied.

Question 9

The PPV and NPV are calculated through? IN Bayes’ theorem?

Answer 9

the posterior probability of the diseased after the test results are known.

Question 10

When does one calculate the PPV and NPV?

Answer 10

If one knows the sensitivity, specificity and pre-test probability of the diseased in population = prevalence .

Question 11

When disease prevalence is high?

Answer 11

The PPV increases and the NPV decreases

Question 12

The likelihood ratio?

Answer 12

Todennäköisyys suhde.

The likelihood ratio is defined as the ratio of two density functions of test results conditional in diseased and non-diseased subjects.

Todennäköisyyssuhde määritellään testitulosten kahden tiheysfunktion suhteena, jotka ovat ehdollisia sairailla ja ei-sairailla koehenkilöillä.

Question 13

The positive LR of test?

Answer 13

Is nothing more than the ratio = suhde of sensitivity to 1-specificity

The LR+ is ranged from 0 to infinity

Question 14

What is the worst case about LR+?

When does that happen?

Answer 14

When LR+=0

When sensitivity becomes close to 0

Question 15

The largest value of LR+ oocurs when?

Answer 15

Specificity tends to be close to 1 and the sensitivity also be close to 1.

Thus the higher value of LR+ has a greater information value for diagnostic test.

Question 16

On the other hand the negative likelihood ratio?

Answer 16

LR- is the ratio of probability of negative test in diseased to non diseased.

Question 17

The lower (eli close to 0) LR- has?

Answer 17

A greater information values of a negative test. The larger value of LR- has lower information values.

Question 18

he Bayesian analysis combines the data?

Answer 18

likelihood ratio of test and prior odds of disease in order to obtain the posterior odds of disease among positive and negative test results

Question 19

Posterior odds of disease = ?

Answer 19

likelihood ratio × prior odds of disease

Question 20

TPF?

Answer 20

True Positive Fraction (Sensitivity)= TP/ (TP+FN)= a/(a+c)

Question 21

FNF?

Answer 21

False Negative Fraction (1-Sensitivity)= FN/ (TP+FN)= c/(a+c)

Question 22

TNF?

Answer 22

True Negative Fraction (Specificity)= TN/ (TN+FP)= d/(b+d)

Question 23

FPF

Answer 23

False Positive Fraction (1-specificity)= FP/ (TN+FP)= b/(b+d)

Question 24

PPV

Answer 24

Positive Predicted Value=TP/(TP+FP)=a/(a+b)

Question 25

NPV

Answer 25

Negative Predicted Value=TN/(TN+FN)=d/(c+d)

Question 26

What does the conventional analyses consider?

Answer 26

the sensitivity and the specificity of a diagnostic test as the primary indices of accuracy since these indices are considered to be independent of the prior probability of disease.

Question 27

Why is using a single sensitivity and a single specificity as measures of accuracy problematic?

Answer 27

these measures depend on a diagnostic criterion (i.e. cut-off) for positivity which is often chosen arbitrarily. For example, one observer may choose a lenient decision criterion and the other may choose a stringent decision criterion for positivity.

Eli nämä mittaukset riippuvat diagnostisesta kriteereistä eli raja-arvoista positiiviselle, mikä taas voidaan valita sattumanvaraisesti. Joku voi valita lievästi/helposti päätöksen ja toinen taas tiukan päätöksen kriteereistä positiivisuudelle.

Question 28

What does ROC analysis do?

Answer 28

ROC analysis circumvents this arbitrariness.

Question 29

Why total information content of diagnostic test can be summarized either it’s LR+ or LR- ?

Answer 29

higher value of LR+ has a greater information value for diagnostic test. The lower (i.e close to 0) LR- has a greater information values of a negative test